REF report on Performance of Wind Farms 2013

52
The Performance of Wind Farms in the United Kingdom and Denmark Gordon Hughes

Transcript of REF report on Performance of Wind Farms 2013

Page 1: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 152

The Performance of Wind Farms

in the United Kingdom and

Denmark

Gordon Hughes

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 252

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 352

The Performance of Wind Farms

in the United Kingdom and Denmark

Gordon Hughes

2012

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 452

copy Gordon Hughes 2012

Cover background image by Simon Gray (httpwwwstar-oneorguk) via stockxchng

Published by the Renewable Energy Foundation

57ndash58 Russell Square

London WC1B 4HSwwwreorguk

he Renewable Energy Foundation is a registered charity in England and Wales (1107360)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 552

Contents

List o Figures and ables 4

About the author 5

About the Renewable Energy Foundation 5

Authorrsquos Note 6

Note on the Data 6

Executive Summary 7

Te Perormance o Wind Farms in the United Kingdom and Denmark 9

Introduction 9

Age-perormance curves 10

Differences between wind arms 12Implications or uture policy and perormance 15

Conclusion 21

Appendix Data and Methods 23

A Data or the United Kingdom 23

B Data or Denmark 25

C Specification and estimation methods 26

D Period fixed effects vs normalisation by wind speeds 28

E Estimation results or the UK 30F Estimation results or Denmark 36

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 652

List of Figures and TablesList of Figures and Tables

Figure 1 Perormance degradation due to age using equal weights 11

Figure 2 Perormance degradation due to age using capacity weights 12

Figure 3 Range o unit fixed effects by category 13

Figure 4 Initial load actors by year o commissioning or onshore wind 14

Figure 5 Impact o perormance degradation on discounted cumulative output 16

Figure 6A Projected load actors or UK onshore wind (total capacity o 10 GW in 2020) 18

Figure 6B Projected load actors or UK onshore wind (total capacity o 15 GW in 2020) 18

Figure 7 Projected load actors or UK offshore wind in 2020 20

Figure 8 Box plots o load actors by age or UK onshore wind arms 24

Figure 9A Additive age-perormance curves or UK onshore wind arms 32

Figure 9B Multiplicative age-perormance curves or UK onshore wind arms 32

Figure 10 UK onshore age-perormance curves using equal and capacity weights 33

Figure 11A Residuals by age or perormance curves using equal weights 34

Figure 11B Residuals by age or perormance curves using capacity weights 34

Figure 12 Unit fixed effects by year o commissioning or England and Scotland 35

Figure 13A Age-perormance curves or Danish onshore wind arms 37

Figure 13B Age-perormance curves or Danish offshore wind arms 37

Figure 14 Danish age-perormance curves using equal and capacity weights 38

able 1 Average load actors by year and country () 40

able 2 Estimation results or UK onshore wind arms 41

able 3 Equations or trends in unit fixed effects 43

able 4 Estimation results or Danish wind arms 44

able 5 Perormance by year o commissioning or Danish wind arms 46

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 752

About the author

Dr Gordon Hughes is a Proessor o Economics at the University o Edinburgh where he teaches courses in

the Economics o Natural Resources and Public Economics He was a senior adviser on energy and environ-

mental policy at the World Bank until 2001 He has advised governments on the design and implementationo environmental policies and was responsible or some o the World Bankrsquos most important environmental

guidelines

About the Renewable Energy Foundation

he Renewable Energy Foundation is a registered charity promoting sustainable development or the benefit

o the public by means o energy conservation and the use o renewable energy

REF is supported by private donation and has no political affiliation or corporate membership In pursuit o

its principal goals REF highlights the need or an overall energy policy that is balanced ecologically sensi-

tive and effective

We aim to raise public awareness o the issues and encourage inormed debate regarding a structured energy

policy that is both ecologically sensitive and practical he issues o climate change and security o energy

supply are complex and closely intertwined REF contributes to the debate surrounding these issues by

commissioning reports to provide an independent and authoritative source o inormation

For urther inormation see wwwreorguk

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 852

Authorrsquos Note

Note he views expressed in this paper are strictly personal and do not reflect the position o any organ-

isation with which I am associated I am grateul to John Constable Lee Moroney and a reeree or their

comments on earlier drafs o the paper he results o this study have been presented at seminars andlectures to various groups in Edinburgh and Glasgow I have benefited rom the questions and comments

rom participants in those events

Gordon Hughes

School o Economics

University o Edinburgh

gahughesedacuk

Note on the Data

For the convenience o interested researchers both the raw and cleaned data employed in this analysis will

published alongside the electronic version o the study on the REF website wwwreorguk

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 7

Executive Summary

1 Onshore wind turbines represent a relatively mature technology which ought to have achieved

a satisactory level o reliability in operation as plants age Unortunately detailed analysis o

the relationship between age and perormance gives a rather different picture or both theUnited Kingdom and Denmark with a significant decline in the average load actor o onshore

wind arms adjusted or wind availability as they get older An even more dramatic decline is

observed or offshore wind arms in Denmark but this may be a reflection o the immaturity

o the technology

2 he study has used data on the monthly output o wind arms in the UK and Denmark reported

under regulatory arrangements and schemes or subsidising renewable energy Normalised

age-perormance curves have been estimated using standard statistical techniques which allow

or differences between sites and over time in wind resources and other actors

3 he normalised load actor or UK onshore wind arms declines rom a peak o about 24 at

age 1 to 15 at age 10 and 11 at age 15 he decline in the normalised load actor or Danish

onshore wind arms is slower but still significant with a all rom a peak o 22 to 18 at age 15

On the other hand or offshore wind arms in Denmark the normalised load actor alls rom

39 at age 0 to 15 at age 10 he reasons or the observed declines in normalised load actors

cannot be ully assessed using the data available but outages due to mechanical breakdowns

appear to be a contributory actor

4 Analysis o site-specific perormance reveals that the average normalised load actor o new UK

onshore wind arms at age 1 (the peak year o operation) declined significantly rom 2000 to2011 In addition larger wind arms have systematically worse perormance than smaller wind

arms Adjusted or age and wind availability the overall perormance o wind arms in the UK

has deteriorated markedly since the beginning o the century

5 hese findings have important implications or policy towards wind generation in the UK First

they suggest that the subsidy regime is extremely generous i investment in new wind arms

is profitable despite the decline in perormance due to age and over time Second meeting

the UK Governmentrsquos targets or wind generation will require a much higher level o wind

capacity ndash and thus capital investment ndash than current projections imply hird the structure

o contracts offered to wind generators under the proposed reorm o the electricity market

should be modified since ew wind arms will operate or more than 12ndash15 years

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 9

The Performance of Wind Farms in the United Kingdom

and Denmark

Introduction

1 Any assessment o the costs o wind power must rely heavily upon assumptions about theaverage load actor that will be achieved by new wind installations over their lietime It is

standard practice to calculate average load actors by year and country or onshore and offshore

installations as shown in able 1 (page 40)1 However such estimates do not provide a

reliable statistical basis or assessing the uture perormance o wind arms in aggregate Part

o the reason is that the amount o wind in any month or year is influenced by long term mete-

orological cycles that have periods o many years notably the North Atlantic Oscillation In

addition average load actors do not allow or changes in the composition o wind installations

by location age size and other actors

2 As the number o wind arms operating in developed countries has grown engineers have an

opportunity to analyse operating experience over extended periods o time With any (relatively)

novel technology the incidence o temporary or permanent breakdowns may be expected to all

over time so it is difficult to disentangle the effects o age on operating perormance rom the

technological immaturity o turbines installed 10 or 20 years ago Nonetheless the technology

or onshore wind turbines has been reasonably mature since the early 2000s his is documented

by data produced by the US Department o Energy which shows that the decline in capital costs

that is characteristic o immature technologies slowed afer 2000 and was reversed afer 20042

3 With at least 10 years o operating data since 2000 it should be possible to examine how thetypical operating perormance o wind installations in the United Kingdom and Denmark

varies with the age o the turbines he estimated age-perormance curves should then be

incorporated in estimates o levelised costs which are ofen used to compare the costs o wind

and other orms o generation

4 he Appendix provides a detailed description o the data and methods employed or this study

In both the UK and Denmark wind operators have a large incentive to produce electricity

whenever there is sufficient wind available since marginal operating costs are small while the

operator receives a much higher price In the UK this price is the sum o the market price plus

the value o (a) the Renewable Obligation Certificates (ROCs) awarded or each MWh o elec-

tricity produced and (b) the associated exemption rom the Climate Change Levy In practice

the effective market value o ROCs and associated incentives means that an onshore wind

operator earns roughly double the average market price o electricity per MWh In Denmark

most onshore wind capacity receives a price premium equivalent to about pound28 per MWh on top

1 he load actor is calculated as the ratio o the amount o electricity actually produced by a turbine or wind arm

over a period o a month or a year divided by the amount o output that would have been produced had it operated

at ull nameplate capacity or the entire period his is expressed as a percentage so that reported load actors lie

between 0 and 100

2 R Wiser amp M Bolinger (2012) 2011 Wind Technologies Market Report Lawrence Berkeley National Laboratory

US Department o Energy It should be noted that the installed cost o wind plants has allen since 2010 but these

changes reflect the cyclical orces o demand and supply that are well established across the electricity sector and

which apply to ossil-uel as well as renewable generators

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1252

10 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

o the market price or about 125 years o operation at a typical load actor3 Offshore capacity

is eligible or a similar price premium but most o it is contracted via tenders that offer a guar-

anteed eed-in tariff he tender tariffs have varied rom pound56 per MWh to twice that figure

extending over 16-18 years o operation at a typical load actor

Age-performance curves

5 A standard indicator o the operating perormance o a wind turbine is its normalised load

actor defined as the total output over some period divided by the maximum potential output

adjusted or wind availability Normalisation or wind availability is not straightorward I

detailed data on wind speeds at specific locations is available then the potential output rom

individual turbines or wind arms can be calculated using standard power curves Since wind

speeds can vary substantially over small distances such calculations are both difficult to imple-

ment and prone to errors when applied to large numbers o wind generators using public

sources o data4

6 Many analyses work in the opposite direction For example the Danish Wind Index ndash the

longest running series o data on wind availability in Europe ndash is constructed rom the common

component o monthly variations in wind output rom a large sample o wind turbines5 his

study adopts a similar approach Wind availability is treated as an unobserved actor whose

contribution to monthly variations in wind output is estimated by a series o fixed period

(monthyear) effects in a statistical model o monthly output or all wind arms he reasons

or adopting this approach are discussed in Section D o the Appendix his explains why the

method must perorm better than the easible alternative o using a single index o average

wind speed or the UK or Denmark

7 he analysis relies upon standard techniques that are widely used in the biological and medical

sciences as well economics and engineering Full details are given in the Appendix he idea is

to treat the actual load actor (ie without adjustment or wind conditions) or a specific wind

generator in any time period as being determined by a set o components which reflect specific

site conditions and the age-related perormance o wind plants in general together with the

period fixed effects and an uncorrelated random error he onshore wind datasets or the UK

and Denmark used or the analysis are large with monthly observations on 282 installations in

the UK and 823 installations in Denmark with an age range rom 0 to 19 years he offshore

wind dataset or Denmark is rather smaller covering only 30 installations but it can be used to

obtain reasonable estimates o perormance up to age 10 he results discussed here are based

upon least squares estimation (see page 27) combined with standard errors that are robust to

various departures rom classical assumptions about the nature o the data generation process

8 he results o the statistical analysis demonstrate an unambiguous and statistically significant

decline in the operating perormance o wind arms as they grow older Figure 1 shows the

3 Conversions to GBP are based on an exchange rate o pound1 = DKK 918 (as at 19 October 2012)

4 All wind generators collect their own inormation on wind speeds in order to manage their plants but they do not

publish such inormation and will not have access to inormation collected by other operators

5 Details o the construction o the Danish Wind Index are given at wwwvindstatdk A paper by Boccard ndash N Boccard

(2009) lsquoCapacity actor o wind power realized values vs estimatesrsquo Energy Policy Vol 37 pp 2679-88 ndash includes a

discussion o the construction o similar indices o wind availability in Germany the UK Netherlands and Sweden

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 11

simplest variant o the normalised age-perormance curve or onshore UK wind installations

together with the equivalent curves or onshore and offshore Danish installations he curves

illustrated are calculated rom the multiplicative error components model described in the

Appendix with log(load actor) as the dependent variable and a quadratic in plant age to repre-

sent the variation in plant perormance with age9 he rate o decline in perormance is greatest or offshore installations in Denmark with a all

rom load actors o over 40 at ages 0 and 1 to less than 15 by at ages 9 and 10 Onshore

installations in the UK show a more rapid rate o decline ndash 09 percentage points per year over

the first 10 years o operation ndash than is the case or Denmark though the normalised Danish

perormance curve lies below the UK curve or the first our years For the UK the normalised

load actor alls to just over 15 at age 10 and to 11 at age 15 With such low load actors it

seems likely that many wind arms will be re-powered ndash ie the turbines will be replaced ndash once

they reach the age o 10 or at most 15

Figure 1 Performance degradation due to age using equal weights

0

5

10

15

20

25

30

35

40

45

50

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

L o

a d f a c t o r ( )

Age (years)

UK onshore DK onshore DK offshore

Note Normalised load actors in Source Authorrsquos estimates

10 here are two plausible explanations or the observed decline in average load actors as wind

arms age he first is that the turbines become less efficient over time as a result o mechanical

wear and tear erosion o the turbine blades and related actors he second is that the turbines

experience more requent breakdowns and their operators take more time to bring them back

into service because they are less concerned about the perormance o older plants Both

reasons may be relevant in different circumstances and it is not possible to identiy a primary

explanation rom the data he requency o extended shutdowns does seem to increase with

age but this could be a reflection o the timing o planned maintenance operations rather than

breakdowns Whatever the cause the reduction in perormance with age is much greater than

would be expected or thermal generating plants

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1452

12 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

11 he age-perormance curves in Figure 1 are derived rom equations estimated by giving equal

weight to each wind arm irrespective o its generating capacity Figure 2 illustrates similar

curves but in this case the estimates are constructed by weighting each wind arm by its generat-

ing capacity reerred to as capacity-weighted his gives a better representation o perormance

degradation per MW o generating capacity he striking result is that the rate o decline inthe perormance o UK onshore wind installations is significantly aster when capacity weights

are used which implies that large wind arms in the UK experience a more rapid decline than

smaller ones he differences are smaller in Denmark but the capacity-weighted decline in

perormance is more rapid than the equal weights decline or Danish onshore installations

Figure 2 Performance degradation due to age using capacity weights

0

5

10

15

20

25

30

35

40

45

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

L o a d f a c t o r ( )

Age (years)

UK onshore DK onshore DK offshore

Note Normalised load actors in Source Authorrsquos estimates

Differences between wind farms

12 he specification o the statistical model means that the unit fixed effects identiy the specific

perormance characteristics o individual wind arms afer adjusting or wind availability and

age hese characteristics may reflect the site andor design o the installation as well as the

way in which it is operated A positive value or the unit fixed effect is equivalent to shifing

the perormance curve up by some percentage applied uniormly at each age while a negative

value shifs it down It ollows that the distribution o unit fixed effects across installations can

be used to compare the relative effectiveness o new installations by location date o commis-

sioning etc Care is required when making comparisons between or example UK and Danish

onshore wind installations because these have been normalised separately so the unit fixed

effects reer to different underlying averages6

6 Some care is required in interpreting the unit fixed effects By deault the average value o the unit fixed effects or

each group is zero but these averages are taken over all observations or each installation in the basic sample For the

purpose o the comparisons in Figure 3 the unit fixed effects have been adjusted so that the average values or each

group are zero when using one value or each installation

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 13

13 o give a sense o the numbers unit fixed effects o (a) 02 or (b) -02 mean that the base load

actor given the age o the plant is multiplied by (a) exp(02) = 122 or (b) exp(-02) = 082 Since

the base load actor or a plant o age 0 is 231 these fixed effects translate to load actors in

the first year o operation o (a) 282 and (b) 189 For a wind arm with 50 MW o gener-

ating capacity the difference amounts to an additional 40700 MWh o electricity output peryear Using the standard figures cited by RenewableUK this would be sufficient to supply 12300

homes or a year his range (040) is slightly greater than the interquartile range o unit fixed

effects or UK onshore wind arms (035) Again or comparison operators o thermal power

plants would regard any difference o even 10 up or down in efficiency relative to what is

expected as a striking and perhaps worrying indication o either good or bad perormance

Figure 3 Range of unit fixed effects by category

- 1

- 5

0

5

1

A d j f i x e d e f f e c t ndash q u a d r a t i c

UK onshore DK onshore DK offshore

Source Authorrsquos estimates

14 A notable eature o the distributions o the unit fixed effects that are summarised in the box

plots shown in Figure 3 is the range o variation between the best and worst perormers espe-

cially or UK onshore and Danish offshore wind arms Box plots are constructed so that the

top and bottom o the box correspond to the upper and lower quartiles while the bars at topand bottom correspond to the maximum and minimum values excluding outliers7 A plant at

the upper quartile o the distribution o UK onshore wind arms will generate an annual output

that is about 40 higher than a plant at the lower quartile he equivalent figure or Danish

onshore wind plants is an increase o 30 while or Danish offshore wind arms the difference

between the upper and lower quartiles is equivalent to 25 times the output o a lower quartile

plant By any standard such differences are important

7 he upper quartile is the point in a distribution such that the 25 o values exceed it while the lower quartile is the

point in a distribution such that 75 o values exceed it he line in the middle o the box marks the median the

point exceeded by 50 o values Outliers are defined as values that lie outside the range o ukey adjacent values

which are defined as the upperlower quartiles +- 15 x interquartile range

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1652

14 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

15 Outside the range between the upper and lower quartiles the top and bottom segments o

the distributions extend urther up and down or UK onshore wind arms than or those in

Denmark he rather different pattern or Danish offshore wind installations is partly a conse-

quence o the relatively small size o the sample but the evidence points to considerably greater

variability in the perormance o offshore wind installations than or onshore installations16 It is somewhat surprising to observe the magnitude o the differences in the perormance in

onshore wind arms in the UK Given the nature o the subsidy regimes and the high capital

cost o developing new installations it might be expected that operators have a strong incen-

tive to identiy the best locations and then choose equipment that will deliver the maximum

amount o electricity output at a high level o reliability For UK onshore operators it seems that

good or bad perormance is somewhat o a lottery However i the subsidies provided by ROCs

are sufficient to underwrite investment in inefficient plants ndash as appears to be the case ndash then

those subsidies are extremely generous or plants that operate close to the efficient rontier

As location is likely to be the main actor that determines the perormance o a specific plant

relative to all other plants the inerence must be that many wind plants have been developed on

sites with poor wind characteristics

Figure 4 Initial load factors by year of commissioning for onshore wind

10

15

20

25

30

35

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

I n i t i a l l o a d f a c t o r ( )

Year commissioned

UK onshore DK onshore Linear (UK onshore) Linear (DK onshore)

Source Authorrsquos estimates

17 here is another disturbing characteristic o UK wind arms which is revealed by examination

o the unit fixed effects and illustrated in Figure 4 It might be expected that the average peror-

mance o recently installed units would be higher than that o older units his should reflect

improvements in turbine design and reliability as well as in the identification o good sites or

wind generation However that is not the pattern or onshore wind arms in the UK he figure

shows the average values o the initial load actors calculated using equally weighted data by

year o commissioning or onshore wind arms in the UK and Denmark together with fitted

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 15

trend lines8 here has been a clear decline in the initial load actors in the UK which has not

occurred in Denmark his pattern is confirmed by more detailed statistical results shown in

the Appendix which take account o location and size o installation

18 he decline in initial perormance is particularly obvious afer 2005 when the rate o commis-

sioning new capacity increased sharply rom less than 500 MW in the 5 years 2000ndash04 to morethan 500 MW per year rom 2005 onwards In contrast the peak years or commissioning new

onshore wind arms in Denmark were 2000ndash02 when 920 MW was installed he rate o new

building (or replacement) picked up again in 2008 but it is still much lower than at the turn o

century Not only is the perormance o the onshore wind plants in the UK commissioned in

recent years significantly worse than that or wind arms commissioned beore 2005 but it is

possible that this is a direct consequence o an overly rapid expansion o capacity In essence

the evidence suggests that the industry does not have the capability to identiy develop and

operate new onshore wind arms at the rate envisaged by UK government targets while main-

taining a satisactory average level o perormance

Implications for future policy and performance

19 he pattern o perormance degradation with age and over time identified in this analysis has

implications or the assessment o new investment in wind arms and or the design o the

proposed Electricity Market Reorm (EMR) here are two related issues that are affected by

the results

a) What are the implications or the design o policies intended to promote the adoption

and expansion o wind generation in the UK his includes the structure o current

subsidies as well as the specification o the contracts that are proposed under the EMR A

related question concerns the expected lie o new wind arms

b) How might the average perormance o wind generation develop over the next decade

his is absolutely critical to any assessment o the amount o wind capacity that will be

required to meet the UK governmentrsquos targets or the share o electricity ndash and more

generally energy ndash rom renewable sources he government has assumed that the

average load actors or both onshore and offshore wind arms will either remain stable

or increase in uture he prospects will be very different i these assumptions are not

borne out

20 he first question can be addressed by considering the discounted sum o cumulative net

output rom a new wind arm on the assumption that its perormance matches the averages

identified in the analysis his takes account o both perormance degradation over time plus

the annual cost o maintaining the plant which may be expressed as a deduction rom gross

production so as to calculate the contribution o expected output at age s to recovering the

original capital cost o building the plant hese figures have to be discounted back to the date

o commissioning in order to assess the net present value o the initial investment in the plant

he curves or constant and observed perormance using a discount rate o 9 (a typical cost o

capital or wind generation) are shown in Figure 5 he black line is based on the ull set o age

8 he initial load actors are calculated as the normalised load actor or UKDanish wind arms at age 0 multiplied by

where is the unit fixed effect or installation i

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1852

16 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

effects while the red dashed line is based on the smoothed quadratic perormance curve and

the blue dashed line shows what the pattern would be i operating perormance was not affected

by age he calculation is extended out to age 25 since DECCrsquos standard comparisons assume

that wind turbines have a lie o 25 years

Figure 5 Impact of performance degradation on discounted cumulative output

0

10

20

30

40

50

60

70

80

90

100

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

o

f d i s c o u n t e d c u m u l a t i v e o u t p u t

Age of plant (years)

Constant Quadratic Full age effects

Source Authorrsquos estimates

21 he key points in the figure are that 80 o the discounted cumulative output o a new wind

plant is likely to be produced in the first 10 years o its lie and 90 in the first 14 years his

is consistent with the structure o Danish subsidies or onshore wind arms which extend over

a typical period o 125 years Since sites or wind arms are scarce and involve the payment o

significant rents that may be linked to output it is very unlikely that any new installation o

wind turbines will have an expected lie o more than 15 years Instead as has happened in the

past wind operators will have a strong incentive to decommission plants afer no more than 15

years and replace the turbines with newer equipment

22 As a consequence any economic assessment o wind generation should not be based on an

expected lie which is longer than 15 years In recent work reported in evidence to the House

o Commons Select Committee on Energy and Climate Change I assumed that wind plants

would have a residual value equal to 20 o their initial cost in real terms at the end o 15 years

he analysis in this paper suggests that this is too avourable an assumption Given the costs o

decommissioning old turbines the residual value is likely to be well below 10 o their initial

cost and the decision point may be at 10 rather than 15 years

23 he corollary o this observation is that it makes little sense to offer long term contracts or 20

years or more that guarantee prices or eed-in tariffs (Fis) to wind operators A contract length

o 10 to 12 years would be sufficient to remove most o the market risk associated with invest-

ment in wind generation In this respect the subsidy arrangements implemented in Denmark

are better designed

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 17

24 At the same time the offer o subsidies andor guaranteed prices may have serious adverse

consequences or the efficiency o wind generation Returning to the unit fixed effects measur-

ing the perormance o wind plants commissioned afer 2005 only 28 out o 159 units have an

operating perormance that exceeds the average or the period hose 28 units account or 360

MW o capacity whereas the remaining 131 units account or 2810 MW o capacity Not onlyare recent plants less efficient than the average or the whole period but the plants which are

below-average in efficiency are typically larger than the more efficient ones and account or a

disproportionate share o the recent additions to generating capacity

25 In mid-2012 there was approximately 4500 MW o onshore wind capacity operating in the

UK By 2020 it is expected that there will be at least 10000 MW (based on proposed wind

arms with planning consent) and perhaps as much 15000 MW o wind capacity in operation

he government is relying upon a substantial contribution rom onshore wind arms towards

meeting its targets or renewable energy in 2020 o provide context the governmentrsquos targets

or renewable energy in 2020 speciy a total o 234 Wh o energy rom all renewable sources

o which 90ndash140 Wh is expected to come rom wind and biomass electricity generation9

he projections or biomass electricity are extremely vague and have to be reconciled with

an assumed increase o 3ndash4 times in the amount o biomass used separately or heat So in

concrete terms it is reasonable to assume that wind generation will account or at least 90 Wh

with 24ndash32 Wh rom onshore generation and the remainder rom offshore

26 he average load actor or onshore wind over the 10 year period rom April 2002 to March

2012 was 256 which is influenced by the large amounts o new capacity added rom 2005

onwards I this load actor were achieved in 2020 the total amount o onshore wind capacityrequired to meet the onshore generation target would be 107 to 143 GW he average load

actor or offshore wind installations (on an unchanged configuration basis) or the five year

period 2007ndash11 was 32010 I this load actor were achieved in 2020 the total amount o

offshore wind capacity required to meet the remainder o the 90 Wh target would be 206

to 235 GW though these estimates would all to 180 to 205 GW i the average load actor

matched the historic average value o 367 or Danish offshore wind installations

27 he question that has to be addressed is whether these load actors are consistent with (a) the

investment programmes that would be required to achieve the total amounts o generation

capacity implied by the governmentrsquos targets and (b) the age profiles o the perormance o

wind arms discussed in this paper he short answer is that this is simply not possible i the

initial load actor or new wind arms continues to decline at the rate observed the past decade

So even as a starting point it is necessary to assume that this underlying deterioration in wind

arm perormance ceases his would be a major change in itsel and it is not obvious what

actors might lead to such a result

28 he results illustrated below are based on a vintage model o perormance that takes account

o the actual expansion o capacity up to the end o 2011 plus whatever investment in new or

replacement turbines is required to achieve the target levels o wind capacity in 2020 o high-light the ull range o potential outcomes these targets are either 10 GW or 15 GW o onshore

9 Department o Energy and Climate Change ndash UK Renewable Energy Road Map July 2011 Figure 2

10 he equivalent average load actor or onshore wind arms was 264

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2052

18 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

wind and either 18 or 24 GW o offshore wind he levels o investment required to meet these

targets depend upon the expected lie o wind turbines For onshore turbines an expected lie

o 10 years is consistent with re-powering decisions over the last decade while an expected lie

o 15 years seems to be the maximum consistent with the age-perormance profile For offshore

turbines the analysis uses expected lives o 15 and 20 years

Figure 6A Projected load factors for UK onshore wind (total capacity of 10 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

Figure 6B Projected load factors for UK onshore wind (total capacity of 15 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

29 Figure 6A shows projected values o the average normalised load actors or UK onshore wind

arms in 2015 and 2020 using a target capacity in 2020 equal to 10 GW plus the age-peror-

mance profiles or the UK and Denmark (DK) Since vintage effects are important in this

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 19

calculation the profile o new or replacement investment affects the results In this case the

level o gross investment is assumed to be constant rom 2013 onwards at either 930 or 1220

MW per year ndash the range is defined by the assumed lie o turbines Achieving a target o 15

GW in 2020 (Figure 6B) would require a level o investment rom 2013 onwards o either 1540

or 1930 MW per year Such figures are 3ndash4 times the average level o investment over theperiod 2005ndash11 so the lower target may provide a better indication o what will happen

30 All o the estimates are calibrated to match the observed average load actor in 2007ndash11 so that

they should be interpreted as relative rather than absolute values Using the age-perormance

profile or UK onshore wind installations the average normalised load actor is likely to all by

about 19 rom 2007ndash11 to 2015 and by 23ndash28 to 2020 I the age-perormance curves or

Danish onshore installations were to apply the reduction in the average normalised load actor

would be considerably less but there would still be a all o up to 5 by 2020

31 In the medium term the all is greater i turbines operate or 15 years because the average age

o all turbines will be higher and thus the degradation in perormance due to age will have a

larger effect By 2020 this effect reduces the average normalised load actor rom 203 or a 10

year lie to 195 or a 15 year lie here is a complex economic and commercial trade-off here

o meet a target o 10 GW in 2020 it is necessary to increase the level o investment in onshore

capacity by 30 rom 2013 Almost all o this will take the orm o re-powering existing wind

arms11 he costs are airly high because this will usually involve new development consents

decommissioning the original turbines and replacing them with a smaller number o larger and

more powerul turbines

32 Estimates o re-powering costs are generally treated as being commercially sensitive but broadproject costs cited in specialist magazines suggest that the unit costs are likely to be at least

pound1000 per kW or more than pound300 million per year in aggregate On the other hand the increase

in the amount o energy generated would be about 12 Wh or a 10 year turbine lie rather

than a 15 year turbine lie though this does not allow or the loss o production during re-pow-

ering Under the best circumstances the payback period may be quite short but the best option

rom a broader economic and social perspective is not obvious

33 Even under the best scenario 10 GW o onshore wind capacity will only generate 178 Wh o

electricity in 2020 well below the lower end o the target range or onshore wind generation

specified in DECCrsquos road map In act allowing or interactions between the age-perormance

profile and investment the amount o onshore wind capacity required in 2020 to generate 24

Wh would be a minimum o 134 GW At the top end o DECCrsquos target range the capacity

required in 2020 to generate 32 Wh would be a minimum o 178 GW Since these estimates

are based on a 10 year turbine lie the level o investment required rom 2013 onwards would

be 1650 MW per year or the bottom o the range and 2350 MW or the top end o the range

In the last 8 years total investment in onshore wind has only ever exceeded 600 MW in one

11 Re-powering is usually defined as the ull replacement o existing turbines whereas retrofitting covers the replace-

ment o various mechanical or electrical components In the data analysis re-powering involves the creation o a

new unit whereas retrofitting affects the perormance o a continuing unit Hence the potential benefits o retrofit-

ting are taken into account in constructing the age-perormance curves

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2252

20 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

year (883 MW in 2008) so it seems rather unlikely that even the lower end o DECCrsquos range or

onshore wind generation will actually be realised

34 O course this assessment would be different i UK onshore wind operators were able to match

the Danish age-perormance curve here are many actors which differentiate the UK and

Danish wind industries he average scale o wind arms in the UK is larger they are moreremote and there appears to be significantly greater variation between operating and potential

new sites in achieved perormance It would thereore require a very optimistic observer to

conclude that the difference in age-perormance curves between the two countries will disap-

pear in the short or medium term

Figure 7 Projected load factors for UK offshore wind in 2020

20

22

24

26

28

30

32

34

18 GW ndash life 15 yrs 18 GW ndash life 20 yrs 24 GW ndash life 15 yrs 24 GW ndash life 20 yrs

L o a d f a c t o r ( )

2007ndash11 2015 2020

Note Based on Danish age-perormance curves Source Authorrsquos estimates

35 Figure 7 shows the results rom carrying out a similar exercise or offshore wind arms in the

UK he only evidence on the age-perormance curves or offshore installations is based upon

the Danish experience so these have been used to construct the projections Further the profile

o investment assumes a progressive increase in additions to capacity rom 2013 to 2016 and

then a stable level o investment rom 2017 to 2020 he DECC road map involves a very large

commitment to offshore wind investment ndash in the range 25-35 GW o new capacity annually

rom 2016 onwards As a consequence the average age o offshore capacity in 2020 will be quite

low and the ull effects o the age-perormance curves will only be elt during the decade ollow-

ing 2020 Even so the average normalised load actor in 2020 will be about 15 lower than or

2007ndash11 while by 2025 it will be about 30 below the base level

36 On these projections 232 GW o offshore capacity is the minimum necessary to generate 58

Wh o electricity rom offshore wind in 2020 his requirement will increase to 298 GW by

2025 simply to offset the effect o the ageing o offshore generation capacity on total output

37 Combining the estimates or onshore and offshore wind generation the total amount o wind

capacity required in 2020 to produce 90 Wh o electricity rom wind will be in the range 39

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 21

to 41 GW once allowance is made or the observed age-perormance curves or both onshore

and offshore installations hese estimates are roughly a third higher than official predictions o

28ndash31 GW o wind capacity in 2020

38 urning the figures round gives a slightly different picture o the reality o the governmentrsquos

strategy on renewable electricity generation A total o 30 GW o wind capacity in 2020 (12GW onshore 18 GW offshore) will generate about 64 Wh o electricity in a normal year

Since electricity generation rom wind and biomass are expected to account or 50 o renew-

able energy in 2020 the consequence is that biomass is expected to account or 53 Wh In

2011 landfill gas and sewage sludge digestion accounted or 44 o electricity rom bioenergy

(DUKES ables 64) here are limited opportunities to expand production rom these sources

so a continuation o historic rates o growth would mean that renewable generation rom other

sources o bioenergy would have to increase rom 72 Wh in 2011 to 46 Wh in 2020 here

are severe constraints on increasing the amounts o electricity generated rom animal biomass

biodegradable municipal solid waste and anaerobic digestion so it would be optimistic to

assume that these will contribute more than 6-8 Wh in 2020 (up rom 26 Wh in 2011)

39 In practice i the projections or renewable energy are to be believed they rest upon an assump-

tion that there will be a large increase in the amount o electricity that is generated rom plant

biomass ndash either on its own or through co-firing with coal his increase will involve increas-

ing the amount o electricity generated in these ways rom 46 Wh in 2011 to 38-40 Wh in

2020 Since the amounts o straw and UK-grown timber available or this purpose are strictly

limited this will come down to the amount o wood chips that can be imported and the cost o

doing this40 he historic average load actor or biomass plants is just over 60 so about 7500 MW o

biomass capacity will be required to supply 40 Wh o electricity he weight o wood chips per

MWh o electricity depends upon moisture content storage boiler design and other actors but

an indicative range is 085 to 115 tonnes per MWh o wood chips with a moisture content o

30 According to DECC and Forestry Commission estimates the UK imported 265 million

tonnes o steam coal and 26 million tonnes o other wood products (including wood chips)

in 2011 It seems that the net effect o the governmentrsquos strategy will be to replace 15 million

tonnes o coal imported in 2011 by 32 to 44 million tonnes o imported wood chips his is a

somewhat strange way o increasing energy security

Conclusion

41 Wind power is a highly capital-intensive technology or generating electricity Its merits rely

entirely upon a substitution o capital or uel inputs he same is true or hydro or tidal or wave

power In comparison with hydro power wind is a low quality resource because o its variability

and because it cannot be stored So the economic case or wind power must rest on obtaining

the most out o the wind that is available

42 While the decline in the achieved perormance o onshore wind turbines in Denmark is

much less than that or the UK or offshore nonetheless the decline in expected output under

standardised wind conditions over 10 years is 10 unweighted and 13 capacity weighted

hese declines accelerate afer age 10 so that the reductions in perormance are 17 and 20

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2452

22 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

respectively afer 15 years For UK onshore wind arms the reduction in perormance due to

age is much worse at 27 unweighted and 69 capacity weighted by age 10

43 Evidence on the perormance o Danish offshore installations is both restricted and so poor

that there may be concern that the results are affected by a small number o outliers Still the

sample contains a reasonable number o sites with at least 5 years o operating experience andthe decline in perormance by age 5 is 38 unweighted and 26 capacity weighted

44 In addition to these results there is strong evidence that the average normalised load actor or

new onshore wind installations in the UK has allen significantly over the period rom 2000 to

2011 his is consistent with a pattern in which the most avourable sites are developed first

Equally it could mean that wind developers have been unable to keep up with the rate o new

investment while maintaining the quality o development and operations For example the site

design or selection o turbine characteristics may make less effective use o the available wind

resources or the sites available than was the case in the past

45 Whatever the reasons the deterioration in initial perormance means that the expected returns

rom the expansion in wind capacity both or investors and in terms o the reduction in CO983090

emissions have been alling without a concomitant decrease in the private and social costs

that are borne by customers and the general public Clearly this is unsatisactory at best and it

suggests that the benefits claimed or current policies cannot be taken at ace value

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 23

Appendix Data and Methods

A Data for the United Kingdom

he raw data used or this study was extracted by the Renewable Energy Foundation rom the

Renewables and CHP Register database compiled by Ogem12 his inormation is used in the

administration o the market in Renewable Obligation Certificates (ROCs) he Renewables

Obligation is the primary mechanism by which large scale generators o renewable electric-

ity receive subsidies so the operators o wind arms have a very strong incentive to submit

complete and up-to-date inormation to the database he data extracted covered all onshore

wind generators with at least 05 MW o generating capacity which are eligible or ROCs or the

10 year period April 2002 to March 2012 he key variables are the monthly output o electric-

ity (used or the purpose o allocating ROCs) and nominal generating capacity here are 282

reporting units in the final dataset A reporting unit may correspond to an entire wind arm orto separate phases o development Complications arise when an existing wind arm is re-pow-

ered ie when old wind turbines are replaced by newer and usually more powerul turbines

he database reports this as a change in the generating capacity o the recording unit but in

this analysis repowering is treated as the termination o the old recording unit and the creation

o a new recording unit

A considerable amount o data checking and cleaning was required beore the data could

be analysed he first step was to add the age o each reporting unit his was calculated by

reerence to the month in which the turbines were commissioned For wind arms in operation

in or beore June 2002 the commissioning date was obtained rom various online sources ndash

notably the thewindpowernet database o UK wind arms13 which was cross-checked against

the inormation provided on the websites o wind operators and other sources For recording

units whose first data relates to dates afer June 2002 the commissioning date was initially

assumed to be the month or which data is first reported Cross-checks were carried out or

all recording units which commenced reporting in the period July 2002 to December 2003 In

more than 80 o cases the externally-reported commissioning date is within one month o the

first month or which data is reported he exceptions ndash Bu Farm and Mablethorpe ndash appear to

be due to delays by small operators in reporting data to Ogem

In a number o cases it is clear rom the data that plants were not in ull operation at the

time when data was first reported to the database here are two indicators o partial or delayed

commissioning

bull For an initial series o months the calculated load actor is very low ndash usually below 5 ndash

and then increases abruptly In such cases the data or the early months is dropped and the

commissioning date is treated as the month immediately preceding the first month with a

lsquonormalrsquo load actor14

12 httpswwwrenewablesandchpogemgovuk

13 See httpwwwthewindpowernetwindarms_list_enphp

14 he month in which a plant is reported as having been commissioned is not included in the analysis

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2652

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 25

the decline in the plantrsquos load actor as it ages As an example consider a plant whose reported

capacity was increased rom 99 MW to 10 MW at age 3 I this was simply a change in report-

ing practice this would reduce the estimated load actor rom age 3 onwards By increasing

the initial capacity rom 99 MW to 10 MW ndash or by reducing the capacity rom 10 MW to 99

MW rom age 3 onwards ndash this bias is removed It is not important which adjustment is madebecause the difference between the alternative assumptions is captured by the unit fixed effect

he numbers o separate sites and o observations per site mean that it is not possible to

provide a simple visual representation o the data Figure 8 provides a summary by illustrating

the distributions o load actors by the age o the wind arm pooled across all years he ranges

o the load actors or each age are wide while the medians show no clear trend by age hus a

naiumlve analysis might conclude that there is nothing to investigate However as will be explained

below the distributions mask crucial differences between the perormance o wind arms as

they age because they do not control or the differences in location and wind availability over

time

B Data for Denmark

he data or Danish wind arms used in this study comes rom a database compiled by the

Danish Energy Agency covering the characteristics and perormance o all wind turbines rom

2002 up to the end o August 201216 he basic recording unit or the register is the individual

wind turbine but where a number o turbines are connected to a single meter an identical

average output per turbine is recorded or all turbines in the group he presence o what are in

effect repeated observations increases the noise in the data In addition the model that is used

or the analysis assumes that site operator and turbine characteristics are common or all o theunits that make up a single wind arm o address this point separate wind arms have been

identified by grouping all turbines which have identical values or turbine manuacturer and

type local authority date o commissioning and date o decommissioning (where applicable)

he register contains inormation on 7569 turbines commissioned rom 1977 onwards his

includes a large number o small turbines which would not all within the scope o the ROC

scheme in the UK Hence or comparability with the UK data all turbines commissioned

prior to 1992 were dropped as well as all wind arms with a generating capacity o less than 1

MW he resulting sample covered 823 onshore wind arms and 30 offshore wind arms with

monthly load actors or various periods rom January 2002 to August 2012 Given the cut-off

or wind arms commissioned prior to 1992 the maximum age or an onshore wind arm was 20

years and the maximum age or an offshore wind arm was 17 years

he total capacity o the onshore wind arms recorded in the Danish database was 2757 MW

he average size o an onshore wind arm in Denmark is small ndash only 33 MW as compared

with 161 MW or the UK ndash because 84 o the sample consists o installations with between

1 and 4 turbines he largest onshore installation has 39 turbines with a capacity o 234 MW

while the onshore installation with the greatest capacity has 19 turbines with a capacity o 584

MW Overall the sample o Danish onshore wind arms is older and smaller than the sample

o UK onshore wind arms reflecting the much longer history o wind generation in Denmark

16 httpwwwensdkEN-USINFOFACSANDFIGURESENERGY_SAISICS_AND_INDICAORS

OVERVIEWOFHEENERGYSECORREGISEROFWINDURBINESSiderForsideaspx

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2852

26 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

he larger scale o UK wind arms may account or some o the differences in perormance

curves or the two countries

he sample o offshore wind arms in Denmark is quite small with 30 installations in

total Still it is useul to analyse the perormance o these installations since there is very little

comparable data or the UK he sample covers 394 turbines with an average o just over 13turbines per installation and an average capacity o 288 MW he largest installation which

was commissioned in 2003 has 72 turbines with a total capacity o 1656 MW Again offshore

installations are smaller and older in Denmark than is the current pattern in the UK he first

two wind arms in the sample were commissioned in the years 1995 and 2000 but the sample

increases significantly rom 2002 onwards o avoid reliance upon a sample o 1 or 2 wind

arms to determine specific coefficients or ages gt 10 years all observations with an age greater

than 10 were omitted rom the analysis17

A somewhat different problem with the offshore wind arm data arose because the coeffi-

cients estimated using capacity weights (see below) were heavily influenced by the largest windarm his has had an extremely erratic profile o perormance over time with a very low average

load actor he treatment o this wind arm is discussed in more detail in Section F below

C Specication and estimation methods

As Figure 8 illustrates it is necessary to go beyond simple summary statistics in order to assess

whether there is any systematic relationship between age and the average perormance o wind

arms he perormance o wind arms varies over both time and space some months and years

have more or less wind than the long term average while specific characteristics o wind arms

ndash including location type o turbine and operating regime ndash will influence the perormance oeach plant under identical wind conditions

he combination o (a) site-specific characteristics which are constant over time and (b)

period-specific characteristics which are constant across wind arms can be represented by

what statisticians call an error components model with fixed effects or each wind arm and

each time period he dependent variable in the model consists o a sequence o load actors ndash

denoted by or unit i in month t ndash or each wind arm (the panel unit) over time he load

actors are calculated by dividing the total output in MWh by (24 x number o days in month x

reported generating capacity in MW) and are multiplied by 100 to convert to percentages

he additive version o the error components model may be written as

(1)

in which denotes the age o plant i in period t and is some unction that is either a

representation o or an approximation to the normalised age-perormance relationship he

multiplicative version is similar except that is replaced by where ln() denotes the

natural logarithm he terms are wind arm or unit fixed effects which capture actors such

as location type o turbine operating regimes etc he terms are period fixed effects which

capture the influence o wind conditions over the UK or Denmark as a whole as well as seasonal

maintenance and demand Without loss o generality it may be assumed that the fixed effects are

17 For the avoidance o doubt this means that no attempt was made to estimate the effect o age on the perormance o

offshore wind arms beyond age 10 while the data or the oldest wind arms was included in the analysis up to and

including age 10

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 27

normalised so that ie all o the error components have a zero mean while the

mean o the random error is zero by definition he terms are random errors which capture

random variations in wind conditions that are not specific to the site turbine breakdowns and

other actors that are uncorrelated with either the site or the time period It is assumed that the

mean o the random error is zero while additional assumptions depend upon the method o

estimation that is adopted

he model specified in equation (1) can be estimated in a variety o ways18 A crucial issue is

whether the are or may be significantly correlated with the age terms or any other independent

regressors I it is assumed that there is no such correlation the model is usually reerred to as

a random effects model In that case the equation can be estimated by a variant o least squares

(LS) generalised least squares (GLS) generalised methods o moments (GMM) or maximum

likelihood (ML) ndash all o which should be yield parameter estimates that will converge in prob-

ability to the true parameter values as the sample size increases An alternative but less restric-

tive specification is known as the fixed effects model in which no assumption is made about

the correlation between the unit fixed effects and other regressors he limitation o the fixed

effects model is that it cannot be used to estimate parameters or regressors that vary across

panel units but are constant over time or a panel unit (time-invariant regressors) ndash eg the

year in which a wind arm was commissioned its location or its rated generating capacity he

influence o such variables is captured by the unit fixed effects

In order to avoid the risk o obtaining biased estimates o the age effects the results reported

in this note are based upon estimates using the fixed effects model he influence o time-invar-

iant regressors is examined in a second stage by regressing the estimates o the unit fixed effectson these variables he simplest method o estimating the fixed effects model is known as the

lsquowithinrsquo estimator in which the mean value or each unit is subtracted rom all observations or

that unit For example i is a general polynomial o order M in A equation (1) becomes

(2)

so that the model can be estimated by least square with the inclusion o dummy variables or

each period t he variance-covariance matrix o the coefficients have been estimated by using

(a) a robust sandwich estimator adjusted or clustering by panel unit and (b) a bootstrap esti-

mator with 400 repetitions he estimates o the standard errors will be consistent even i the

random error component is serially correlated within panel units andor has different variances

across panel units19 Since the number o panel units and the average number o time periods

per panel units are both large (other than or Danish offshore wind arms) the least square esti-

mates o the parameters will be unbiased under these assumptions

he bootstrap method is expensive to carry out but it provides a useul cross-check on

the standard errors generated by other methods Unortunately the bootstrap method o

18 See Chapter 21 o AC Cameron amp PK rivedi ndash Microeconometrics Methods and Applications (Cambridge

Cambridge University Press 2005) or a standard textbook treatment o estimation methods or panel data models

19 he estimation was carried out using the xtreg procedure in Stata Version 12 he methods o constructing the

sandwich and bootstrap estimates o the variance-covariance matrix are described in Chapters 8 amp 13 o AC

Cameron amp PK rivedi ndash Microeconometrics using Stata (Revised Edition College Station exas Stata Press 2010)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 2: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 252

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 352

The Performance of Wind Farms

in the United Kingdom and Denmark

Gordon Hughes

2012

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 452

copy Gordon Hughes 2012

Cover background image by Simon Gray (httpwwwstar-oneorguk) via stockxchng

Published by the Renewable Energy Foundation

57ndash58 Russell Square

London WC1B 4HSwwwreorguk

he Renewable Energy Foundation is a registered charity in England and Wales (1107360)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 552

Contents

List o Figures and ables 4

About the author 5

About the Renewable Energy Foundation 5

Authorrsquos Note 6

Note on the Data 6

Executive Summary 7

Te Perormance o Wind Farms in the United Kingdom and Denmark 9

Introduction 9

Age-perormance curves 10

Differences between wind arms 12Implications or uture policy and perormance 15

Conclusion 21

Appendix Data and Methods 23

A Data or the United Kingdom 23

B Data or Denmark 25

C Specification and estimation methods 26

D Period fixed effects vs normalisation by wind speeds 28

E Estimation results or the UK 30F Estimation results or Denmark 36

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 652

List of Figures and TablesList of Figures and Tables

Figure 1 Perormance degradation due to age using equal weights 11

Figure 2 Perormance degradation due to age using capacity weights 12

Figure 3 Range o unit fixed effects by category 13

Figure 4 Initial load actors by year o commissioning or onshore wind 14

Figure 5 Impact o perormance degradation on discounted cumulative output 16

Figure 6A Projected load actors or UK onshore wind (total capacity o 10 GW in 2020) 18

Figure 6B Projected load actors or UK onshore wind (total capacity o 15 GW in 2020) 18

Figure 7 Projected load actors or UK offshore wind in 2020 20

Figure 8 Box plots o load actors by age or UK onshore wind arms 24

Figure 9A Additive age-perormance curves or UK onshore wind arms 32

Figure 9B Multiplicative age-perormance curves or UK onshore wind arms 32

Figure 10 UK onshore age-perormance curves using equal and capacity weights 33

Figure 11A Residuals by age or perormance curves using equal weights 34

Figure 11B Residuals by age or perormance curves using capacity weights 34

Figure 12 Unit fixed effects by year o commissioning or England and Scotland 35

Figure 13A Age-perormance curves or Danish onshore wind arms 37

Figure 13B Age-perormance curves or Danish offshore wind arms 37

Figure 14 Danish age-perormance curves using equal and capacity weights 38

able 1 Average load actors by year and country () 40

able 2 Estimation results or UK onshore wind arms 41

able 3 Equations or trends in unit fixed effects 43

able 4 Estimation results or Danish wind arms 44

able 5 Perormance by year o commissioning or Danish wind arms 46

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 752

About the author

Dr Gordon Hughes is a Proessor o Economics at the University o Edinburgh where he teaches courses in

the Economics o Natural Resources and Public Economics He was a senior adviser on energy and environ-

mental policy at the World Bank until 2001 He has advised governments on the design and implementationo environmental policies and was responsible or some o the World Bankrsquos most important environmental

guidelines

About the Renewable Energy Foundation

he Renewable Energy Foundation is a registered charity promoting sustainable development or the benefit

o the public by means o energy conservation and the use o renewable energy

REF is supported by private donation and has no political affiliation or corporate membership In pursuit o

its principal goals REF highlights the need or an overall energy policy that is balanced ecologically sensi-

tive and effective

We aim to raise public awareness o the issues and encourage inormed debate regarding a structured energy

policy that is both ecologically sensitive and practical he issues o climate change and security o energy

supply are complex and closely intertwined REF contributes to the debate surrounding these issues by

commissioning reports to provide an independent and authoritative source o inormation

For urther inormation see wwwreorguk

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 852

Authorrsquos Note

Note he views expressed in this paper are strictly personal and do not reflect the position o any organ-

isation with which I am associated I am grateul to John Constable Lee Moroney and a reeree or their

comments on earlier drafs o the paper he results o this study have been presented at seminars andlectures to various groups in Edinburgh and Glasgow I have benefited rom the questions and comments

rom participants in those events

Gordon Hughes

School o Economics

University o Edinburgh

gahughesedacuk

Note on the Data

For the convenience o interested researchers both the raw and cleaned data employed in this analysis will

published alongside the electronic version o the study on the REF website wwwreorguk

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 7

Executive Summary

1 Onshore wind turbines represent a relatively mature technology which ought to have achieved

a satisactory level o reliability in operation as plants age Unortunately detailed analysis o

the relationship between age and perormance gives a rather different picture or both theUnited Kingdom and Denmark with a significant decline in the average load actor o onshore

wind arms adjusted or wind availability as they get older An even more dramatic decline is

observed or offshore wind arms in Denmark but this may be a reflection o the immaturity

o the technology

2 he study has used data on the monthly output o wind arms in the UK and Denmark reported

under regulatory arrangements and schemes or subsidising renewable energy Normalised

age-perormance curves have been estimated using standard statistical techniques which allow

or differences between sites and over time in wind resources and other actors

3 he normalised load actor or UK onshore wind arms declines rom a peak o about 24 at

age 1 to 15 at age 10 and 11 at age 15 he decline in the normalised load actor or Danish

onshore wind arms is slower but still significant with a all rom a peak o 22 to 18 at age 15

On the other hand or offshore wind arms in Denmark the normalised load actor alls rom

39 at age 0 to 15 at age 10 he reasons or the observed declines in normalised load actors

cannot be ully assessed using the data available but outages due to mechanical breakdowns

appear to be a contributory actor

4 Analysis o site-specific perormance reveals that the average normalised load actor o new UK

onshore wind arms at age 1 (the peak year o operation) declined significantly rom 2000 to2011 In addition larger wind arms have systematically worse perormance than smaller wind

arms Adjusted or age and wind availability the overall perormance o wind arms in the UK

has deteriorated markedly since the beginning o the century

5 hese findings have important implications or policy towards wind generation in the UK First

they suggest that the subsidy regime is extremely generous i investment in new wind arms

is profitable despite the decline in perormance due to age and over time Second meeting

the UK Governmentrsquos targets or wind generation will require a much higher level o wind

capacity ndash and thus capital investment ndash than current projections imply hird the structure

o contracts offered to wind generators under the proposed reorm o the electricity market

should be modified since ew wind arms will operate or more than 12ndash15 years

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 9

The Performance of Wind Farms in the United Kingdom

and Denmark

Introduction

1 Any assessment o the costs o wind power must rely heavily upon assumptions about theaverage load actor that will be achieved by new wind installations over their lietime It is

standard practice to calculate average load actors by year and country or onshore and offshore

installations as shown in able 1 (page 40)1 However such estimates do not provide a

reliable statistical basis or assessing the uture perormance o wind arms in aggregate Part

o the reason is that the amount o wind in any month or year is influenced by long term mete-

orological cycles that have periods o many years notably the North Atlantic Oscillation In

addition average load actors do not allow or changes in the composition o wind installations

by location age size and other actors

2 As the number o wind arms operating in developed countries has grown engineers have an

opportunity to analyse operating experience over extended periods o time With any (relatively)

novel technology the incidence o temporary or permanent breakdowns may be expected to all

over time so it is difficult to disentangle the effects o age on operating perormance rom the

technological immaturity o turbines installed 10 or 20 years ago Nonetheless the technology

or onshore wind turbines has been reasonably mature since the early 2000s his is documented

by data produced by the US Department o Energy which shows that the decline in capital costs

that is characteristic o immature technologies slowed afer 2000 and was reversed afer 20042

3 With at least 10 years o operating data since 2000 it should be possible to examine how thetypical operating perormance o wind installations in the United Kingdom and Denmark

varies with the age o the turbines he estimated age-perormance curves should then be

incorporated in estimates o levelised costs which are ofen used to compare the costs o wind

and other orms o generation

4 he Appendix provides a detailed description o the data and methods employed or this study

In both the UK and Denmark wind operators have a large incentive to produce electricity

whenever there is sufficient wind available since marginal operating costs are small while the

operator receives a much higher price In the UK this price is the sum o the market price plus

the value o (a) the Renewable Obligation Certificates (ROCs) awarded or each MWh o elec-

tricity produced and (b) the associated exemption rom the Climate Change Levy In practice

the effective market value o ROCs and associated incentives means that an onshore wind

operator earns roughly double the average market price o electricity per MWh In Denmark

most onshore wind capacity receives a price premium equivalent to about pound28 per MWh on top

1 he load actor is calculated as the ratio o the amount o electricity actually produced by a turbine or wind arm

over a period o a month or a year divided by the amount o output that would have been produced had it operated

at ull nameplate capacity or the entire period his is expressed as a percentage so that reported load actors lie

between 0 and 100

2 R Wiser amp M Bolinger (2012) 2011 Wind Technologies Market Report Lawrence Berkeley National Laboratory

US Department o Energy It should be noted that the installed cost o wind plants has allen since 2010 but these

changes reflect the cyclical orces o demand and supply that are well established across the electricity sector and

which apply to ossil-uel as well as renewable generators

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1252

10 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

o the market price or about 125 years o operation at a typical load actor3 Offshore capacity

is eligible or a similar price premium but most o it is contracted via tenders that offer a guar-

anteed eed-in tariff he tender tariffs have varied rom pound56 per MWh to twice that figure

extending over 16-18 years o operation at a typical load actor

Age-performance curves

5 A standard indicator o the operating perormance o a wind turbine is its normalised load

actor defined as the total output over some period divided by the maximum potential output

adjusted or wind availability Normalisation or wind availability is not straightorward I

detailed data on wind speeds at specific locations is available then the potential output rom

individual turbines or wind arms can be calculated using standard power curves Since wind

speeds can vary substantially over small distances such calculations are both difficult to imple-

ment and prone to errors when applied to large numbers o wind generators using public

sources o data4

6 Many analyses work in the opposite direction For example the Danish Wind Index ndash the

longest running series o data on wind availability in Europe ndash is constructed rom the common

component o monthly variations in wind output rom a large sample o wind turbines5 his

study adopts a similar approach Wind availability is treated as an unobserved actor whose

contribution to monthly variations in wind output is estimated by a series o fixed period

(monthyear) effects in a statistical model o monthly output or all wind arms he reasons

or adopting this approach are discussed in Section D o the Appendix his explains why the

method must perorm better than the easible alternative o using a single index o average

wind speed or the UK or Denmark

7 he analysis relies upon standard techniques that are widely used in the biological and medical

sciences as well economics and engineering Full details are given in the Appendix he idea is

to treat the actual load actor (ie without adjustment or wind conditions) or a specific wind

generator in any time period as being determined by a set o components which reflect specific

site conditions and the age-related perormance o wind plants in general together with the

period fixed effects and an uncorrelated random error he onshore wind datasets or the UK

and Denmark used or the analysis are large with monthly observations on 282 installations in

the UK and 823 installations in Denmark with an age range rom 0 to 19 years he offshore

wind dataset or Denmark is rather smaller covering only 30 installations but it can be used to

obtain reasonable estimates o perormance up to age 10 he results discussed here are based

upon least squares estimation (see page 27) combined with standard errors that are robust to

various departures rom classical assumptions about the nature o the data generation process

8 he results o the statistical analysis demonstrate an unambiguous and statistically significant

decline in the operating perormance o wind arms as they grow older Figure 1 shows the

3 Conversions to GBP are based on an exchange rate o pound1 = DKK 918 (as at 19 October 2012)

4 All wind generators collect their own inormation on wind speeds in order to manage their plants but they do not

publish such inormation and will not have access to inormation collected by other operators

5 Details o the construction o the Danish Wind Index are given at wwwvindstatdk A paper by Boccard ndash N Boccard

(2009) lsquoCapacity actor o wind power realized values vs estimatesrsquo Energy Policy Vol 37 pp 2679-88 ndash includes a

discussion o the construction o similar indices o wind availability in Germany the UK Netherlands and Sweden

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 11

simplest variant o the normalised age-perormance curve or onshore UK wind installations

together with the equivalent curves or onshore and offshore Danish installations he curves

illustrated are calculated rom the multiplicative error components model described in the

Appendix with log(load actor) as the dependent variable and a quadratic in plant age to repre-

sent the variation in plant perormance with age9 he rate o decline in perormance is greatest or offshore installations in Denmark with a all

rom load actors o over 40 at ages 0 and 1 to less than 15 by at ages 9 and 10 Onshore

installations in the UK show a more rapid rate o decline ndash 09 percentage points per year over

the first 10 years o operation ndash than is the case or Denmark though the normalised Danish

perormance curve lies below the UK curve or the first our years For the UK the normalised

load actor alls to just over 15 at age 10 and to 11 at age 15 With such low load actors it

seems likely that many wind arms will be re-powered ndash ie the turbines will be replaced ndash once

they reach the age o 10 or at most 15

Figure 1 Performance degradation due to age using equal weights

0

5

10

15

20

25

30

35

40

45

50

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

L o

a d f a c t o r ( )

Age (years)

UK onshore DK onshore DK offshore

Note Normalised load actors in Source Authorrsquos estimates

10 here are two plausible explanations or the observed decline in average load actors as wind

arms age he first is that the turbines become less efficient over time as a result o mechanical

wear and tear erosion o the turbine blades and related actors he second is that the turbines

experience more requent breakdowns and their operators take more time to bring them back

into service because they are less concerned about the perormance o older plants Both

reasons may be relevant in different circumstances and it is not possible to identiy a primary

explanation rom the data he requency o extended shutdowns does seem to increase with

age but this could be a reflection o the timing o planned maintenance operations rather than

breakdowns Whatever the cause the reduction in perormance with age is much greater than

would be expected or thermal generating plants

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1452

12 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

11 he age-perormance curves in Figure 1 are derived rom equations estimated by giving equal

weight to each wind arm irrespective o its generating capacity Figure 2 illustrates similar

curves but in this case the estimates are constructed by weighting each wind arm by its generat-

ing capacity reerred to as capacity-weighted his gives a better representation o perormance

degradation per MW o generating capacity he striking result is that the rate o decline inthe perormance o UK onshore wind installations is significantly aster when capacity weights

are used which implies that large wind arms in the UK experience a more rapid decline than

smaller ones he differences are smaller in Denmark but the capacity-weighted decline in

perormance is more rapid than the equal weights decline or Danish onshore installations

Figure 2 Performance degradation due to age using capacity weights

0

5

10

15

20

25

30

35

40

45

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

L o a d f a c t o r ( )

Age (years)

UK onshore DK onshore DK offshore

Note Normalised load actors in Source Authorrsquos estimates

Differences between wind farms

12 he specification o the statistical model means that the unit fixed effects identiy the specific

perormance characteristics o individual wind arms afer adjusting or wind availability and

age hese characteristics may reflect the site andor design o the installation as well as the

way in which it is operated A positive value or the unit fixed effect is equivalent to shifing

the perormance curve up by some percentage applied uniormly at each age while a negative

value shifs it down It ollows that the distribution o unit fixed effects across installations can

be used to compare the relative effectiveness o new installations by location date o commis-

sioning etc Care is required when making comparisons between or example UK and Danish

onshore wind installations because these have been normalised separately so the unit fixed

effects reer to different underlying averages6

6 Some care is required in interpreting the unit fixed effects By deault the average value o the unit fixed effects or

each group is zero but these averages are taken over all observations or each installation in the basic sample For the

purpose o the comparisons in Figure 3 the unit fixed effects have been adjusted so that the average values or each

group are zero when using one value or each installation

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 13

13 o give a sense o the numbers unit fixed effects o (a) 02 or (b) -02 mean that the base load

actor given the age o the plant is multiplied by (a) exp(02) = 122 or (b) exp(-02) = 082 Since

the base load actor or a plant o age 0 is 231 these fixed effects translate to load actors in

the first year o operation o (a) 282 and (b) 189 For a wind arm with 50 MW o gener-

ating capacity the difference amounts to an additional 40700 MWh o electricity output peryear Using the standard figures cited by RenewableUK this would be sufficient to supply 12300

homes or a year his range (040) is slightly greater than the interquartile range o unit fixed

effects or UK onshore wind arms (035) Again or comparison operators o thermal power

plants would regard any difference o even 10 up or down in efficiency relative to what is

expected as a striking and perhaps worrying indication o either good or bad perormance

Figure 3 Range of unit fixed effects by category

- 1

- 5

0

5

1

A d j f i x e d e f f e c t ndash q u a d r a t i c

UK onshore DK onshore DK offshore

Source Authorrsquos estimates

14 A notable eature o the distributions o the unit fixed effects that are summarised in the box

plots shown in Figure 3 is the range o variation between the best and worst perormers espe-

cially or UK onshore and Danish offshore wind arms Box plots are constructed so that the

top and bottom o the box correspond to the upper and lower quartiles while the bars at topand bottom correspond to the maximum and minimum values excluding outliers7 A plant at

the upper quartile o the distribution o UK onshore wind arms will generate an annual output

that is about 40 higher than a plant at the lower quartile he equivalent figure or Danish

onshore wind plants is an increase o 30 while or Danish offshore wind arms the difference

between the upper and lower quartiles is equivalent to 25 times the output o a lower quartile

plant By any standard such differences are important

7 he upper quartile is the point in a distribution such that the 25 o values exceed it while the lower quartile is the

point in a distribution such that 75 o values exceed it he line in the middle o the box marks the median the

point exceeded by 50 o values Outliers are defined as values that lie outside the range o ukey adjacent values

which are defined as the upperlower quartiles +- 15 x interquartile range

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1652

14 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

15 Outside the range between the upper and lower quartiles the top and bottom segments o

the distributions extend urther up and down or UK onshore wind arms than or those in

Denmark he rather different pattern or Danish offshore wind installations is partly a conse-

quence o the relatively small size o the sample but the evidence points to considerably greater

variability in the perormance o offshore wind installations than or onshore installations16 It is somewhat surprising to observe the magnitude o the differences in the perormance in

onshore wind arms in the UK Given the nature o the subsidy regimes and the high capital

cost o developing new installations it might be expected that operators have a strong incen-

tive to identiy the best locations and then choose equipment that will deliver the maximum

amount o electricity output at a high level o reliability For UK onshore operators it seems that

good or bad perormance is somewhat o a lottery However i the subsidies provided by ROCs

are sufficient to underwrite investment in inefficient plants ndash as appears to be the case ndash then

those subsidies are extremely generous or plants that operate close to the efficient rontier

As location is likely to be the main actor that determines the perormance o a specific plant

relative to all other plants the inerence must be that many wind plants have been developed on

sites with poor wind characteristics

Figure 4 Initial load factors by year of commissioning for onshore wind

10

15

20

25

30

35

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

I n i t i a l l o a d f a c t o r ( )

Year commissioned

UK onshore DK onshore Linear (UK onshore) Linear (DK onshore)

Source Authorrsquos estimates

17 here is another disturbing characteristic o UK wind arms which is revealed by examination

o the unit fixed effects and illustrated in Figure 4 It might be expected that the average peror-

mance o recently installed units would be higher than that o older units his should reflect

improvements in turbine design and reliability as well as in the identification o good sites or

wind generation However that is not the pattern or onshore wind arms in the UK he figure

shows the average values o the initial load actors calculated using equally weighted data by

year o commissioning or onshore wind arms in the UK and Denmark together with fitted

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 15

trend lines8 here has been a clear decline in the initial load actors in the UK which has not

occurred in Denmark his pattern is confirmed by more detailed statistical results shown in

the Appendix which take account o location and size o installation

18 he decline in initial perormance is particularly obvious afer 2005 when the rate o commis-

sioning new capacity increased sharply rom less than 500 MW in the 5 years 2000ndash04 to morethan 500 MW per year rom 2005 onwards In contrast the peak years or commissioning new

onshore wind arms in Denmark were 2000ndash02 when 920 MW was installed he rate o new

building (or replacement) picked up again in 2008 but it is still much lower than at the turn o

century Not only is the perormance o the onshore wind plants in the UK commissioned in

recent years significantly worse than that or wind arms commissioned beore 2005 but it is

possible that this is a direct consequence o an overly rapid expansion o capacity In essence

the evidence suggests that the industry does not have the capability to identiy develop and

operate new onshore wind arms at the rate envisaged by UK government targets while main-

taining a satisactory average level o perormance

Implications for future policy and performance

19 he pattern o perormance degradation with age and over time identified in this analysis has

implications or the assessment o new investment in wind arms and or the design o the

proposed Electricity Market Reorm (EMR) here are two related issues that are affected by

the results

a) What are the implications or the design o policies intended to promote the adoption

and expansion o wind generation in the UK his includes the structure o current

subsidies as well as the specification o the contracts that are proposed under the EMR A

related question concerns the expected lie o new wind arms

b) How might the average perormance o wind generation develop over the next decade

his is absolutely critical to any assessment o the amount o wind capacity that will be

required to meet the UK governmentrsquos targets or the share o electricity ndash and more

generally energy ndash rom renewable sources he government has assumed that the

average load actors or both onshore and offshore wind arms will either remain stable

or increase in uture he prospects will be very different i these assumptions are not

borne out

20 he first question can be addressed by considering the discounted sum o cumulative net

output rom a new wind arm on the assumption that its perormance matches the averages

identified in the analysis his takes account o both perormance degradation over time plus

the annual cost o maintaining the plant which may be expressed as a deduction rom gross

production so as to calculate the contribution o expected output at age s to recovering the

original capital cost o building the plant hese figures have to be discounted back to the date

o commissioning in order to assess the net present value o the initial investment in the plant

he curves or constant and observed perormance using a discount rate o 9 (a typical cost o

capital or wind generation) are shown in Figure 5 he black line is based on the ull set o age

8 he initial load actors are calculated as the normalised load actor or UKDanish wind arms at age 0 multiplied by

where is the unit fixed effect or installation i

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1852

16 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

effects while the red dashed line is based on the smoothed quadratic perormance curve and

the blue dashed line shows what the pattern would be i operating perormance was not affected

by age he calculation is extended out to age 25 since DECCrsquos standard comparisons assume

that wind turbines have a lie o 25 years

Figure 5 Impact of performance degradation on discounted cumulative output

0

10

20

30

40

50

60

70

80

90

100

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

o

f d i s c o u n t e d c u m u l a t i v e o u t p u t

Age of plant (years)

Constant Quadratic Full age effects

Source Authorrsquos estimates

21 he key points in the figure are that 80 o the discounted cumulative output o a new wind

plant is likely to be produced in the first 10 years o its lie and 90 in the first 14 years his

is consistent with the structure o Danish subsidies or onshore wind arms which extend over

a typical period o 125 years Since sites or wind arms are scarce and involve the payment o

significant rents that may be linked to output it is very unlikely that any new installation o

wind turbines will have an expected lie o more than 15 years Instead as has happened in the

past wind operators will have a strong incentive to decommission plants afer no more than 15

years and replace the turbines with newer equipment

22 As a consequence any economic assessment o wind generation should not be based on an

expected lie which is longer than 15 years In recent work reported in evidence to the House

o Commons Select Committee on Energy and Climate Change I assumed that wind plants

would have a residual value equal to 20 o their initial cost in real terms at the end o 15 years

he analysis in this paper suggests that this is too avourable an assumption Given the costs o

decommissioning old turbines the residual value is likely to be well below 10 o their initial

cost and the decision point may be at 10 rather than 15 years

23 he corollary o this observation is that it makes little sense to offer long term contracts or 20

years or more that guarantee prices or eed-in tariffs (Fis) to wind operators A contract length

o 10 to 12 years would be sufficient to remove most o the market risk associated with invest-

ment in wind generation In this respect the subsidy arrangements implemented in Denmark

are better designed

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 17

24 At the same time the offer o subsidies andor guaranteed prices may have serious adverse

consequences or the efficiency o wind generation Returning to the unit fixed effects measur-

ing the perormance o wind plants commissioned afer 2005 only 28 out o 159 units have an

operating perormance that exceeds the average or the period hose 28 units account or 360

MW o capacity whereas the remaining 131 units account or 2810 MW o capacity Not onlyare recent plants less efficient than the average or the whole period but the plants which are

below-average in efficiency are typically larger than the more efficient ones and account or a

disproportionate share o the recent additions to generating capacity

25 In mid-2012 there was approximately 4500 MW o onshore wind capacity operating in the

UK By 2020 it is expected that there will be at least 10000 MW (based on proposed wind

arms with planning consent) and perhaps as much 15000 MW o wind capacity in operation

he government is relying upon a substantial contribution rom onshore wind arms towards

meeting its targets or renewable energy in 2020 o provide context the governmentrsquos targets

or renewable energy in 2020 speciy a total o 234 Wh o energy rom all renewable sources

o which 90ndash140 Wh is expected to come rom wind and biomass electricity generation9

he projections or biomass electricity are extremely vague and have to be reconciled with

an assumed increase o 3ndash4 times in the amount o biomass used separately or heat So in

concrete terms it is reasonable to assume that wind generation will account or at least 90 Wh

with 24ndash32 Wh rom onshore generation and the remainder rom offshore

26 he average load actor or onshore wind over the 10 year period rom April 2002 to March

2012 was 256 which is influenced by the large amounts o new capacity added rom 2005

onwards I this load actor were achieved in 2020 the total amount o onshore wind capacityrequired to meet the onshore generation target would be 107 to 143 GW he average load

actor or offshore wind installations (on an unchanged configuration basis) or the five year

period 2007ndash11 was 32010 I this load actor were achieved in 2020 the total amount o

offshore wind capacity required to meet the remainder o the 90 Wh target would be 206

to 235 GW though these estimates would all to 180 to 205 GW i the average load actor

matched the historic average value o 367 or Danish offshore wind installations

27 he question that has to be addressed is whether these load actors are consistent with (a) the

investment programmes that would be required to achieve the total amounts o generation

capacity implied by the governmentrsquos targets and (b) the age profiles o the perormance o

wind arms discussed in this paper he short answer is that this is simply not possible i the

initial load actor or new wind arms continues to decline at the rate observed the past decade

So even as a starting point it is necessary to assume that this underlying deterioration in wind

arm perormance ceases his would be a major change in itsel and it is not obvious what

actors might lead to such a result

28 he results illustrated below are based on a vintage model o perormance that takes account

o the actual expansion o capacity up to the end o 2011 plus whatever investment in new or

replacement turbines is required to achieve the target levels o wind capacity in 2020 o high-light the ull range o potential outcomes these targets are either 10 GW or 15 GW o onshore

9 Department o Energy and Climate Change ndash UK Renewable Energy Road Map July 2011 Figure 2

10 he equivalent average load actor or onshore wind arms was 264

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2052

18 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

wind and either 18 or 24 GW o offshore wind he levels o investment required to meet these

targets depend upon the expected lie o wind turbines For onshore turbines an expected lie

o 10 years is consistent with re-powering decisions over the last decade while an expected lie

o 15 years seems to be the maximum consistent with the age-perormance profile For offshore

turbines the analysis uses expected lives o 15 and 20 years

Figure 6A Projected load factors for UK onshore wind (total capacity of 10 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

Figure 6B Projected load factors for UK onshore wind (total capacity of 15 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

29 Figure 6A shows projected values o the average normalised load actors or UK onshore wind

arms in 2015 and 2020 using a target capacity in 2020 equal to 10 GW plus the age-peror-

mance profiles or the UK and Denmark (DK) Since vintage effects are important in this

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 19

calculation the profile o new or replacement investment affects the results In this case the

level o gross investment is assumed to be constant rom 2013 onwards at either 930 or 1220

MW per year ndash the range is defined by the assumed lie o turbines Achieving a target o 15

GW in 2020 (Figure 6B) would require a level o investment rom 2013 onwards o either 1540

or 1930 MW per year Such figures are 3ndash4 times the average level o investment over theperiod 2005ndash11 so the lower target may provide a better indication o what will happen

30 All o the estimates are calibrated to match the observed average load actor in 2007ndash11 so that

they should be interpreted as relative rather than absolute values Using the age-perormance

profile or UK onshore wind installations the average normalised load actor is likely to all by

about 19 rom 2007ndash11 to 2015 and by 23ndash28 to 2020 I the age-perormance curves or

Danish onshore installations were to apply the reduction in the average normalised load actor

would be considerably less but there would still be a all o up to 5 by 2020

31 In the medium term the all is greater i turbines operate or 15 years because the average age

o all turbines will be higher and thus the degradation in perormance due to age will have a

larger effect By 2020 this effect reduces the average normalised load actor rom 203 or a 10

year lie to 195 or a 15 year lie here is a complex economic and commercial trade-off here

o meet a target o 10 GW in 2020 it is necessary to increase the level o investment in onshore

capacity by 30 rom 2013 Almost all o this will take the orm o re-powering existing wind

arms11 he costs are airly high because this will usually involve new development consents

decommissioning the original turbines and replacing them with a smaller number o larger and

more powerul turbines

32 Estimates o re-powering costs are generally treated as being commercially sensitive but broadproject costs cited in specialist magazines suggest that the unit costs are likely to be at least

pound1000 per kW or more than pound300 million per year in aggregate On the other hand the increase

in the amount o energy generated would be about 12 Wh or a 10 year turbine lie rather

than a 15 year turbine lie though this does not allow or the loss o production during re-pow-

ering Under the best circumstances the payback period may be quite short but the best option

rom a broader economic and social perspective is not obvious

33 Even under the best scenario 10 GW o onshore wind capacity will only generate 178 Wh o

electricity in 2020 well below the lower end o the target range or onshore wind generation

specified in DECCrsquos road map In act allowing or interactions between the age-perormance

profile and investment the amount o onshore wind capacity required in 2020 to generate 24

Wh would be a minimum o 134 GW At the top end o DECCrsquos target range the capacity

required in 2020 to generate 32 Wh would be a minimum o 178 GW Since these estimates

are based on a 10 year turbine lie the level o investment required rom 2013 onwards would

be 1650 MW per year or the bottom o the range and 2350 MW or the top end o the range

In the last 8 years total investment in onshore wind has only ever exceeded 600 MW in one

11 Re-powering is usually defined as the ull replacement o existing turbines whereas retrofitting covers the replace-

ment o various mechanical or electrical components In the data analysis re-powering involves the creation o a

new unit whereas retrofitting affects the perormance o a continuing unit Hence the potential benefits o retrofit-

ting are taken into account in constructing the age-perormance curves

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2252

20 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

year (883 MW in 2008) so it seems rather unlikely that even the lower end o DECCrsquos range or

onshore wind generation will actually be realised

34 O course this assessment would be different i UK onshore wind operators were able to match

the Danish age-perormance curve here are many actors which differentiate the UK and

Danish wind industries he average scale o wind arms in the UK is larger they are moreremote and there appears to be significantly greater variation between operating and potential

new sites in achieved perormance It would thereore require a very optimistic observer to

conclude that the difference in age-perormance curves between the two countries will disap-

pear in the short or medium term

Figure 7 Projected load factors for UK offshore wind in 2020

20

22

24

26

28

30

32

34

18 GW ndash life 15 yrs 18 GW ndash life 20 yrs 24 GW ndash life 15 yrs 24 GW ndash life 20 yrs

L o a d f a c t o r ( )

2007ndash11 2015 2020

Note Based on Danish age-perormance curves Source Authorrsquos estimates

35 Figure 7 shows the results rom carrying out a similar exercise or offshore wind arms in the

UK he only evidence on the age-perormance curves or offshore installations is based upon

the Danish experience so these have been used to construct the projections Further the profile

o investment assumes a progressive increase in additions to capacity rom 2013 to 2016 and

then a stable level o investment rom 2017 to 2020 he DECC road map involves a very large

commitment to offshore wind investment ndash in the range 25-35 GW o new capacity annually

rom 2016 onwards As a consequence the average age o offshore capacity in 2020 will be quite

low and the ull effects o the age-perormance curves will only be elt during the decade ollow-

ing 2020 Even so the average normalised load actor in 2020 will be about 15 lower than or

2007ndash11 while by 2025 it will be about 30 below the base level

36 On these projections 232 GW o offshore capacity is the minimum necessary to generate 58

Wh o electricity rom offshore wind in 2020 his requirement will increase to 298 GW by

2025 simply to offset the effect o the ageing o offshore generation capacity on total output

37 Combining the estimates or onshore and offshore wind generation the total amount o wind

capacity required in 2020 to produce 90 Wh o electricity rom wind will be in the range 39

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 21

to 41 GW once allowance is made or the observed age-perormance curves or both onshore

and offshore installations hese estimates are roughly a third higher than official predictions o

28ndash31 GW o wind capacity in 2020

38 urning the figures round gives a slightly different picture o the reality o the governmentrsquos

strategy on renewable electricity generation A total o 30 GW o wind capacity in 2020 (12GW onshore 18 GW offshore) will generate about 64 Wh o electricity in a normal year

Since electricity generation rom wind and biomass are expected to account or 50 o renew-

able energy in 2020 the consequence is that biomass is expected to account or 53 Wh In

2011 landfill gas and sewage sludge digestion accounted or 44 o electricity rom bioenergy

(DUKES ables 64) here are limited opportunities to expand production rom these sources

so a continuation o historic rates o growth would mean that renewable generation rom other

sources o bioenergy would have to increase rom 72 Wh in 2011 to 46 Wh in 2020 here

are severe constraints on increasing the amounts o electricity generated rom animal biomass

biodegradable municipal solid waste and anaerobic digestion so it would be optimistic to

assume that these will contribute more than 6-8 Wh in 2020 (up rom 26 Wh in 2011)

39 In practice i the projections or renewable energy are to be believed they rest upon an assump-

tion that there will be a large increase in the amount o electricity that is generated rom plant

biomass ndash either on its own or through co-firing with coal his increase will involve increas-

ing the amount o electricity generated in these ways rom 46 Wh in 2011 to 38-40 Wh in

2020 Since the amounts o straw and UK-grown timber available or this purpose are strictly

limited this will come down to the amount o wood chips that can be imported and the cost o

doing this40 he historic average load actor or biomass plants is just over 60 so about 7500 MW o

biomass capacity will be required to supply 40 Wh o electricity he weight o wood chips per

MWh o electricity depends upon moisture content storage boiler design and other actors but

an indicative range is 085 to 115 tonnes per MWh o wood chips with a moisture content o

30 According to DECC and Forestry Commission estimates the UK imported 265 million

tonnes o steam coal and 26 million tonnes o other wood products (including wood chips)

in 2011 It seems that the net effect o the governmentrsquos strategy will be to replace 15 million

tonnes o coal imported in 2011 by 32 to 44 million tonnes o imported wood chips his is a

somewhat strange way o increasing energy security

Conclusion

41 Wind power is a highly capital-intensive technology or generating electricity Its merits rely

entirely upon a substitution o capital or uel inputs he same is true or hydro or tidal or wave

power In comparison with hydro power wind is a low quality resource because o its variability

and because it cannot be stored So the economic case or wind power must rest on obtaining

the most out o the wind that is available

42 While the decline in the achieved perormance o onshore wind turbines in Denmark is

much less than that or the UK or offshore nonetheless the decline in expected output under

standardised wind conditions over 10 years is 10 unweighted and 13 capacity weighted

hese declines accelerate afer age 10 so that the reductions in perormance are 17 and 20

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2452

22 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

respectively afer 15 years For UK onshore wind arms the reduction in perormance due to

age is much worse at 27 unweighted and 69 capacity weighted by age 10

43 Evidence on the perormance o Danish offshore installations is both restricted and so poor

that there may be concern that the results are affected by a small number o outliers Still the

sample contains a reasonable number o sites with at least 5 years o operating experience andthe decline in perormance by age 5 is 38 unweighted and 26 capacity weighted

44 In addition to these results there is strong evidence that the average normalised load actor or

new onshore wind installations in the UK has allen significantly over the period rom 2000 to

2011 his is consistent with a pattern in which the most avourable sites are developed first

Equally it could mean that wind developers have been unable to keep up with the rate o new

investment while maintaining the quality o development and operations For example the site

design or selection o turbine characteristics may make less effective use o the available wind

resources or the sites available than was the case in the past

45 Whatever the reasons the deterioration in initial perormance means that the expected returns

rom the expansion in wind capacity both or investors and in terms o the reduction in CO983090

emissions have been alling without a concomitant decrease in the private and social costs

that are borne by customers and the general public Clearly this is unsatisactory at best and it

suggests that the benefits claimed or current policies cannot be taken at ace value

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 23

Appendix Data and Methods

A Data for the United Kingdom

he raw data used or this study was extracted by the Renewable Energy Foundation rom the

Renewables and CHP Register database compiled by Ogem12 his inormation is used in the

administration o the market in Renewable Obligation Certificates (ROCs) he Renewables

Obligation is the primary mechanism by which large scale generators o renewable electric-

ity receive subsidies so the operators o wind arms have a very strong incentive to submit

complete and up-to-date inormation to the database he data extracted covered all onshore

wind generators with at least 05 MW o generating capacity which are eligible or ROCs or the

10 year period April 2002 to March 2012 he key variables are the monthly output o electric-

ity (used or the purpose o allocating ROCs) and nominal generating capacity here are 282

reporting units in the final dataset A reporting unit may correspond to an entire wind arm orto separate phases o development Complications arise when an existing wind arm is re-pow-

ered ie when old wind turbines are replaced by newer and usually more powerul turbines

he database reports this as a change in the generating capacity o the recording unit but in

this analysis repowering is treated as the termination o the old recording unit and the creation

o a new recording unit

A considerable amount o data checking and cleaning was required beore the data could

be analysed he first step was to add the age o each reporting unit his was calculated by

reerence to the month in which the turbines were commissioned For wind arms in operation

in or beore June 2002 the commissioning date was obtained rom various online sources ndash

notably the thewindpowernet database o UK wind arms13 which was cross-checked against

the inormation provided on the websites o wind operators and other sources For recording

units whose first data relates to dates afer June 2002 the commissioning date was initially

assumed to be the month or which data is first reported Cross-checks were carried out or

all recording units which commenced reporting in the period July 2002 to December 2003 In

more than 80 o cases the externally-reported commissioning date is within one month o the

first month or which data is reported he exceptions ndash Bu Farm and Mablethorpe ndash appear to

be due to delays by small operators in reporting data to Ogem

In a number o cases it is clear rom the data that plants were not in ull operation at the

time when data was first reported to the database here are two indicators o partial or delayed

commissioning

bull For an initial series o months the calculated load actor is very low ndash usually below 5 ndash

and then increases abruptly In such cases the data or the early months is dropped and the

commissioning date is treated as the month immediately preceding the first month with a

lsquonormalrsquo load actor14

12 httpswwwrenewablesandchpogemgovuk

13 See httpwwwthewindpowernetwindarms_list_enphp

14 he month in which a plant is reported as having been commissioned is not included in the analysis

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2652

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 25

the decline in the plantrsquos load actor as it ages As an example consider a plant whose reported

capacity was increased rom 99 MW to 10 MW at age 3 I this was simply a change in report-

ing practice this would reduce the estimated load actor rom age 3 onwards By increasing

the initial capacity rom 99 MW to 10 MW ndash or by reducing the capacity rom 10 MW to 99

MW rom age 3 onwards ndash this bias is removed It is not important which adjustment is madebecause the difference between the alternative assumptions is captured by the unit fixed effect

he numbers o separate sites and o observations per site mean that it is not possible to

provide a simple visual representation o the data Figure 8 provides a summary by illustrating

the distributions o load actors by the age o the wind arm pooled across all years he ranges

o the load actors or each age are wide while the medians show no clear trend by age hus a

naiumlve analysis might conclude that there is nothing to investigate However as will be explained

below the distributions mask crucial differences between the perormance o wind arms as

they age because they do not control or the differences in location and wind availability over

time

B Data for Denmark

he data or Danish wind arms used in this study comes rom a database compiled by the

Danish Energy Agency covering the characteristics and perormance o all wind turbines rom

2002 up to the end o August 201216 he basic recording unit or the register is the individual

wind turbine but where a number o turbines are connected to a single meter an identical

average output per turbine is recorded or all turbines in the group he presence o what are in

effect repeated observations increases the noise in the data In addition the model that is used

or the analysis assumes that site operator and turbine characteristics are common or all o theunits that make up a single wind arm o address this point separate wind arms have been

identified by grouping all turbines which have identical values or turbine manuacturer and

type local authority date o commissioning and date o decommissioning (where applicable)

he register contains inormation on 7569 turbines commissioned rom 1977 onwards his

includes a large number o small turbines which would not all within the scope o the ROC

scheme in the UK Hence or comparability with the UK data all turbines commissioned

prior to 1992 were dropped as well as all wind arms with a generating capacity o less than 1

MW he resulting sample covered 823 onshore wind arms and 30 offshore wind arms with

monthly load actors or various periods rom January 2002 to August 2012 Given the cut-off

or wind arms commissioned prior to 1992 the maximum age or an onshore wind arm was 20

years and the maximum age or an offshore wind arm was 17 years

he total capacity o the onshore wind arms recorded in the Danish database was 2757 MW

he average size o an onshore wind arm in Denmark is small ndash only 33 MW as compared

with 161 MW or the UK ndash because 84 o the sample consists o installations with between

1 and 4 turbines he largest onshore installation has 39 turbines with a capacity o 234 MW

while the onshore installation with the greatest capacity has 19 turbines with a capacity o 584

MW Overall the sample o Danish onshore wind arms is older and smaller than the sample

o UK onshore wind arms reflecting the much longer history o wind generation in Denmark

16 httpwwwensdkEN-USINFOFACSANDFIGURESENERGY_SAISICS_AND_INDICAORS

OVERVIEWOFHEENERGYSECORREGISEROFWINDURBINESSiderForsideaspx

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2852

26 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

he larger scale o UK wind arms may account or some o the differences in perormance

curves or the two countries

he sample o offshore wind arms in Denmark is quite small with 30 installations in

total Still it is useul to analyse the perormance o these installations since there is very little

comparable data or the UK he sample covers 394 turbines with an average o just over 13turbines per installation and an average capacity o 288 MW he largest installation which

was commissioned in 2003 has 72 turbines with a total capacity o 1656 MW Again offshore

installations are smaller and older in Denmark than is the current pattern in the UK he first

two wind arms in the sample were commissioned in the years 1995 and 2000 but the sample

increases significantly rom 2002 onwards o avoid reliance upon a sample o 1 or 2 wind

arms to determine specific coefficients or ages gt 10 years all observations with an age greater

than 10 were omitted rom the analysis17

A somewhat different problem with the offshore wind arm data arose because the coeffi-

cients estimated using capacity weights (see below) were heavily influenced by the largest windarm his has had an extremely erratic profile o perormance over time with a very low average

load actor he treatment o this wind arm is discussed in more detail in Section F below

C Specication and estimation methods

As Figure 8 illustrates it is necessary to go beyond simple summary statistics in order to assess

whether there is any systematic relationship between age and the average perormance o wind

arms he perormance o wind arms varies over both time and space some months and years

have more or less wind than the long term average while specific characteristics o wind arms

ndash including location type o turbine and operating regime ndash will influence the perormance oeach plant under identical wind conditions

he combination o (a) site-specific characteristics which are constant over time and (b)

period-specific characteristics which are constant across wind arms can be represented by

what statisticians call an error components model with fixed effects or each wind arm and

each time period he dependent variable in the model consists o a sequence o load actors ndash

denoted by or unit i in month t ndash or each wind arm (the panel unit) over time he load

actors are calculated by dividing the total output in MWh by (24 x number o days in month x

reported generating capacity in MW) and are multiplied by 100 to convert to percentages

he additive version o the error components model may be written as

(1)

in which denotes the age o plant i in period t and is some unction that is either a

representation o or an approximation to the normalised age-perormance relationship he

multiplicative version is similar except that is replaced by where ln() denotes the

natural logarithm he terms are wind arm or unit fixed effects which capture actors such

as location type o turbine operating regimes etc he terms are period fixed effects which

capture the influence o wind conditions over the UK or Denmark as a whole as well as seasonal

maintenance and demand Without loss o generality it may be assumed that the fixed effects are

17 For the avoidance o doubt this means that no attempt was made to estimate the effect o age on the perormance o

offshore wind arms beyond age 10 while the data or the oldest wind arms was included in the analysis up to and

including age 10

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 27

normalised so that ie all o the error components have a zero mean while the

mean o the random error is zero by definition he terms are random errors which capture

random variations in wind conditions that are not specific to the site turbine breakdowns and

other actors that are uncorrelated with either the site or the time period It is assumed that the

mean o the random error is zero while additional assumptions depend upon the method o

estimation that is adopted

he model specified in equation (1) can be estimated in a variety o ways18 A crucial issue is

whether the are or may be significantly correlated with the age terms or any other independent

regressors I it is assumed that there is no such correlation the model is usually reerred to as

a random effects model In that case the equation can be estimated by a variant o least squares

(LS) generalised least squares (GLS) generalised methods o moments (GMM) or maximum

likelihood (ML) ndash all o which should be yield parameter estimates that will converge in prob-

ability to the true parameter values as the sample size increases An alternative but less restric-

tive specification is known as the fixed effects model in which no assumption is made about

the correlation between the unit fixed effects and other regressors he limitation o the fixed

effects model is that it cannot be used to estimate parameters or regressors that vary across

panel units but are constant over time or a panel unit (time-invariant regressors) ndash eg the

year in which a wind arm was commissioned its location or its rated generating capacity he

influence o such variables is captured by the unit fixed effects

In order to avoid the risk o obtaining biased estimates o the age effects the results reported

in this note are based upon estimates using the fixed effects model he influence o time-invar-

iant regressors is examined in a second stage by regressing the estimates o the unit fixed effectson these variables he simplest method o estimating the fixed effects model is known as the

lsquowithinrsquo estimator in which the mean value or each unit is subtracted rom all observations or

that unit For example i is a general polynomial o order M in A equation (1) becomes

(2)

so that the model can be estimated by least square with the inclusion o dummy variables or

each period t he variance-covariance matrix o the coefficients have been estimated by using

(a) a robust sandwich estimator adjusted or clustering by panel unit and (b) a bootstrap esti-

mator with 400 repetitions he estimates o the standard errors will be consistent even i the

random error component is serially correlated within panel units andor has different variances

across panel units19 Since the number o panel units and the average number o time periods

per panel units are both large (other than or Danish offshore wind arms) the least square esti-

mates o the parameters will be unbiased under these assumptions

he bootstrap method is expensive to carry out but it provides a useul cross-check on

the standard errors generated by other methods Unortunately the bootstrap method o

18 See Chapter 21 o AC Cameron amp PK rivedi ndash Microeconometrics Methods and Applications (Cambridge

Cambridge University Press 2005) or a standard textbook treatment o estimation methods or panel data models

19 he estimation was carried out using the xtreg procedure in Stata Version 12 he methods o constructing the

sandwich and bootstrap estimates o the variance-covariance matrix are described in Chapters 8 amp 13 o AC

Cameron amp PK rivedi ndash Microeconometrics using Stata (Revised Edition College Station exas Stata Press 2010)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 3: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 352

The Performance of Wind Farms

in the United Kingdom and Denmark

Gordon Hughes

2012

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 452

copy Gordon Hughes 2012

Cover background image by Simon Gray (httpwwwstar-oneorguk) via stockxchng

Published by the Renewable Energy Foundation

57ndash58 Russell Square

London WC1B 4HSwwwreorguk

he Renewable Energy Foundation is a registered charity in England and Wales (1107360)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 552

Contents

List o Figures and ables 4

About the author 5

About the Renewable Energy Foundation 5

Authorrsquos Note 6

Note on the Data 6

Executive Summary 7

Te Perormance o Wind Farms in the United Kingdom and Denmark 9

Introduction 9

Age-perormance curves 10

Differences between wind arms 12Implications or uture policy and perormance 15

Conclusion 21

Appendix Data and Methods 23

A Data or the United Kingdom 23

B Data or Denmark 25

C Specification and estimation methods 26

D Period fixed effects vs normalisation by wind speeds 28

E Estimation results or the UK 30F Estimation results or Denmark 36

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 652

List of Figures and TablesList of Figures and Tables

Figure 1 Perormance degradation due to age using equal weights 11

Figure 2 Perormance degradation due to age using capacity weights 12

Figure 3 Range o unit fixed effects by category 13

Figure 4 Initial load actors by year o commissioning or onshore wind 14

Figure 5 Impact o perormance degradation on discounted cumulative output 16

Figure 6A Projected load actors or UK onshore wind (total capacity o 10 GW in 2020) 18

Figure 6B Projected load actors or UK onshore wind (total capacity o 15 GW in 2020) 18

Figure 7 Projected load actors or UK offshore wind in 2020 20

Figure 8 Box plots o load actors by age or UK onshore wind arms 24

Figure 9A Additive age-perormance curves or UK onshore wind arms 32

Figure 9B Multiplicative age-perormance curves or UK onshore wind arms 32

Figure 10 UK onshore age-perormance curves using equal and capacity weights 33

Figure 11A Residuals by age or perormance curves using equal weights 34

Figure 11B Residuals by age or perormance curves using capacity weights 34

Figure 12 Unit fixed effects by year o commissioning or England and Scotland 35

Figure 13A Age-perormance curves or Danish onshore wind arms 37

Figure 13B Age-perormance curves or Danish offshore wind arms 37

Figure 14 Danish age-perormance curves using equal and capacity weights 38

able 1 Average load actors by year and country () 40

able 2 Estimation results or UK onshore wind arms 41

able 3 Equations or trends in unit fixed effects 43

able 4 Estimation results or Danish wind arms 44

able 5 Perormance by year o commissioning or Danish wind arms 46

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 752

About the author

Dr Gordon Hughes is a Proessor o Economics at the University o Edinburgh where he teaches courses in

the Economics o Natural Resources and Public Economics He was a senior adviser on energy and environ-

mental policy at the World Bank until 2001 He has advised governments on the design and implementationo environmental policies and was responsible or some o the World Bankrsquos most important environmental

guidelines

About the Renewable Energy Foundation

he Renewable Energy Foundation is a registered charity promoting sustainable development or the benefit

o the public by means o energy conservation and the use o renewable energy

REF is supported by private donation and has no political affiliation or corporate membership In pursuit o

its principal goals REF highlights the need or an overall energy policy that is balanced ecologically sensi-

tive and effective

We aim to raise public awareness o the issues and encourage inormed debate regarding a structured energy

policy that is both ecologically sensitive and practical he issues o climate change and security o energy

supply are complex and closely intertwined REF contributes to the debate surrounding these issues by

commissioning reports to provide an independent and authoritative source o inormation

For urther inormation see wwwreorguk

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 852

Authorrsquos Note

Note he views expressed in this paper are strictly personal and do not reflect the position o any organ-

isation with which I am associated I am grateul to John Constable Lee Moroney and a reeree or their

comments on earlier drafs o the paper he results o this study have been presented at seminars andlectures to various groups in Edinburgh and Glasgow I have benefited rom the questions and comments

rom participants in those events

Gordon Hughes

School o Economics

University o Edinburgh

gahughesedacuk

Note on the Data

For the convenience o interested researchers both the raw and cleaned data employed in this analysis will

published alongside the electronic version o the study on the REF website wwwreorguk

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 7

Executive Summary

1 Onshore wind turbines represent a relatively mature technology which ought to have achieved

a satisactory level o reliability in operation as plants age Unortunately detailed analysis o

the relationship between age and perormance gives a rather different picture or both theUnited Kingdom and Denmark with a significant decline in the average load actor o onshore

wind arms adjusted or wind availability as they get older An even more dramatic decline is

observed or offshore wind arms in Denmark but this may be a reflection o the immaturity

o the technology

2 he study has used data on the monthly output o wind arms in the UK and Denmark reported

under regulatory arrangements and schemes or subsidising renewable energy Normalised

age-perormance curves have been estimated using standard statistical techniques which allow

or differences between sites and over time in wind resources and other actors

3 he normalised load actor or UK onshore wind arms declines rom a peak o about 24 at

age 1 to 15 at age 10 and 11 at age 15 he decline in the normalised load actor or Danish

onshore wind arms is slower but still significant with a all rom a peak o 22 to 18 at age 15

On the other hand or offshore wind arms in Denmark the normalised load actor alls rom

39 at age 0 to 15 at age 10 he reasons or the observed declines in normalised load actors

cannot be ully assessed using the data available but outages due to mechanical breakdowns

appear to be a contributory actor

4 Analysis o site-specific perormance reveals that the average normalised load actor o new UK

onshore wind arms at age 1 (the peak year o operation) declined significantly rom 2000 to2011 In addition larger wind arms have systematically worse perormance than smaller wind

arms Adjusted or age and wind availability the overall perormance o wind arms in the UK

has deteriorated markedly since the beginning o the century

5 hese findings have important implications or policy towards wind generation in the UK First

they suggest that the subsidy regime is extremely generous i investment in new wind arms

is profitable despite the decline in perormance due to age and over time Second meeting

the UK Governmentrsquos targets or wind generation will require a much higher level o wind

capacity ndash and thus capital investment ndash than current projections imply hird the structure

o contracts offered to wind generators under the proposed reorm o the electricity market

should be modified since ew wind arms will operate or more than 12ndash15 years

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 9

The Performance of Wind Farms in the United Kingdom

and Denmark

Introduction

1 Any assessment o the costs o wind power must rely heavily upon assumptions about theaverage load actor that will be achieved by new wind installations over their lietime It is

standard practice to calculate average load actors by year and country or onshore and offshore

installations as shown in able 1 (page 40)1 However such estimates do not provide a

reliable statistical basis or assessing the uture perormance o wind arms in aggregate Part

o the reason is that the amount o wind in any month or year is influenced by long term mete-

orological cycles that have periods o many years notably the North Atlantic Oscillation In

addition average load actors do not allow or changes in the composition o wind installations

by location age size and other actors

2 As the number o wind arms operating in developed countries has grown engineers have an

opportunity to analyse operating experience over extended periods o time With any (relatively)

novel technology the incidence o temporary or permanent breakdowns may be expected to all

over time so it is difficult to disentangle the effects o age on operating perormance rom the

technological immaturity o turbines installed 10 or 20 years ago Nonetheless the technology

or onshore wind turbines has been reasonably mature since the early 2000s his is documented

by data produced by the US Department o Energy which shows that the decline in capital costs

that is characteristic o immature technologies slowed afer 2000 and was reversed afer 20042

3 With at least 10 years o operating data since 2000 it should be possible to examine how thetypical operating perormance o wind installations in the United Kingdom and Denmark

varies with the age o the turbines he estimated age-perormance curves should then be

incorporated in estimates o levelised costs which are ofen used to compare the costs o wind

and other orms o generation

4 he Appendix provides a detailed description o the data and methods employed or this study

In both the UK and Denmark wind operators have a large incentive to produce electricity

whenever there is sufficient wind available since marginal operating costs are small while the

operator receives a much higher price In the UK this price is the sum o the market price plus

the value o (a) the Renewable Obligation Certificates (ROCs) awarded or each MWh o elec-

tricity produced and (b) the associated exemption rom the Climate Change Levy In practice

the effective market value o ROCs and associated incentives means that an onshore wind

operator earns roughly double the average market price o electricity per MWh In Denmark

most onshore wind capacity receives a price premium equivalent to about pound28 per MWh on top

1 he load actor is calculated as the ratio o the amount o electricity actually produced by a turbine or wind arm

over a period o a month or a year divided by the amount o output that would have been produced had it operated

at ull nameplate capacity or the entire period his is expressed as a percentage so that reported load actors lie

between 0 and 100

2 R Wiser amp M Bolinger (2012) 2011 Wind Technologies Market Report Lawrence Berkeley National Laboratory

US Department o Energy It should be noted that the installed cost o wind plants has allen since 2010 but these

changes reflect the cyclical orces o demand and supply that are well established across the electricity sector and

which apply to ossil-uel as well as renewable generators

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1252

10 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

o the market price or about 125 years o operation at a typical load actor3 Offshore capacity

is eligible or a similar price premium but most o it is contracted via tenders that offer a guar-

anteed eed-in tariff he tender tariffs have varied rom pound56 per MWh to twice that figure

extending over 16-18 years o operation at a typical load actor

Age-performance curves

5 A standard indicator o the operating perormance o a wind turbine is its normalised load

actor defined as the total output over some period divided by the maximum potential output

adjusted or wind availability Normalisation or wind availability is not straightorward I

detailed data on wind speeds at specific locations is available then the potential output rom

individual turbines or wind arms can be calculated using standard power curves Since wind

speeds can vary substantially over small distances such calculations are both difficult to imple-

ment and prone to errors when applied to large numbers o wind generators using public

sources o data4

6 Many analyses work in the opposite direction For example the Danish Wind Index ndash the

longest running series o data on wind availability in Europe ndash is constructed rom the common

component o monthly variations in wind output rom a large sample o wind turbines5 his

study adopts a similar approach Wind availability is treated as an unobserved actor whose

contribution to monthly variations in wind output is estimated by a series o fixed period

(monthyear) effects in a statistical model o monthly output or all wind arms he reasons

or adopting this approach are discussed in Section D o the Appendix his explains why the

method must perorm better than the easible alternative o using a single index o average

wind speed or the UK or Denmark

7 he analysis relies upon standard techniques that are widely used in the biological and medical

sciences as well economics and engineering Full details are given in the Appendix he idea is

to treat the actual load actor (ie without adjustment or wind conditions) or a specific wind

generator in any time period as being determined by a set o components which reflect specific

site conditions and the age-related perormance o wind plants in general together with the

period fixed effects and an uncorrelated random error he onshore wind datasets or the UK

and Denmark used or the analysis are large with monthly observations on 282 installations in

the UK and 823 installations in Denmark with an age range rom 0 to 19 years he offshore

wind dataset or Denmark is rather smaller covering only 30 installations but it can be used to

obtain reasonable estimates o perormance up to age 10 he results discussed here are based

upon least squares estimation (see page 27) combined with standard errors that are robust to

various departures rom classical assumptions about the nature o the data generation process

8 he results o the statistical analysis demonstrate an unambiguous and statistically significant

decline in the operating perormance o wind arms as they grow older Figure 1 shows the

3 Conversions to GBP are based on an exchange rate o pound1 = DKK 918 (as at 19 October 2012)

4 All wind generators collect their own inormation on wind speeds in order to manage their plants but they do not

publish such inormation and will not have access to inormation collected by other operators

5 Details o the construction o the Danish Wind Index are given at wwwvindstatdk A paper by Boccard ndash N Boccard

(2009) lsquoCapacity actor o wind power realized values vs estimatesrsquo Energy Policy Vol 37 pp 2679-88 ndash includes a

discussion o the construction o similar indices o wind availability in Germany the UK Netherlands and Sweden

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 11

simplest variant o the normalised age-perormance curve or onshore UK wind installations

together with the equivalent curves or onshore and offshore Danish installations he curves

illustrated are calculated rom the multiplicative error components model described in the

Appendix with log(load actor) as the dependent variable and a quadratic in plant age to repre-

sent the variation in plant perormance with age9 he rate o decline in perormance is greatest or offshore installations in Denmark with a all

rom load actors o over 40 at ages 0 and 1 to less than 15 by at ages 9 and 10 Onshore

installations in the UK show a more rapid rate o decline ndash 09 percentage points per year over

the first 10 years o operation ndash than is the case or Denmark though the normalised Danish

perormance curve lies below the UK curve or the first our years For the UK the normalised

load actor alls to just over 15 at age 10 and to 11 at age 15 With such low load actors it

seems likely that many wind arms will be re-powered ndash ie the turbines will be replaced ndash once

they reach the age o 10 or at most 15

Figure 1 Performance degradation due to age using equal weights

0

5

10

15

20

25

30

35

40

45

50

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

L o

a d f a c t o r ( )

Age (years)

UK onshore DK onshore DK offshore

Note Normalised load actors in Source Authorrsquos estimates

10 here are two plausible explanations or the observed decline in average load actors as wind

arms age he first is that the turbines become less efficient over time as a result o mechanical

wear and tear erosion o the turbine blades and related actors he second is that the turbines

experience more requent breakdowns and their operators take more time to bring them back

into service because they are less concerned about the perormance o older plants Both

reasons may be relevant in different circumstances and it is not possible to identiy a primary

explanation rom the data he requency o extended shutdowns does seem to increase with

age but this could be a reflection o the timing o planned maintenance operations rather than

breakdowns Whatever the cause the reduction in perormance with age is much greater than

would be expected or thermal generating plants

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1452

12 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

11 he age-perormance curves in Figure 1 are derived rom equations estimated by giving equal

weight to each wind arm irrespective o its generating capacity Figure 2 illustrates similar

curves but in this case the estimates are constructed by weighting each wind arm by its generat-

ing capacity reerred to as capacity-weighted his gives a better representation o perormance

degradation per MW o generating capacity he striking result is that the rate o decline inthe perormance o UK onshore wind installations is significantly aster when capacity weights

are used which implies that large wind arms in the UK experience a more rapid decline than

smaller ones he differences are smaller in Denmark but the capacity-weighted decline in

perormance is more rapid than the equal weights decline or Danish onshore installations

Figure 2 Performance degradation due to age using capacity weights

0

5

10

15

20

25

30

35

40

45

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

L o a d f a c t o r ( )

Age (years)

UK onshore DK onshore DK offshore

Note Normalised load actors in Source Authorrsquos estimates

Differences between wind farms

12 he specification o the statistical model means that the unit fixed effects identiy the specific

perormance characteristics o individual wind arms afer adjusting or wind availability and

age hese characteristics may reflect the site andor design o the installation as well as the

way in which it is operated A positive value or the unit fixed effect is equivalent to shifing

the perormance curve up by some percentage applied uniormly at each age while a negative

value shifs it down It ollows that the distribution o unit fixed effects across installations can

be used to compare the relative effectiveness o new installations by location date o commis-

sioning etc Care is required when making comparisons between or example UK and Danish

onshore wind installations because these have been normalised separately so the unit fixed

effects reer to different underlying averages6

6 Some care is required in interpreting the unit fixed effects By deault the average value o the unit fixed effects or

each group is zero but these averages are taken over all observations or each installation in the basic sample For the

purpose o the comparisons in Figure 3 the unit fixed effects have been adjusted so that the average values or each

group are zero when using one value or each installation

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 13

13 o give a sense o the numbers unit fixed effects o (a) 02 or (b) -02 mean that the base load

actor given the age o the plant is multiplied by (a) exp(02) = 122 or (b) exp(-02) = 082 Since

the base load actor or a plant o age 0 is 231 these fixed effects translate to load actors in

the first year o operation o (a) 282 and (b) 189 For a wind arm with 50 MW o gener-

ating capacity the difference amounts to an additional 40700 MWh o electricity output peryear Using the standard figures cited by RenewableUK this would be sufficient to supply 12300

homes or a year his range (040) is slightly greater than the interquartile range o unit fixed

effects or UK onshore wind arms (035) Again or comparison operators o thermal power

plants would regard any difference o even 10 up or down in efficiency relative to what is

expected as a striking and perhaps worrying indication o either good or bad perormance

Figure 3 Range of unit fixed effects by category

- 1

- 5

0

5

1

A d j f i x e d e f f e c t ndash q u a d r a t i c

UK onshore DK onshore DK offshore

Source Authorrsquos estimates

14 A notable eature o the distributions o the unit fixed effects that are summarised in the box

plots shown in Figure 3 is the range o variation between the best and worst perormers espe-

cially or UK onshore and Danish offshore wind arms Box plots are constructed so that the

top and bottom o the box correspond to the upper and lower quartiles while the bars at topand bottom correspond to the maximum and minimum values excluding outliers7 A plant at

the upper quartile o the distribution o UK onshore wind arms will generate an annual output

that is about 40 higher than a plant at the lower quartile he equivalent figure or Danish

onshore wind plants is an increase o 30 while or Danish offshore wind arms the difference

between the upper and lower quartiles is equivalent to 25 times the output o a lower quartile

plant By any standard such differences are important

7 he upper quartile is the point in a distribution such that the 25 o values exceed it while the lower quartile is the

point in a distribution such that 75 o values exceed it he line in the middle o the box marks the median the

point exceeded by 50 o values Outliers are defined as values that lie outside the range o ukey adjacent values

which are defined as the upperlower quartiles +- 15 x interquartile range

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1652

14 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

15 Outside the range between the upper and lower quartiles the top and bottom segments o

the distributions extend urther up and down or UK onshore wind arms than or those in

Denmark he rather different pattern or Danish offshore wind installations is partly a conse-

quence o the relatively small size o the sample but the evidence points to considerably greater

variability in the perormance o offshore wind installations than or onshore installations16 It is somewhat surprising to observe the magnitude o the differences in the perormance in

onshore wind arms in the UK Given the nature o the subsidy regimes and the high capital

cost o developing new installations it might be expected that operators have a strong incen-

tive to identiy the best locations and then choose equipment that will deliver the maximum

amount o electricity output at a high level o reliability For UK onshore operators it seems that

good or bad perormance is somewhat o a lottery However i the subsidies provided by ROCs

are sufficient to underwrite investment in inefficient plants ndash as appears to be the case ndash then

those subsidies are extremely generous or plants that operate close to the efficient rontier

As location is likely to be the main actor that determines the perormance o a specific plant

relative to all other plants the inerence must be that many wind plants have been developed on

sites with poor wind characteristics

Figure 4 Initial load factors by year of commissioning for onshore wind

10

15

20

25

30

35

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

I n i t i a l l o a d f a c t o r ( )

Year commissioned

UK onshore DK onshore Linear (UK onshore) Linear (DK onshore)

Source Authorrsquos estimates

17 here is another disturbing characteristic o UK wind arms which is revealed by examination

o the unit fixed effects and illustrated in Figure 4 It might be expected that the average peror-

mance o recently installed units would be higher than that o older units his should reflect

improvements in turbine design and reliability as well as in the identification o good sites or

wind generation However that is not the pattern or onshore wind arms in the UK he figure

shows the average values o the initial load actors calculated using equally weighted data by

year o commissioning or onshore wind arms in the UK and Denmark together with fitted

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 15

trend lines8 here has been a clear decline in the initial load actors in the UK which has not

occurred in Denmark his pattern is confirmed by more detailed statistical results shown in

the Appendix which take account o location and size o installation

18 he decline in initial perormance is particularly obvious afer 2005 when the rate o commis-

sioning new capacity increased sharply rom less than 500 MW in the 5 years 2000ndash04 to morethan 500 MW per year rom 2005 onwards In contrast the peak years or commissioning new

onshore wind arms in Denmark were 2000ndash02 when 920 MW was installed he rate o new

building (or replacement) picked up again in 2008 but it is still much lower than at the turn o

century Not only is the perormance o the onshore wind plants in the UK commissioned in

recent years significantly worse than that or wind arms commissioned beore 2005 but it is

possible that this is a direct consequence o an overly rapid expansion o capacity In essence

the evidence suggests that the industry does not have the capability to identiy develop and

operate new onshore wind arms at the rate envisaged by UK government targets while main-

taining a satisactory average level o perormance

Implications for future policy and performance

19 he pattern o perormance degradation with age and over time identified in this analysis has

implications or the assessment o new investment in wind arms and or the design o the

proposed Electricity Market Reorm (EMR) here are two related issues that are affected by

the results

a) What are the implications or the design o policies intended to promote the adoption

and expansion o wind generation in the UK his includes the structure o current

subsidies as well as the specification o the contracts that are proposed under the EMR A

related question concerns the expected lie o new wind arms

b) How might the average perormance o wind generation develop over the next decade

his is absolutely critical to any assessment o the amount o wind capacity that will be

required to meet the UK governmentrsquos targets or the share o electricity ndash and more

generally energy ndash rom renewable sources he government has assumed that the

average load actors or both onshore and offshore wind arms will either remain stable

or increase in uture he prospects will be very different i these assumptions are not

borne out

20 he first question can be addressed by considering the discounted sum o cumulative net

output rom a new wind arm on the assumption that its perormance matches the averages

identified in the analysis his takes account o both perormance degradation over time plus

the annual cost o maintaining the plant which may be expressed as a deduction rom gross

production so as to calculate the contribution o expected output at age s to recovering the

original capital cost o building the plant hese figures have to be discounted back to the date

o commissioning in order to assess the net present value o the initial investment in the plant

he curves or constant and observed perormance using a discount rate o 9 (a typical cost o

capital or wind generation) are shown in Figure 5 he black line is based on the ull set o age

8 he initial load actors are calculated as the normalised load actor or UKDanish wind arms at age 0 multiplied by

where is the unit fixed effect or installation i

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1852

16 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

effects while the red dashed line is based on the smoothed quadratic perormance curve and

the blue dashed line shows what the pattern would be i operating perormance was not affected

by age he calculation is extended out to age 25 since DECCrsquos standard comparisons assume

that wind turbines have a lie o 25 years

Figure 5 Impact of performance degradation on discounted cumulative output

0

10

20

30

40

50

60

70

80

90

100

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

o

f d i s c o u n t e d c u m u l a t i v e o u t p u t

Age of plant (years)

Constant Quadratic Full age effects

Source Authorrsquos estimates

21 he key points in the figure are that 80 o the discounted cumulative output o a new wind

plant is likely to be produced in the first 10 years o its lie and 90 in the first 14 years his

is consistent with the structure o Danish subsidies or onshore wind arms which extend over

a typical period o 125 years Since sites or wind arms are scarce and involve the payment o

significant rents that may be linked to output it is very unlikely that any new installation o

wind turbines will have an expected lie o more than 15 years Instead as has happened in the

past wind operators will have a strong incentive to decommission plants afer no more than 15

years and replace the turbines with newer equipment

22 As a consequence any economic assessment o wind generation should not be based on an

expected lie which is longer than 15 years In recent work reported in evidence to the House

o Commons Select Committee on Energy and Climate Change I assumed that wind plants

would have a residual value equal to 20 o their initial cost in real terms at the end o 15 years

he analysis in this paper suggests that this is too avourable an assumption Given the costs o

decommissioning old turbines the residual value is likely to be well below 10 o their initial

cost and the decision point may be at 10 rather than 15 years

23 he corollary o this observation is that it makes little sense to offer long term contracts or 20

years or more that guarantee prices or eed-in tariffs (Fis) to wind operators A contract length

o 10 to 12 years would be sufficient to remove most o the market risk associated with invest-

ment in wind generation In this respect the subsidy arrangements implemented in Denmark

are better designed

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 17

24 At the same time the offer o subsidies andor guaranteed prices may have serious adverse

consequences or the efficiency o wind generation Returning to the unit fixed effects measur-

ing the perormance o wind plants commissioned afer 2005 only 28 out o 159 units have an

operating perormance that exceeds the average or the period hose 28 units account or 360

MW o capacity whereas the remaining 131 units account or 2810 MW o capacity Not onlyare recent plants less efficient than the average or the whole period but the plants which are

below-average in efficiency are typically larger than the more efficient ones and account or a

disproportionate share o the recent additions to generating capacity

25 In mid-2012 there was approximately 4500 MW o onshore wind capacity operating in the

UK By 2020 it is expected that there will be at least 10000 MW (based on proposed wind

arms with planning consent) and perhaps as much 15000 MW o wind capacity in operation

he government is relying upon a substantial contribution rom onshore wind arms towards

meeting its targets or renewable energy in 2020 o provide context the governmentrsquos targets

or renewable energy in 2020 speciy a total o 234 Wh o energy rom all renewable sources

o which 90ndash140 Wh is expected to come rom wind and biomass electricity generation9

he projections or biomass electricity are extremely vague and have to be reconciled with

an assumed increase o 3ndash4 times in the amount o biomass used separately or heat So in

concrete terms it is reasonable to assume that wind generation will account or at least 90 Wh

with 24ndash32 Wh rom onshore generation and the remainder rom offshore

26 he average load actor or onshore wind over the 10 year period rom April 2002 to March

2012 was 256 which is influenced by the large amounts o new capacity added rom 2005

onwards I this load actor were achieved in 2020 the total amount o onshore wind capacityrequired to meet the onshore generation target would be 107 to 143 GW he average load

actor or offshore wind installations (on an unchanged configuration basis) or the five year

period 2007ndash11 was 32010 I this load actor were achieved in 2020 the total amount o

offshore wind capacity required to meet the remainder o the 90 Wh target would be 206

to 235 GW though these estimates would all to 180 to 205 GW i the average load actor

matched the historic average value o 367 or Danish offshore wind installations

27 he question that has to be addressed is whether these load actors are consistent with (a) the

investment programmes that would be required to achieve the total amounts o generation

capacity implied by the governmentrsquos targets and (b) the age profiles o the perormance o

wind arms discussed in this paper he short answer is that this is simply not possible i the

initial load actor or new wind arms continues to decline at the rate observed the past decade

So even as a starting point it is necessary to assume that this underlying deterioration in wind

arm perormance ceases his would be a major change in itsel and it is not obvious what

actors might lead to such a result

28 he results illustrated below are based on a vintage model o perormance that takes account

o the actual expansion o capacity up to the end o 2011 plus whatever investment in new or

replacement turbines is required to achieve the target levels o wind capacity in 2020 o high-light the ull range o potential outcomes these targets are either 10 GW or 15 GW o onshore

9 Department o Energy and Climate Change ndash UK Renewable Energy Road Map July 2011 Figure 2

10 he equivalent average load actor or onshore wind arms was 264

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2052

18 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

wind and either 18 or 24 GW o offshore wind he levels o investment required to meet these

targets depend upon the expected lie o wind turbines For onshore turbines an expected lie

o 10 years is consistent with re-powering decisions over the last decade while an expected lie

o 15 years seems to be the maximum consistent with the age-perormance profile For offshore

turbines the analysis uses expected lives o 15 and 20 years

Figure 6A Projected load factors for UK onshore wind (total capacity of 10 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

Figure 6B Projected load factors for UK onshore wind (total capacity of 15 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

29 Figure 6A shows projected values o the average normalised load actors or UK onshore wind

arms in 2015 and 2020 using a target capacity in 2020 equal to 10 GW plus the age-peror-

mance profiles or the UK and Denmark (DK) Since vintage effects are important in this

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 19

calculation the profile o new or replacement investment affects the results In this case the

level o gross investment is assumed to be constant rom 2013 onwards at either 930 or 1220

MW per year ndash the range is defined by the assumed lie o turbines Achieving a target o 15

GW in 2020 (Figure 6B) would require a level o investment rom 2013 onwards o either 1540

or 1930 MW per year Such figures are 3ndash4 times the average level o investment over theperiod 2005ndash11 so the lower target may provide a better indication o what will happen

30 All o the estimates are calibrated to match the observed average load actor in 2007ndash11 so that

they should be interpreted as relative rather than absolute values Using the age-perormance

profile or UK onshore wind installations the average normalised load actor is likely to all by

about 19 rom 2007ndash11 to 2015 and by 23ndash28 to 2020 I the age-perormance curves or

Danish onshore installations were to apply the reduction in the average normalised load actor

would be considerably less but there would still be a all o up to 5 by 2020

31 In the medium term the all is greater i turbines operate or 15 years because the average age

o all turbines will be higher and thus the degradation in perormance due to age will have a

larger effect By 2020 this effect reduces the average normalised load actor rom 203 or a 10

year lie to 195 or a 15 year lie here is a complex economic and commercial trade-off here

o meet a target o 10 GW in 2020 it is necessary to increase the level o investment in onshore

capacity by 30 rom 2013 Almost all o this will take the orm o re-powering existing wind

arms11 he costs are airly high because this will usually involve new development consents

decommissioning the original turbines and replacing them with a smaller number o larger and

more powerul turbines

32 Estimates o re-powering costs are generally treated as being commercially sensitive but broadproject costs cited in specialist magazines suggest that the unit costs are likely to be at least

pound1000 per kW or more than pound300 million per year in aggregate On the other hand the increase

in the amount o energy generated would be about 12 Wh or a 10 year turbine lie rather

than a 15 year turbine lie though this does not allow or the loss o production during re-pow-

ering Under the best circumstances the payback period may be quite short but the best option

rom a broader economic and social perspective is not obvious

33 Even under the best scenario 10 GW o onshore wind capacity will only generate 178 Wh o

electricity in 2020 well below the lower end o the target range or onshore wind generation

specified in DECCrsquos road map In act allowing or interactions between the age-perormance

profile and investment the amount o onshore wind capacity required in 2020 to generate 24

Wh would be a minimum o 134 GW At the top end o DECCrsquos target range the capacity

required in 2020 to generate 32 Wh would be a minimum o 178 GW Since these estimates

are based on a 10 year turbine lie the level o investment required rom 2013 onwards would

be 1650 MW per year or the bottom o the range and 2350 MW or the top end o the range

In the last 8 years total investment in onshore wind has only ever exceeded 600 MW in one

11 Re-powering is usually defined as the ull replacement o existing turbines whereas retrofitting covers the replace-

ment o various mechanical or electrical components In the data analysis re-powering involves the creation o a

new unit whereas retrofitting affects the perormance o a continuing unit Hence the potential benefits o retrofit-

ting are taken into account in constructing the age-perormance curves

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2252

20 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

year (883 MW in 2008) so it seems rather unlikely that even the lower end o DECCrsquos range or

onshore wind generation will actually be realised

34 O course this assessment would be different i UK onshore wind operators were able to match

the Danish age-perormance curve here are many actors which differentiate the UK and

Danish wind industries he average scale o wind arms in the UK is larger they are moreremote and there appears to be significantly greater variation between operating and potential

new sites in achieved perormance It would thereore require a very optimistic observer to

conclude that the difference in age-perormance curves between the two countries will disap-

pear in the short or medium term

Figure 7 Projected load factors for UK offshore wind in 2020

20

22

24

26

28

30

32

34

18 GW ndash life 15 yrs 18 GW ndash life 20 yrs 24 GW ndash life 15 yrs 24 GW ndash life 20 yrs

L o a d f a c t o r ( )

2007ndash11 2015 2020

Note Based on Danish age-perormance curves Source Authorrsquos estimates

35 Figure 7 shows the results rom carrying out a similar exercise or offshore wind arms in the

UK he only evidence on the age-perormance curves or offshore installations is based upon

the Danish experience so these have been used to construct the projections Further the profile

o investment assumes a progressive increase in additions to capacity rom 2013 to 2016 and

then a stable level o investment rom 2017 to 2020 he DECC road map involves a very large

commitment to offshore wind investment ndash in the range 25-35 GW o new capacity annually

rom 2016 onwards As a consequence the average age o offshore capacity in 2020 will be quite

low and the ull effects o the age-perormance curves will only be elt during the decade ollow-

ing 2020 Even so the average normalised load actor in 2020 will be about 15 lower than or

2007ndash11 while by 2025 it will be about 30 below the base level

36 On these projections 232 GW o offshore capacity is the minimum necessary to generate 58

Wh o electricity rom offshore wind in 2020 his requirement will increase to 298 GW by

2025 simply to offset the effect o the ageing o offshore generation capacity on total output

37 Combining the estimates or onshore and offshore wind generation the total amount o wind

capacity required in 2020 to produce 90 Wh o electricity rom wind will be in the range 39

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 21

to 41 GW once allowance is made or the observed age-perormance curves or both onshore

and offshore installations hese estimates are roughly a third higher than official predictions o

28ndash31 GW o wind capacity in 2020

38 urning the figures round gives a slightly different picture o the reality o the governmentrsquos

strategy on renewable electricity generation A total o 30 GW o wind capacity in 2020 (12GW onshore 18 GW offshore) will generate about 64 Wh o electricity in a normal year

Since electricity generation rom wind and biomass are expected to account or 50 o renew-

able energy in 2020 the consequence is that biomass is expected to account or 53 Wh In

2011 landfill gas and sewage sludge digestion accounted or 44 o electricity rom bioenergy

(DUKES ables 64) here are limited opportunities to expand production rom these sources

so a continuation o historic rates o growth would mean that renewable generation rom other

sources o bioenergy would have to increase rom 72 Wh in 2011 to 46 Wh in 2020 here

are severe constraints on increasing the amounts o electricity generated rom animal biomass

biodegradable municipal solid waste and anaerobic digestion so it would be optimistic to

assume that these will contribute more than 6-8 Wh in 2020 (up rom 26 Wh in 2011)

39 In practice i the projections or renewable energy are to be believed they rest upon an assump-

tion that there will be a large increase in the amount o electricity that is generated rom plant

biomass ndash either on its own or through co-firing with coal his increase will involve increas-

ing the amount o electricity generated in these ways rom 46 Wh in 2011 to 38-40 Wh in

2020 Since the amounts o straw and UK-grown timber available or this purpose are strictly

limited this will come down to the amount o wood chips that can be imported and the cost o

doing this40 he historic average load actor or biomass plants is just over 60 so about 7500 MW o

biomass capacity will be required to supply 40 Wh o electricity he weight o wood chips per

MWh o electricity depends upon moisture content storage boiler design and other actors but

an indicative range is 085 to 115 tonnes per MWh o wood chips with a moisture content o

30 According to DECC and Forestry Commission estimates the UK imported 265 million

tonnes o steam coal and 26 million tonnes o other wood products (including wood chips)

in 2011 It seems that the net effect o the governmentrsquos strategy will be to replace 15 million

tonnes o coal imported in 2011 by 32 to 44 million tonnes o imported wood chips his is a

somewhat strange way o increasing energy security

Conclusion

41 Wind power is a highly capital-intensive technology or generating electricity Its merits rely

entirely upon a substitution o capital or uel inputs he same is true or hydro or tidal or wave

power In comparison with hydro power wind is a low quality resource because o its variability

and because it cannot be stored So the economic case or wind power must rest on obtaining

the most out o the wind that is available

42 While the decline in the achieved perormance o onshore wind turbines in Denmark is

much less than that or the UK or offshore nonetheless the decline in expected output under

standardised wind conditions over 10 years is 10 unweighted and 13 capacity weighted

hese declines accelerate afer age 10 so that the reductions in perormance are 17 and 20

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2452

22 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

respectively afer 15 years For UK onshore wind arms the reduction in perormance due to

age is much worse at 27 unweighted and 69 capacity weighted by age 10

43 Evidence on the perormance o Danish offshore installations is both restricted and so poor

that there may be concern that the results are affected by a small number o outliers Still the

sample contains a reasonable number o sites with at least 5 years o operating experience andthe decline in perormance by age 5 is 38 unweighted and 26 capacity weighted

44 In addition to these results there is strong evidence that the average normalised load actor or

new onshore wind installations in the UK has allen significantly over the period rom 2000 to

2011 his is consistent with a pattern in which the most avourable sites are developed first

Equally it could mean that wind developers have been unable to keep up with the rate o new

investment while maintaining the quality o development and operations For example the site

design or selection o turbine characteristics may make less effective use o the available wind

resources or the sites available than was the case in the past

45 Whatever the reasons the deterioration in initial perormance means that the expected returns

rom the expansion in wind capacity both or investors and in terms o the reduction in CO983090

emissions have been alling without a concomitant decrease in the private and social costs

that are borne by customers and the general public Clearly this is unsatisactory at best and it

suggests that the benefits claimed or current policies cannot be taken at ace value

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 23

Appendix Data and Methods

A Data for the United Kingdom

he raw data used or this study was extracted by the Renewable Energy Foundation rom the

Renewables and CHP Register database compiled by Ogem12 his inormation is used in the

administration o the market in Renewable Obligation Certificates (ROCs) he Renewables

Obligation is the primary mechanism by which large scale generators o renewable electric-

ity receive subsidies so the operators o wind arms have a very strong incentive to submit

complete and up-to-date inormation to the database he data extracted covered all onshore

wind generators with at least 05 MW o generating capacity which are eligible or ROCs or the

10 year period April 2002 to March 2012 he key variables are the monthly output o electric-

ity (used or the purpose o allocating ROCs) and nominal generating capacity here are 282

reporting units in the final dataset A reporting unit may correspond to an entire wind arm orto separate phases o development Complications arise when an existing wind arm is re-pow-

ered ie when old wind turbines are replaced by newer and usually more powerul turbines

he database reports this as a change in the generating capacity o the recording unit but in

this analysis repowering is treated as the termination o the old recording unit and the creation

o a new recording unit

A considerable amount o data checking and cleaning was required beore the data could

be analysed he first step was to add the age o each reporting unit his was calculated by

reerence to the month in which the turbines were commissioned For wind arms in operation

in or beore June 2002 the commissioning date was obtained rom various online sources ndash

notably the thewindpowernet database o UK wind arms13 which was cross-checked against

the inormation provided on the websites o wind operators and other sources For recording

units whose first data relates to dates afer June 2002 the commissioning date was initially

assumed to be the month or which data is first reported Cross-checks were carried out or

all recording units which commenced reporting in the period July 2002 to December 2003 In

more than 80 o cases the externally-reported commissioning date is within one month o the

first month or which data is reported he exceptions ndash Bu Farm and Mablethorpe ndash appear to

be due to delays by small operators in reporting data to Ogem

In a number o cases it is clear rom the data that plants were not in ull operation at the

time when data was first reported to the database here are two indicators o partial or delayed

commissioning

bull For an initial series o months the calculated load actor is very low ndash usually below 5 ndash

and then increases abruptly In such cases the data or the early months is dropped and the

commissioning date is treated as the month immediately preceding the first month with a

lsquonormalrsquo load actor14

12 httpswwwrenewablesandchpogemgovuk

13 See httpwwwthewindpowernetwindarms_list_enphp

14 he month in which a plant is reported as having been commissioned is not included in the analysis

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2652

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 25

the decline in the plantrsquos load actor as it ages As an example consider a plant whose reported

capacity was increased rom 99 MW to 10 MW at age 3 I this was simply a change in report-

ing practice this would reduce the estimated load actor rom age 3 onwards By increasing

the initial capacity rom 99 MW to 10 MW ndash or by reducing the capacity rom 10 MW to 99

MW rom age 3 onwards ndash this bias is removed It is not important which adjustment is madebecause the difference between the alternative assumptions is captured by the unit fixed effect

he numbers o separate sites and o observations per site mean that it is not possible to

provide a simple visual representation o the data Figure 8 provides a summary by illustrating

the distributions o load actors by the age o the wind arm pooled across all years he ranges

o the load actors or each age are wide while the medians show no clear trend by age hus a

naiumlve analysis might conclude that there is nothing to investigate However as will be explained

below the distributions mask crucial differences between the perormance o wind arms as

they age because they do not control or the differences in location and wind availability over

time

B Data for Denmark

he data or Danish wind arms used in this study comes rom a database compiled by the

Danish Energy Agency covering the characteristics and perormance o all wind turbines rom

2002 up to the end o August 201216 he basic recording unit or the register is the individual

wind turbine but where a number o turbines are connected to a single meter an identical

average output per turbine is recorded or all turbines in the group he presence o what are in

effect repeated observations increases the noise in the data In addition the model that is used

or the analysis assumes that site operator and turbine characteristics are common or all o theunits that make up a single wind arm o address this point separate wind arms have been

identified by grouping all turbines which have identical values or turbine manuacturer and

type local authority date o commissioning and date o decommissioning (where applicable)

he register contains inormation on 7569 turbines commissioned rom 1977 onwards his

includes a large number o small turbines which would not all within the scope o the ROC

scheme in the UK Hence or comparability with the UK data all turbines commissioned

prior to 1992 were dropped as well as all wind arms with a generating capacity o less than 1

MW he resulting sample covered 823 onshore wind arms and 30 offshore wind arms with

monthly load actors or various periods rom January 2002 to August 2012 Given the cut-off

or wind arms commissioned prior to 1992 the maximum age or an onshore wind arm was 20

years and the maximum age or an offshore wind arm was 17 years

he total capacity o the onshore wind arms recorded in the Danish database was 2757 MW

he average size o an onshore wind arm in Denmark is small ndash only 33 MW as compared

with 161 MW or the UK ndash because 84 o the sample consists o installations with between

1 and 4 turbines he largest onshore installation has 39 turbines with a capacity o 234 MW

while the onshore installation with the greatest capacity has 19 turbines with a capacity o 584

MW Overall the sample o Danish onshore wind arms is older and smaller than the sample

o UK onshore wind arms reflecting the much longer history o wind generation in Denmark

16 httpwwwensdkEN-USINFOFACSANDFIGURESENERGY_SAISICS_AND_INDICAORS

OVERVIEWOFHEENERGYSECORREGISEROFWINDURBINESSiderForsideaspx

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2852

26 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

he larger scale o UK wind arms may account or some o the differences in perormance

curves or the two countries

he sample o offshore wind arms in Denmark is quite small with 30 installations in

total Still it is useul to analyse the perormance o these installations since there is very little

comparable data or the UK he sample covers 394 turbines with an average o just over 13turbines per installation and an average capacity o 288 MW he largest installation which

was commissioned in 2003 has 72 turbines with a total capacity o 1656 MW Again offshore

installations are smaller and older in Denmark than is the current pattern in the UK he first

two wind arms in the sample were commissioned in the years 1995 and 2000 but the sample

increases significantly rom 2002 onwards o avoid reliance upon a sample o 1 or 2 wind

arms to determine specific coefficients or ages gt 10 years all observations with an age greater

than 10 were omitted rom the analysis17

A somewhat different problem with the offshore wind arm data arose because the coeffi-

cients estimated using capacity weights (see below) were heavily influenced by the largest windarm his has had an extremely erratic profile o perormance over time with a very low average

load actor he treatment o this wind arm is discussed in more detail in Section F below

C Specication and estimation methods

As Figure 8 illustrates it is necessary to go beyond simple summary statistics in order to assess

whether there is any systematic relationship between age and the average perormance o wind

arms he perormance o wind arms varies over both time and space some months and years

have more or less wind than the long term average while specific characteristics o wind arms

ndash including location type o turbine and operating regime ndash will influence the perormance oeach plant under identical wind conditions

he combination o (a) site-specific characteristics which are constant over time and (b)

period-specific characteristics which are constant across wind arms can be represented by

what statisticians call an error components model with fixed effects or each wind arm and

each time period he dependent variable in the model consists o a sequence o load actors ndash

denoted by or unit i in month t ndash or each wind arm (the panel unit) over time he load

actors are calculated by dividing the total output in MWh by (24 x number o days in month x

reported generating capacity in MW) and are multiplied by 100 to convert to percentages

he additive version o the error components model may be written as

(1)

in which denotes the age o plant i in period t and is some unction that is either a

representation o or an approximation to the normalised age-perormance relationship he

multiplicative version is similar except that is replaced by where ln() denotes the

natural logarithm he terms are wind arm or unit fixed effects which capture actors such

as location type o turbine operating regimes etc he terms are period fixed effects which

capture the influence o wind conditions over the UK or Denmark as a whole as well as seasonal

maintenance and demand Without loss o generality it may be assumed that the fixed effects are

17 For the avoidance o doubt this means that no attempt was made to estimate the effect o age on the perormance o

offshore wind arms beyond age 10 while the data or the oldest wind arms was included in the analysis up to and

including age 10

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 27

normalised so that ie all o the error components have a zero mean while the

mean o the random error is zero by definition he terms are random errors which capture

random variations in wind conditions that are not specific to the site turbine breakdowns and

other actors that are uncorrelated with either the site or the time period It is assumed that the

mean o the random error is zero while additional assumptions depend upon the method o

estimation that is adopted

he model specified in equation (1) can be estimated in a variety o ways18 A crucial issue is

whether the are or may be significantly correlated with the age terms or any other independent

regressors I it is assumed that there is no such correlation the model is usually reerred to as

a random effects model In that case the equation can be estimated by a variant o least squares

(LS) generalised least squares (GLS) generalised methods o moments (GMM) or maximum

likelihood (ML) ndash all o which should be yield parameter estimates that will converge in prob-

ability to the true parameter values as the sample size increases An alternative but less restric-

tive specification is known as the fixed effects model in which no assumption is made about

the correlation between the unit fixed effects and other regressors he limitation o the fixed

effects model is that it cannot be used to estimate parameters or regressors that vary across

panel units but are constant over time or a panel unit (time-invariant regressors) ndash eg the

year in which a wind arm was commissioned its location or its rated generating capacity he

influence o such variables is captured by the unit fixed effects

In order to avoid the risk o obtaining biased estimates o the age effects the results reported

in this note are based upon estimates using the fixed effects model he influence o time-invar-

iant regressors is examined in a second stage by regressing the estimates o the unit fixed effectson these variables he simplest method o estimating the fixed effects model is known as the

lsquowithinrsquo estimator in which the mean value or each unit is subtracted rom all observations or

that unit For example i is a general polynomial o order M in A equation (1) becomes

(2)

so that the model can be estimated by least square with the inclusion o dummy variables or

each period t he variance-covariance matrix o the coefficients have been estimated by using

(a) a robust sandwich estimator adjusted or clustering by panel unit and (b) a bootstrap esti-

mator with 400 repetitions he estimates o the standard errors will be consistent even i the

random error component is serially correlated within panel units andor has different variances

across panel units19 Since the number o panel units and the average number o time periods

per panel units are both large (other than or Danish offshore wind arms) the least square esti-

mates o the parameters will be unbiased under these assumptions

he bootstrap method is expensive to carry out but it provides a useul cross-check on

the standard errors generated by other methods Unortunately the bootstrap method o

18 See Chapter 21 o AC Cameron amp PK rivedi ndash Microeconometrics Methods and Applications (Cambridge

Cambridge University Press 2005) or a standard textbook treatment o estimation methods or panel data models

19 he estimation was carried out using the xtreg procedure in Stata Version 12 he methods o constructing the

sandwich and bootstrap estimates o the variance-covariance matrix are described in Chapters 8 amp 13 o AC

Cameron amp PK rivedi ndash Microeconometrics using Stata (Revised Edition College Station exas Stata Press 2010)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 4: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 452

copy Gordon Hughes 2012

Cover background image by Simon Gray (httpwwwstar-oneorguk) via stockxchng

Published by the Renewable Energy Foundation

57ndash58 Russell Square

London WC1B 4HSwwwreorguk

he Renewable Energy Foundation is a registered charity in England and Wales (1107360)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 552

Contents

List o Figures and ables 4

About the author 5

About the Renewable Energy Foundation 5

Authorrsquos Note 6

Note on the Data 6

Executive Summary 7

Te Perormance o Wind Farms in the United Kingdom and Denmark 9

Introduction 9

Age-perormance curves 10

Differences between wind arms 12Implications or uture policy and perormance 15

Conclusion 21

Appendix Data and Methods 23

A Data or the United Kingdom 23

B Data or Denmark 25

C Specification and estimation methods 26

D Period fixed effects vs normalisation by wind speeds 28

E Estimation results or the UK 30F Estimation results or Denmark 36

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 652

List of Figures and TablesList of Figures and Tables

Figure 1 Perormance degradation due to age using equal weights 11

Figure 2 Perormance degradation due to age using capacity weights 12

Figure 3 Range o unit fixed effects by category 13

Figure 4 Initial load actors by year o commissioning or onshore wind 14

Figure 5 Impact o perormance degradation on discounted cumulative output 16

Figure 6A Projected load actors or UK onshore wind (total capacity o 10 GW in 2020) 18

Figure 6B Projected load actors or UK onshore wind (total capacity o 15 GW in 2020) 18

Figure 7 Projected load actors or UK offshore wind in 2020 20

Figure 8 Box plots o load actors by age or UK onshore wind arms 24

Figure 9A Additive age-perormance curves or UK onshore wind arms 32

Figure 9B Multiplicative age-perormance curves or UK onshore wind arms 32

Figure 10 UK onshore age-perormance curves using equal and capacity weights 33

Figure 11A Residuals by age or perormance curves using equal weights 34

Figure 11B Residuals by age or perormance curves using capacity weights 34

Figure 12 Unit fixed effects by year o commissioning or England and Scotland 35

Figure 13A Age-perormance curves or Danish onshore wind arms 37

Figure 13B Age-perormance curves or Danish offshore wind arms 37

Figure 14 Danish age-perormance curves using equal and capacity weights 38

able 1 Average load actors by year and country () 40

able 2 Estimation results or UK onshore wind arms 41

able 3 Equations or trends in unit fixed effects 43

able 4 Estimation results or Danish wind arms 44

able 5 Perormance by year o commissioning or Danish wind arms 46

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 752

About the author

Dr Gordon Hughes is a Proessor o Economics at the University o Edinburgh where he teaches courses in

the Economics o Natural Resources and Public Economics He was a senior adviser on energy and environ-

mental policy at the World Bank until 2001 He has advised governments on the design and implementationo environmental policies and was responsible or some o the World Bankrsquos most important environmental

guidelines

About the Renewable Energy Foundation

he Renewable Energy Foundation is a registered charity promoting sustainable development or the benefit

o the public by means o energy conservation and the use o renewable energy

REF is supported by private donation and has no political affiliation or corporate membership In pursuit o

its principal goals REF highlights the need or an overall energy policy that is balanced ecologically sensi-

tive and effective

We aim to raise public awareness o the issues and encourage inormed debate regarding a structured energy

policy that is both ecologically sensitive and practical he issues o climate change and security o energy

supply are complex and closely intertwined REF contributes to the debate surrounding these issues by

commissioning reports to provide an independent and authoritative source o inormation

For urther inormation see wwwreorguk

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 852

Authorrsquos Note

Note he views expressed in this paper are strictly personal and do not reflect the position o any organ-

isation with which I am associated I am grateul to John Constable Lee Moroney and a reeree or their

comments on earlier drafs o the paper he results o this study have been presented at seminars andlectures to various groups in Edinburgh and Glasgow I have benefited rom the questions and comments

rom participants in those events

Gordon Hughes

School o Economics

University o Edinburgh

gahughesedacuk

Note on the Data

For the convenience o interested researchers both the raw and cleaned data employed in this analysis will

published alongside the electronic version o the study on the REF website wwwreorguk

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 7

Executive Summary

1 Onshore wind turbines represent a relatively mature technology which ought to have achieved

a satisactory level o reliability in operation as plants age Unortunately detailed analysis o

the relationship between age and perormance gives a rather different picture or both theUnited Kingdom and Denmark with a significant decline in the average load actor o onshore

wind arms adjusted or wind availability as they get older An even more dramatic decline is

observed or offshore wind arms in Denmark but this may be a reflection o the immaturity

o the technology

2 he study has used data on the monthly output o wind arms in the UK and Denmark reported

under regulatory arrangements and schemes or subsidising renewable energy Normalised

age-perormance curves have been estimated using standard statistical techniques which allow

or differences between sites and over time in wind resources and other actors

3 he normalised load actor or UK onshore wind arms declines rom a peak o about 24 at

age 1 to 15 at age 10 and 11 at age 15 he decline in the normalised load actor or Danish

onshore wind arms is slower but still significant with a all rom a peak o 22 to 18 at age 15

On the other hand or offshore wind arms in Denmark the normalised load actor alls rom

39 at age 0 to 15 at age 10 he reasons or the observed declines in normalised load actors

cannot be ully assessed using the data available but outages due to mechanical breakdowns

appear to be a contributory actor

4 Analysis o site-specific perormance reveals that the average normalised load actor o new UK

onshore wind arms at age 1 (the peak year o operation) declined significantly rom 2000 to2011 In addition larger wind arms have systematically worse perormance than smaller wind

arms Adjusted or age and wind availability the overall perormance o wind arms in the UK

has deteriorated markedly since the beginning o the century

5 hese findings have important implications or policy towards wind generation in the UK First

they suggest that the subsidy regime is extremely generous i investment in new wind arms

is profitable despite the decline in perormance due to age and over time Second meeting

the UK Governmentrsquos targets or wind generation will require a much higher level o wind

capacity ndash and thus capital investment ndash than current projections imply hird the structure

o contracts offered to wind generators under the proposed reorm o the electricity market

should be modified since ew wind arms will operate or more than 12ndash15 years

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 9

The Performance of Wind Farms in the United Kingdom

and Denmark

Introduction

1 Any assessment o the costs o wind power must rely heavily upon assumptions about theaverage load actor that will be achieved by new wind installations over their lietime It is

standard practice to calculate average load actors by year and country or onshore and offshore

installations as shown in able 1 (page 40)1 However such estimates do not provide a

reliable statistical basis or assessing the uture perormance o wind arms in aggregate Part

o the reason is that the amount o wind in any month or year is influenced by long term mete-

orological cycles that have periods o many years notably the North Atlantic Oscillation In

addition average load actors do not allow or changes in the composition o wind installations

by location age size and other actors

2 As the number o wind arms operating in developed countries has grown engineers have an

opportunity to analyse operating experience over extended periods o time With any (relatively)

novel technology the incidence o temporary or permanent breakdowns may be expected to all

over time so it is difficult to disentangle the effects o age on operating perormance rom the

technological immaturity o turbines installed 10 or 20 years ago Nonetheless the technology

or onshore wind turbines has been reasonably mature since the early 2000s his is documented

by data produced by the US Department o Energy which shows that the decline in capital costs

that is characteristic o immature technologies slowed afer 2000 and was reversed afer 20042

3 With at least 10 years o operating data since 2000 it should be possible to examine how thetypical operating perormance o wind installations in the United Kingdom and Denmark

varies with the age o the turbines he estimated age-perormance curves should then be

incorporated in estimates o levelised costs which are ofen used to compare the costs o wind

and other orms o generation

4 he Appendix provides a detailed description o the data and methods employed or this study

In both the UK and Denmark wind operators have a large incentive to produce electricity

whenever there is sufficient wind available since marginal operating costs are small while the

operator receives a much higher price In the UK this price is the sum o the market price plus

the value o (a) the Renewable Obligation Certificates (ROCs) awarded or each MWh o elec-

tricity produced and (b) the associated exemption rom the Climate Change Levy In practice

the effective market value o ROCs and associated incentives means that an onshore wind

operator earns roughly double the average market price o electricity per MWh In Denmark

most onshore wind capacity receives a price premium equivalent to about pound28 per MWh on top

1 he load actor is calculated as the ratio o the amount o electricity actually produced by a turbine or wind arm

over a period o a month or a year divided by the amount o output that would have been produced had it operated

at ull nameplate capacity or the entire period his is expressed as a percentage so that reported load actors lie

between 0 and 100

2 R Wiser amp M Bolinger (2012) 2011 Wind Technologies Market Report Lawrence Berkeley National Laboratory

US Department o Energy It should be noted that the installed cost o wind plants has allen since 2010 but these

changes reflect the cyclical orces o demand and supply that are well established across the electricity sector and

which apply to ossil-uel as well as renewable generators

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1252

10 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

o the market price or about 125 years o operation at a typical load actor3 Offshore capacity

is eligible or a similar price premium but most o it is contracted via tenders that offer a guar-

anteed eed-in tariff he tender tariffs have varied rom pound56 per MWh to twice that figure

extending over 16-18 years o operation at a typical load actor

Age-performance curves

5 A standard indicator o the operating perormance o a wind turbine is its normalised load

actor defined as the total output over some period divided by the maximum potential output

adjusted or wind availability Normalisation or wind availability is not straightorward I

detailed data on wind speeds at specific locations is available then the potential output rom

individual turbines or wind arms can be calculated using standard power curves Since wind

speeds can vary substantially over small distances such calculations are both difficult to imple-

ment and prone to errors when applied to large numbers o wind generators using public

sources o data4

6 Many analyses work in the opposite direction For example the Danish Wind Index ndash the

longest running series o data on wind availability in Europe ndash is constructed rom the common

component o monthly variations in wind output rom a large sample o wind turbines5 his

study adopts a similar approach Wind availability is treated as an unobserved actor whose

contribution to monthly variations in wind output is estimated by a series o fixed period

(monthyear) effects in a statistical model o monthly output or all wind arms he reasons

or adopting this approach are discussed in Section D o the Appendix his explains why the

method must perorm better than the easible alternative o using a single index o average

wind speed or the UK or Denmark

7 he analysis relies upon standard techniques that are widely used in the biological and medical

sciences as well economics and engineering Full details are given in the Appendix he idea is

to treat the actual load actor (ie without adjustment or wind conditions) or a specific wind

generator in any time period as being determined by a set o components which reflect specific

site conditions and the age-related perormance o wind plants in general together with the

period fixed effects and an uncorrelated random error he onshore wind datasets or the UK

and Denmark used or the analysis are large with monthly observations on 282 installations in

the UK and 823 installations in Denmark with an age range rom 0 to 19 years he offshore

wind dataset or Denmark is rather smaller covering only 30 installations but it can be used to

obtain reasonable estimates o perormance up to age 10 he results discussed here are based

upon least squares estimation (see page 27) combined with standard errors that are robust to

various departures rom classical assumptions about the nature o the data generation process

8 he results o the statistical analysis demonstrate an unambiguous and statistically significant

decline in the operating perormance o wind arms as they grow older Figure 1 shows the

3 Conversions to GBP are based on an exchange rate o pound1 = DKK 918 (as at 19 October 2012)

4 All wind generators collect their own inormation on wind speeds in order to manage their plants but they do not

publish such inormation and will not have access to inormation collected by other operators

5 Details o the construction o the Danish Wind Index are given at wwwvindstatdk A paper by Boccard ndash N Boccard

(2009) lsquoCapacity actor o wind power realized values vs estimatesrsquo Energy Policy Vol 37 pp 2679-88 ndash includes a

discussion o the construction o similar indices o wind availability in Germany the UK Netherlands and Sweden

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 11

simplest variant o the normalised age-perormance curve or onshore UK wind installations

together with the equivalent curves or onshore and offshore Danish installations he curves

illustrated are calculated rom the multiplicative error components model described in the

Appendix with log(load actor) as the dependent variable and a quadratic in plant age to repre-

sent the variation in plant perormance with age9 he rate o decline in perormance is greatest or offshore installations in Denmark with a all

rom load actors o over 40 at ages 0 and 1 to less than 15 by at ages 9 and 10 Onshore

installations in the UK show a more rapid rate o decline ndash 09 percentage points per year over

the first 10 years o operation ndash than is the case or Denmark though the normalised Danish

perormance curve lies below the UK curve or the first our years For the UK the normalised

load actor alls to just over 15 at age 10 and to 11 at age 15 With such low load actors it

seems likely that many wind arms will be re-powered ndash ie the turbines will be replaced ndash once

they reach the age o 10 or at most 15

Figure 1 Performance degradation due to age using equal weights

0

5

10

15

20

25

30

35

40

45

50

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

L o

a d f a c t o r ( )

Age (years)

UK onshore DK onshore DK offshore

Note Normalised load actors in Source Authorrsquos estimates

10 here are two plausible explanations or the observed decline in average load actors as wind

arms age he first is that the turbines become less efficient over time as a result o mechanical

wear and tear erosion o the turbine blades and related actors he second is that the turbines

experience more requent breakdowns and their operators take more time to bring them back

into service because they are less concerned about the perormance o older plants Both

reasons may be relevant in different circumstances and it is not possible to identiy a primary

explanation rom the data he requency o extended shutdowns does seem to increase with

age but this could be a reflection o the timing o planned maintenance operations rather than

breakdowns Whatever the cause the reduction in perormance with age is much greater than

would be expected or thermal generating plants

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1452

12 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

11 he age-perormance curves in Figure 1 are derived rom equations estimated by giving equal

weight to each wind arm irrespective o its generating capacity Figure 2 illustrates similar

curves but in this case the estimates are constructed by weighting each wind arm by its generat-

ing capacity reerred to as capacity-weighted his gives a better representation o perormance

degradation per MW o generating capacity he striking result is that the rate o decline inthe perormance o UK onshore wind installations is significantly aster when capacity weights

are used which implies that large wind arms in the UK experience a more rapid decline than

smaller ones he differences are smaller in Denmark but the capacity-weighted decline in

perormance is more rapid than the equal weights decline or Danish onshore installations

Figure 2 Performance degradation due to age using capacity weights

0

5

10

15

20

25

30

35

40

45

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

L o a d f a c t o r ( )

Age (years)

UK onshore DK onshore DK offshore

Note Normalised load actors in Source Authorrsquos estimates

Differences between wind farms

12 he specification o the statistical model means that the unit fixed effects identiy the specific

perormance characteristics o individual wind arms afer adjusting or wind availability and

age hese characteristics may reflect the site andor design o the installation as well as the

way in which it is operated A positive value or the unit fixed effect is equivalent to shifing

the perormance curve up by some percentage applied uniormly at each age while a negative

value shifs it down It ollows that the distribution o unit fixed effects across installations can

be used to compare the relative effectiveness o new installations by location date o commis-

sioning etc Care is required when making comparisons between or example UK and Danish

onshore wind installations because these have been normalised separately so the unit fixed

effects reer to different underlying averages6

6 Some care is required in interpreting the unit fixed effects By deault the average value o the unit fixed effects or

each group is zero but these averages are taken over all observations or each installation in the basic sample For the

purpose o the comparisons in Figure 3 the unit fixed effects have been adjusted so that the average values or each

group are zero when using one value or each installation

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 13

13 o give a sense o the numbers unit fixed effects o (a) 02 or (b) -02 mean that the base load

actor given the age o the plant is multiplied by (a) exp(02) = 122 or (b) exp(-02) = 082 Since

the base load actor or a plant o age 0 is 231 these fixed effects translate to load actors in

the first year o operation o (a) 282 and (b) 189 For a wind arm with 50 MW o gener-

ating capacity the difference amounts to an additional 40700 MWh o electricity output peryear Using the standard figures cited by RenewableUK this would be sufficient to supply 12300

homes or a year his range (040) is slightly greater than the interquartile range o unit fixed

effects or UK onshore wind arms (035) Again or comparison operators o thermal power

plants would regard any difference o even 10 up or down in efficiency relative to what is

expected as a striking and perhaps worrying indication o either good or bad perormance

Figure 3 Range of unit fixed effects by category

- 1

- 5

0

5

1

A d j f i x e d e f f e c t ndash q u a d r a t i c

UK onshore DK onshore DK offshore

Source Authorrsquos estimates

14 A notable eature o the distributions o the unit fixed effects that are summarised in the box

plots shown in Figure 3 is the range o variation between the best and worst perormers espe-

cially or UK onshore and Danish offshore wind arms Box plots are constructed so that the

top and bottom o the box correspond to the upper and lower quartiles while the bars at topand bottom correspond to the maximum and minimum values excluding outliers7 A plant at

the upper quartile o the distribution o UK onshore wind arms will generate an annual output

that is about 40 higher than a plant at the lower quartile he equivalent figure or Danish

onshore wind plants is an increase o 30 while or Danish offshore wind arms the difference

between the upper and lower quartiles is equivalent to 25 times the output o a lower quartile

plant By any standard such differences are important

7 he upper quartile is the point in a distribution such that the 25 o values exceed it while the lower quartile is the

point in a distribution such that 75 o values exceed it he line in the middle o the box marks the median the

point exceeded by 50 o values Outliers are defined as values that lie outside the range o ukey adjacent values

which are defined as the upperlower quartiles +- 15 x interquartile range

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1652

14 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

15 Outside the range between the upper and lower quartiles the top and bottom segments o

the distributions extend urther up and down or UK onshore wind arms than or those in

Denmark he rather different pattern or Danish offshore wind installations is partly a conse-

quence o the relatively small size o the sample but the evidence points to considerably greater

variability in the perormance o offshore wind installations than or onshore installations16 It is somewhat surprising to observe the magnitude o the differences in the perormance in

onshore wind arms in the UK Given the nature o the subsidy regimes and the high capital

cost o developing new installations it might be expected that operators have a strong incen-

tive to identiy the best locations and then choose equipment that will deliver the maximum

amount o electricity output at a high level o reliability For UK onshore operators it seems that

good or bad perormance is somewhat o a lottery However i the subsidies provided by ROCs

are sufficient to underwrite investment in inefficient plants ndash as appears to be the case ndash then

those subsidies are extremely generous or plants that operate close to the efficient rontier

As location is likely to be the main actor that determines the perormance o a specific plant

relative to all other plants the inerence must be that many wind plants have been developed on

sites with poor wind characteristics

Figure 4 Initial load factors by year of commissioning for onshore wind

10

15

20

25

30

35

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

I n i t i a l l o a d f a c t o r ( )

Year commissioned

UK onshore DK onshore Linear (UK onshore) Linear (DK onshore)

Source Authorrsquos estimates

17 here is another disturbing characteristic o UK wind arms which is revealed by examination

o the unit fixed effects and illustrated in Figure 4 It might be expected that the average peror-

mance o recently installed units would be higher than that o older units his should reflect

improvements in turbine design and reliability as well as in the identification o good sites or

wind generation However that is not the pattern or onshore wind arms in the UK he figure

shows the average values o the initial load actors calculated using equally weighted data by

year o commissioning or onshore wind arms in the UK and Denmark together with fitted

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 15

trend lines8 here has been a clear decline in the initial load actors in the UK which has not

occurred in Denmark his pattern is confirmed by more detailed statistical results shown in

the Appendix which take account o location and size o installation

18 he decline in initial perormance is particularly obvious afer 2005 when the rate o commis-

sioning new capacity increased sharply rom less than 500 MW in the 5 years 2000ndash04 to morethan 500 MW per year rom 2005 onwards In contrast the peak years or commissioning new

onshore wind arms in Denmark were 2000ndash02 when 920 MW was installed he rate o new

building (or replacement) picked up again in 2008 but it is still much lower than at the turn o

century Not only is the perormance o the onshore wind plants in the UK commissioned in

recent years significantly worse than that or wind arms commissioned beore 2005 but it is

possible that this is a direct consequence o an overly rapid expansion o capacity In essence

the evidence suggests that the industry does not have the capability to identiy develop and

operate new onshore wind arms at the rate envisaged by UK government targets while main-

taining a satisactory average level o perormance

Implications for future policy and performance

19 he pattern o perormance degradation with age and over time identified in this analysis has

implications or the assessment o new investment in wind arms and or the design o the

proposed Electricity Market Reorm (EMR) here are two related issues that are affected by

the results

a) What are the implications or the design o policies intended to promote the adoption

and expansion o wind generation in the UK his includes the structure o current

subsidies as well as the specification o the contracts that are proposed under the EMR A

related question concerns the expected lie o new wind arms

b) How might the average perormance o wind generation develop over the next decade

his is absolutely critical to any assessment o the amount o wind capacity that will be

required to meet the UK governmentrsquos targets or the share o electricity ndash and more

generally energy ndash rom renewable sources he government has assumed that the

average load actors or both onshore and offshore wind arms will either remain stable

or increase in uture he prospects will be very different i these assumptions are not

borne out

20 he first question can be addressed by considering the discounted sum o cumulative net

output rom a new wind arm on the assumption that its perormance matches the averages

identified in the analysis his takes account o both perormance degradation over time plus

the annual cost o maintaining the plant which may be expressed as a deduction rom gross

production so as to calculate the contribution o expected output at age s to recovering the

original capital cost o building the plant hese figures have to be discounted back to the date

o commissioning in order to assess the net present value o the initial investment in the plant

he curves or constant and observed perormance using a discount rate o 9 (a typical cost o

capital or wind generation) are shown in Figure 5 he black line is based on the ull set o age

8 he initial load actors are calculated as the normalised load actor or UKDanish wind arms at age 0 multiplied by

where is the unit fixed effect or installation i

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1852

16 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

effects while the red dashed line is based on the smoothed quadratic perormance curve and

the blue dashed line shows what the pattern would be i operating perormance was not affected

by age he calculation is extended out to age 25 since DECCrsquos standard comparisons assume

that wind turbines have a lie o 25 years

Figure 5 Impact of performance degradation on discounted cumulative output

0

10

20

30

40

50

60

70

80

90

100

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

o

f d i s c o u n t e d c u m u l a t i v e o u t p u t

Age of plant (years)

Constant Quadratic Full age effects

Source Authorrsquos estimates

21 he key points in the figure are that 80 o the discounted cumulative output o a new wind

plant is likely to be produced in the first 10 years o its lie and 90 in the first 14 years his

is consistent with the structure o Danish subsidies or onshore wind arms which extend over

a typical period o 125 years Since sites or wind arms are scarce and involve the payment o

significant rents that may be linked to output it is very unlikely that any new installation o

wind turbines will have an expected lie o more than 15 years Instead as has happened in the

past wind operators will have a strong incentive to decommission plants afer no more than 15

years and replace the turbines with newer equipment

22 As a consequence any economic assessment o wind generation should not be based on an

expected lie which is longer than 15 years In recent work reported in evidence to the House

o Commons Select Committee on Energy and Climate Change I assumed that wind plants

would have a residual value equal to 20 o their initial cost in real terms at the end o 15 years

he analysis in this paper suggests that this is too avourable an assumption Given the costs o

decommissioning old turbines the residual value is likely to be well below 10 o their initial

cost and the decision point may be at 10 rather than 15 years

23 he corollary o this observation is that it makes little sense to offer long term contracts or 20

years or more that guarantee prices or eed-in tariffs (Fis) to wind operators A contract length

o 10 to 12 years would be sufficient to remove most o the market risk associated with invest-

ment in wind generation In this respect the subsidy arrangements implemented in Denmark

are better designed

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 17

24 At the same time the offer o subsidies andor guaranteed prices may have serious adverse

consequences or the efficiency o wind generation Returning to the unit fixed effects measur-

ing the perormance o wind plants commissioned afer 2005 only 28 out o 159 units have an

operating perormance that exceeds the average or the period hose 28 units account or 360

MW o capacity whereas the remaining 131 units account or 2810 MW o capacity Not onlyare recent plants less efficient than the average or the whole period but the plants which are

below-average in efficiency are typically larger than the more efficient ones and account or a

disproportionate share o the recent additions to generating capacity

25 In mid-2012 there was approximately 4500 MW o onshore wind capacity operating in the

UK By 2020 it is expected that there will be at least 10000 MW (based on proposed wind

arms with planning consent) and perhaps as much 15000 MW o wind capacity in operation

he government is relying upon a substantial contribution rom onshore wind arms towards

meeting its targets or renewable energy in 2020 o provide context the governmentrsquos targets

or renewable energy in 2020 speciy a total o 234 Wh o energy rom all renewable sources

o which 90ndash140 Wh is expected to come rom wind and biomass electricity generation9

he projections or biomass electricity are extremely vague and have to be reconciled with

an assumed increase o 3ndash4 times in the amount o biomass used separately or heat So in

concrete terms it is reasonable to assume that wind generation will account or at least 90 Wh

with 24ndash32 Wh rom onshore generation and the remainder rom offshore

26 he average load actor or onshore wind over the 10 year period rom April 2002 to March

2012 was 256 which is influenced by the large amounts o new capacity added rom 2005

onwards I this load actor were achieved in 2020 the total amount o onshore wind capacityrequired to meet the onshore generation target would be 107 to 143 GW he average load

actor or offshore wind installations (on an unchanged configuration basis) or the five year

period 2007ndash11 was 32010 I this load actor were achieved in 2020 the total amount o

offshore wind capacity required to meet the remainder o the 90 Wh target would be 206

to 235 GW though these estimates would all to 180 to 205 GW i the average load actor

matched the historic average value o 367 or Danish offshore wind installations

27 he question that has to be addressed is whether these load actors are consistent with (a) the

investment programmes that would be required to achieve the total amounts o generation

capacity implied by the governmentrsquos targets and (b) the age profiles o the perormance o

wind arms discussed in this paper he short answer is that this is simply not possible i the

initial load actor or new wind arms continues to decline at the rate observed the past decade

So even as a starting point it is necessary to assume that this underlying deterioration in wind

arm perormance ceases his would be a major change in itsel and it is not obvious what

actors might lead to such a result

28 he results illustrated below are based on a vintage model o perormance that takes account

o the actual expansion o capacity up to the end o 2011 plus whatever investment in new or

replacement turbines is required to achieve the target levels o wind capacity in 2020 o high-light the ull range o potential outcomes these targets are either 10 GW or 15 GW o onshore

9 Department o Energy and Climate Change ndash UK Renewable Energy Road Map July 2011 Figure 2

10 he equivalent average load actor or onshore wind arms was 264

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2052

18 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

wind and either 18 or 24 GW o offshore wind he levels o investment required to meet these

targets depend upon the expected lie o wind turbines For onshore turbines an expected lie

o 10 years is consistent with re-powering decisions over the last decade while an expected lie

o 15 years seems to be the maximum consistent with the age-perormance profile For offshore

turbines the analysis uses expected lives o 15 and 20 years

Figure 6A Projected load factors for UK onshore wind (total capacity of 10 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

Figure 6B Projected load factors for UK onshore wind (total capacity of 15 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

29 Figure 6A shows projected values o the average normalised load actors or UK onshore wind

arms in 2015 and 2020 using a target capacity in 2020 equal to 10 GW plus the age-peror-

mance profiles or the UK and Denmark (DK) Since vintage effects are important in this

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 19

calculation the profile o new or replacement investment affects the results In this case the

level o gross investment is assumed to be constant rom 2013 onwards at either 930 or 1220

MW per year ndash the range is defined by the assumed lie o turbines Achieving a target o 15

GW in 2020 (Figure 6B) would require a level o investment rom 2013 onwards o either 1540

or 1930 MW per year Such figures are 3ndash4 times the average level o investment over theperiod 2005ndash11 so the lower target may provide a better indication o what will happen

30 All o the estimates are calibrated to match the observed average load actor in 2007ndash11 so that

they should be interpreted as relative rather than absolute values Using the age-perormance

profile or UK onshore wind installations the average normalised load actor is likely to all by

about 19 rom 2007ndash11 to 2015 and by 23ndash28 to 2020 I the age-perormance curves or

Danish onshore installations were to apply the reduction in the average normalised load actor

would be considerably less but there would still be a all o up to 5 by 2020

31 In the medium term the all is greater i turbines operate or 15 years because the average age

o all turbines will be higher and thus the degradation in perormance due to age will have a

larger effect By 2020 this effect reduces the average normalised load actor rom 203 or a 10

year lie to 195 or a 15 year lie here is a complex economic and commercial trade-off here

o meet a target o 10 GW in 2020 it is necessary to increase the level o investment in onshore

capacity by 30 rom 2013 Almost all o this will take the orm o re-powering existing wind

arms11 he costs are airly high because this will usually involve new development consents

decommissioning the original turbines and replacing them with a smaller number o larger and

more powerul turbines

32 Estimates o re-powering costs are generally treated as being commercially sensitive but broadproject costs cited in specialist magazines suggest that the unit costs are likely to be at least

pound1000 per kW or more than pound300 million per year in aggregate On the other hand the increase

in the amount o energy generated would be about 12 Wh or a 10 year turbine lie rather

than a 15 year turbine lie though this does not allow or the loss o production during re-pow-

ering Under the best circumstances the payback period may be quite short but the best option

rom a broader economic and social perspective is not obvious

33 Even under the best scenario 10 GW o onshore wind capacity will only generate 178 Wh o

electricity in 2020 well below the lower end o the target range or onshore wind generation

specified in DECCrsquos road map In act allowing or interactions between the age-perormance

profile and investment the amount o onshore wind capacity required in 2020 to generate 24

Wh would be a minimum o 134 GW At the top end o DECCrsquos target range the capacity

required in 2020 to generate 32 Wh would be a minimum o 178 GW Since these estimates

are based on a 10 year turbine lie the level o investment required rom 2013 onwards would

be 1650 MW per year or the bottom o the range and 2350 MW or the top end o the range

In the last 8 years total investment in onshore wind has only ever exceeded 600 MW in one

11 Re-powering is usually defined as the ull replacement o existing turbines whereas retrofitting covers the replace-

ment o various mechanical or electrical components In the data analysis re-powering involves the creation o a

new unit whereas retrofitting affects the perormance o a continuing unit Hence the potential benefits o retrofit-

ting are taken into account in constructing the age-perormance curves

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2252

20 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

year (883 MW in 2008) so it seems rather unlikely that even the lower end o DECCrsquos range or

onshore wind generation will actually be realised

34 O course this assessment would be different i UK onshore wind operators were able to match

the Danish age-perormance curve here are many actors which differentiate the UK and

Danish wind industries he average scale o wind arms in the UK is larger they are moreremote and there appears to be significantly greater variation between operating and potential

new sites in achieved perormance It would thereore require a very optimistic observer to

conclude that the difference in age-perormance curves between the two countries will disap-

pear in the short or medium term

Figure 7 Projected load factors for UK offshore wind in 2020

20

22

24

26

28

30

32

34

18 GW ndash life 15 yrs 18 GW ndash life 20 yrs 24 GW ndash life 15 yrs 24 GW ndash life 20 yrs

L o a d f a c t o r ( )

2007ndash11 2015 2020

Note Based on Danish age-perormance curves Source Authorrsquos estimates

35 Figure 7 shows the results rom carrying out a similar exercise or offshore wind arms in the

UK he only evidence on the age-perormance curves or offshore installations is based upon

the Danish experience so these have been used to construct the projections Further the profile

o investment assumes a progressive increase in additions to capacity rom 2013 to 2016 and

then a stable level o investment rom 2017 to 2020 he DECC road map involves a very large

commitment to offshore wind investment ndash in the range 25-35 GW o new capacity annually

rom 2016 onwards As a consequence the average age o offshore capacity in 2020 will be quite

low and the ull effects o the age-perormance curves will only be elt during the decade ollow-

ing 2020 Even so the average normalised load actor in 2020 will be about 15 lower than or

2007ndash11 while by 2025 it will be about 30 below the base level

36 On these projections 232 GW o offshore capacity is the minimum necessary to generate 58

Wh o electricity rom offshore wind in 2020 his requirement will increase to 298 GW by

2025 simply to offset the effect o the ageing o offshore generation capacity on total output

37 Combining the estimates or onshore and offshore wind generation the total amount o wind

capacity required in 2020 to produce 90 Wh o electricity rom wind will be in the range 39

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 21

to 41 GW once allowance is made or the observed age-perormance curves or both onshore

and offshore installations hese estimates are roughly a third higher than official predictions o

28ndash31 GW o wind capacity in 2020

38 urning the figures round gives a slightly different picture o the reality o the governmentrsquos

strategy on renewable electricity generation A total o 30 GW o wind capacity in 2020 (12GW onshore 18 GW offshore) will generate about 64 Wh o electricity in a normal year

Since electricity generation rom wind and biomass are expected to account or 50 o renew-

able energy in 2020 the consequence is that biomass is expected to account or 53 Wh In

2011 landfill gas and sewage sludge digestion accounted or 44 o electricity rom bioenergy

(DUKES ables 64) here are limited opportunities to expand production rom these sources

so a continuation o historic rates o growth would mean that renewable generation rom other

sources o bioenergy would have to increase rom 72 Wh in 2011 to 46 Wh in 2020 here

are severe constraints on increasing the amounts o electricity generated rom animal biomass

biodegradable municipal solid waste and anaerobic digestion so it would be optimistic to

assume that these will contribute more than 6-8 Wh in 2020 (up rom 26 Wh in 2011)

39 In practice i the projections or renewable energy are to be believed they rest upon an assump-

tion that there will be a large increase in the amount o electricity that is generated rom plant

biomass ndash either on its own or through co-firing with coal his increase will involve increas-

ing the amount o electricity generated in these ways rom 46 Wh in 2011 to 38-40 Wh in

2020 Since the amounts o straw and UK-grown timber available or this purpose are strictly

limited this will come down to the amount o wood chips that can be imported and the cost o

doing this40 he historic average load actor or biomass plants is just over 60 so about 7500 MW o

biomass capacity will be required to supply 40 Wh o electricity he weight o wood chips per

MWh o electricity depends upon moisture content storage boiler design and other actors but

an indicative range is 085 to 115 tonnes per MWh o wood chips with a moisture content o

30 According to DECC and Forestry Commission estimates the UK imported 265 million

tonnes o steam coal and 26 million tonnes o other wood products (including wood chips)

in 2011 It seems that the net effect o the governmentrsquos strategy will be to replace 15 million

tonnes o coal imported in 2011 by 32 to 44 million tonnes o imported wood chips his is a

somewhat strange way o increasing energy security

Conclusion

41 Wind power is a highly capital-intensive technology or generating electricity Its merits rely

entirely upon a substitution o capital or uel inputs he same is true or hydro or tidal or wave

power In comparison with hydro power wind is a low quality resource because o its variability

and because it cannot be stored So the economic case or wind power must rest on obtaining

the most out o the wind that is available

42 While the decline in the achieved perormance o onshore wind turbines in Denmark is

much less than that or the UK or offshore nonetheless the decline in expected output under

standardised wind conditions over 10 years is 10 unweighted and 13 capacity weighted

hese declines accelerate afer age 10 so that the reductions in perormance are 17 and 20

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2452

22 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

respectively afer 15 years For UK onshore wind arms the reduction in perormance due to

age is much worse at 27 unweighted and 69 capacity weighted by age 10

43 Evidence on the perormance o Danish offshore installations is both restricted and so poor

that there may be concern that the results are affected by a small number o outliers Still the

sample contains a reasonable number o sites with at least 5 years o operating experience andthe decline in perormance by age 5 is 38 unweighted and 26 capacity weighted

44 In addition to these results there is strong evidence that the average normalised load actor or

new onshore wind installations in the UK has allen significantly over the period rom 2000 to

2011 his is consistent with a pattern in which the most avourable sites are developed first

Equally it could mean that wind developers have been unable to keep up with the rate o new

investment while maintaining the quality o development and operations For example the site

design or selection o turbine characteristics may make less effective use o the available wind

resources or the sites available than was the case in the past

45 Whatever the reasons the deterioration in initial perormance means that the expected returns

rom the expansion in wind capacity both or investors and in terms o the reduction in CO983090

emissions have been alling without a concomitant decrease in the private and social costs

that are borne by customers and the general public Clearly this is unsatisactory at best and it

suggests that the benefits claimed or current policies cannot be taken at ace value

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 23

Appendix Data and Methods

A Data for the United Kingdom

he raw data used or this study was extracted by the Renewable Energy Foundation rom the

Renewables and CHP Register database compiled by Ogem12 his inormation is used in the

administration o the market in Renewable Obligation Certificates (ROCs) he Renewables

Obligation is the primary mechanism by which large scale generators o renewable electric-

ity receive subsidies so the operators o wind arms have a very strong incentive to submit

complete and up-to-date inormation to the database he data extracted covered all onshore

wind generators with at least 05 MW o generating capacity which are eligible or ROCs or the

10 year period April 2002 to March 2012 he key variables are the monthly output o electric-

ity (used or the purpose o allocating ROCs) and nominal generating capacity here are 282

reporting units in the final dataset A reporting unit may correspond to an entire wind arm orto separate phases o development Complications arise when an existing wind arm is re-pow-

ered ie when old wind turbines are replaced by newer and usually more powerul turbines

he database reports this as a change in the generating capacity o the recording unit but in

this analysis repowering is treated as the termination o the old recording unit and the creation

o a new recording unit

A considerable amount o data checking and cleaning was required beore the data could

be analysed he first step was to add the age o each reporting unit his was calculated by

reerence to the month in which the turbines were commissioned For wind arms in operation

in or beore June 2002 the commissioning date was obtained rom various online sources ndash

notably the thewindpowernet database o UK wind arms13 which was cross-checked against

the inormation provided on the websites o wind operators and other sources For recording

units whose first data relates to dates afer June 2002 the commissioning date was initially

assumed to be the month or which data is first reported Cross-checks were carried out or

all recording units which commenced reporting in the period July 2002 to December 2003 In

more than 80 o cases the externally-reported commissioning date is within one month o the

first month or which data is reported he exceptions ndash Bu Farm and Mablethorpe ndash appear to

be due to delays by small operators in reporting data to Ogem

In a number o cases it is clear rom the data that plants were not in ull operation at the

time when data was first reported to the database here are two indicators o partial or delayed

commissioning

bull For an initial series o months the calculated load actor is very low ndash usually below 5 ndash

and then increases abruptly In such cases the data or the early months is dropped and the

commissioning date is treated as the month immediately preceding the first month with a

lsquonormalrsquo load actor14

12 httpswwwrenewablesandchpogemgovuk

13 See httpwwwthewindpowernetwindarms_list_enphp

14 he month in which a plant is reported as having been commissioned is not included in the analysis

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2652

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 25

the decline in the plantrsquos load actor as it ages As an example consider a plant whose reported

capacity was increased rom 99 MW to 10 MW at age 3 I this was simply a change in report-

ing practice this would reduce the estimated load actor rom age 3 onwards By increasing

the initial capacity rom 99 MW to 10 MW ndash or by reducing the capacity rom 10 MW to 99

MW rom age 3 onwards ndash this bias is removed It is not important which adjustment is madebecause the difference between the alternative assumptions is captured by the unit fixed effect

he numbers o separate sites and o observations per site mean that it is not possible to

provide a simple visual representation o the data Figure 8 provides a summary by illustrating

the distributions o load actors by the age o the wind arm pooled across all years he ranges

o the load actors or each age are wide while the medians show no clear trend by age hus a

naiumlve analysis might conclude that there is nothing to investigate However as will be explained

below the distributions mask crucial differences between the perormance o wind arms as

they age because they do not control or the differences in location and wind availability over

time

B Data for Denmark

he data or Danish wind arms used in this study comes rom a database compiled by the

Danish Energy Agency covering the characteristics and perormance o all wind turbines rom

2002 up to the end o August 201216 he basic recording unit or the register is the individual

wind turbine but where a number o turbines are connected to a single meter an identical

average output per turbine is recorded or all turbines in the group he presence o what are in

effect repeated observations increases the noise in the data In addition the model that is used

or the analysis assumes that site operator and turbine characteristics are common or all o theunits that make up a single wind arm o address this point separate wind arms have been

identified by grouping all turbines which have identical values or turbine manuacturer and

type local authority date o commissioning and date o decommissioning (where applicable)

he register contains inormation on 7569 turbines commissioned rom 1977 onwards his

includes a large number o small turbines which would not all within the scope o the ROC

scheme in the UK Hence or comparability with the UK data all turbines commissioned

prior to 1992 were dropped as well as all wind arms with a generating capacity o less than 1

MW he resulting sample covered 823 onshore wind arms and 30 offshore wind arms with

monthly load actors or various periods rom January 2002 to August 2012 Given the cut-off

or wind arms commissioned prior to 1992 the maximum age or an onshore wind arm was 20

years and the maximum age or an offshore wind arm was 17 years

he total capacity o the onshore wind arms recorded in the Danish database was 2757 MW

he average size o an onshore wind arm in Denmark is small ndash only 33 MW as compared

with 161 MW or the UK ndash because 84 o the sample consists o installations with between

1 and 4 turbines he largest onshore installation has 39 turbines with a capacity o 234 MW

while the onshore installation with the greatest capacity has 19 turbines with a capacity o 584

MW Overall the sample o Danish onshore wind arms is older and smaller than the sample

o UK onshore wind arms reflecting the much longer history o wind generation in Denmark

16 httpwwwensdkEN-USINFOFACSANDFIGURESENERGY_SAISICS_AND_INDICAORS

OVERVIEWOFHEENERGYSECORREGISEROFWINDURBINESSiderForsideaspx

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2852

26 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

he larger scale o UK wind arms may account or some o the differences in perormance

curves or the two countries

he sample o offshore wind arms in Denmark is quite small with 30 installations in

total Still it is useul to analyse the perormance o these installations since there is very little

comparable data or the UK he sample covers 394 turbines with an average o just over 13turbines per installation and an average capacity o 288 MW he largest installation which

was commissioned in 2003 has 72 turbines with a total capacity o 1656 MW Again offshore

installations are smaller and older in Denmark than is the current pattern in the UK he first

two wind arms in the sample were commissioned in the years 1995 and 2000 but the sample

increases significantly rom 2002 onwards o avoid reliance upon a sample o 1 or 2 wind

arms to determine specific coefficients or ages gt 10 years all observations with an age greater

than 10 were omitted rom the analysis17

A somewhat different problem with the offshore wind arm data arose because the coeffi-

cients estimated using capacity weights (see below) were heavily influenced by the largest windarm his has had an extremely erratic profile o perormance over time with a very low average

load actor he treatment o this wind arm is discussed in more detail in Section F below

C Specication and estimation methods

As Figure 8 illustrates it is necessary to go beyond simple summary statistics in order to assess

whether there is any systematic relationship between age and the average perormance o wind

arms he perormance o wind arms varies over both time and space some months and years

have more or less wind than the long term average while specific characteristics o wind arms

ndash including location type o turbine and operating regime ndash will influence the perormance oeach plant under identical wind conditions

he combination o (a) site-specific characteristics which are constant over time and (b)

period-specific characteristics which are constant across wind arms can be represented by

what statisticians call an error components model with fixed effects or each wind arm and

each time period he dependent variable in the model consists o a sequence o load actors ndash

denoted by or unit i in month t ndash or each wind arm (the panel unit) over time he load

actors are calculated by dividing the total output in MWh by (24 x number o days in month x

reported generating capacity in MW) and are multiplied by 100 to convert to percentages

he additive version o the error components model may be written as

(1)

in which denotes the age o plant i in period t and is some unction that is either a

representation o or an approximation to the normalised age-perormance relationship he

multiplicative version is similar except that is replaced by where ln() denotes the

natural logarithm he terms are wind arm or unit fixed effects which capture actors such

as location type o turbine operating regimes etc he terms are period fixed effects which

capture the influence o wind conditions over the UK or Denmark as a whole as well as seasonal

maintenance and demand Without loss o generality it may be assumed that the fixed effects are

17 For the avoidance o doubt this means that no attempt was made to estimate the effect o age on the perormance o

offshore wind arms beyond age 10 while the data or the oldest wind arms was included in the analysis up to and

including age 10

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 27

normalised so that ie all o the error components have a zero mean while the

mean o the random error is zero by definition he terms are random errors which capture

random variations in wind conditions that are not specific to the site turbine breakdowns and

other actors that are uncorrelated with either the site or the time period It is assumed that the

mean o the random error is zero while additional assumptions depend upon the method o

estimation that is adopted

he model specified in equation (1) can be estimated in a variety o ways18 A crucial issue is

whether the are or may be significantly correlated with the age terms or any other independent

regressors I it is assumed that there is no such correlation the model is usually reerred to as

a random effects model In that case the equation can be estimated by a variant o least squares

(LS) generalised least squares (GLS) generalised methods o moments (GMM) or maximum

likelihood (ML) ndash all o which should be yield parameter estimates that will converge in prob-

ability to the true parameter values as the sample size increases An alternative but less restric-

tive specification is known as the fixed effects model in which no assumption is made about

the correlation between the unit fixed effects and other regressors he limitation o the fixed

effects model is that it cannot be used to estimate parameters or regressors that vary across

panel units but are constant over time or a panel unit (time-invariant regressors) ndash eg the

year in which a wind arm was commissioned its location or its rated generating capacity he

influence o such variables is captured by the unit fixed effects

In order to avoid the risk o obtaining biased estimates o the age effects the results reported

in this note are based upon estimates using the fixed effects model he influence o time-invar-

iant regressors is examined in a second stage by regressing the estimates o the unit fixed effectson these variables he simplest method o estimating the fixed effects model is known as the

lsquowithinrsquo estimator in which the mean value or each unit is subtracted rom all observations or

that unit For example i is a general polynomial o order M in A equation (1) becomes

(2)

so that the model can be estimated by least square with the inclusion o dummy variables or

each period t he variance-covariance matrix o the coefficients have been estimated by using

(a) a robust sandwich estimator adjusted or clustering by panel unit and (b) a bootstrap esti-

mator with 400 repetitions he estimates o the standard errors will be consistent even i the

random error component is serially correlated within panel units andor has different variances

across panel units19 Since the number o panel units and the average number o time periods

per panel units are both large (other than or Danish offshore wind arms) the least square esti-

mates o the parameters will be unbiased under these assumptions

he bootstrap method is expensive to carry out but it provides a useul cross-check on

the standard errors generated by other methods Unortunately the bootstrap method o

18 See Chapter 21 o AC Cameron amp PK rivedi ndash Microeconometrics Methods and Applications (Cambridge

Cambridge University Press 2005) or a standard textbook treatment o estimation methods or panel data models

19 he estimation was carried out using the xtreg procedure in Stata Version 12 he methods o constructing the

sandwich and bootstrap estimates o the variance-covariance matrix are described in Chapters 8 amp 13 o AC

Cameron amp PK rivedi ndash Microeconometrics using Stata (Revised Edition College Station exas Stata Press 2010)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 5: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 552

Contents

List o Figures and ables 4

About the author 5

About the Renewable Energy Foundation 5

Authorrsquos Note 6

Note on the Data 6

Executive Summary 7

Te Perormance o Wind Farms in the United Kingdom and Denmark 9

Introduction 9

Age-perormance curves 10

Differences between wind arms 12Implications or uture policy and perormance 15

Conclusion 21

Appendix Data and Methods 23

A Data or the United Kingdom 23

B Data or Denmark 25

C Specification and estimation methods 26

D Period fixed effects vs normalisation by wind speeds 28

E Estimation results or the UK 30F Estimation results or Denmark 36

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 652

List of Figures and TablesList of Figures and Tables

Figure 1 Perormance degradation due to age using equal weights 11

Figure 2 Perormance degradation due to age using capacity weights 12

Figure 3 Range o unit fixed effects by category 13

Figure 4 Initial load actors by year o commissioning or onshore wind 14

Figure 5 Impact o perormance degradation on discounted cumulative output 16

Figure 6A Projected load actors or UK onshore wind (total capacity o 10 GW in 2020) 18

Figure 6B Projected load actors or UK onshore wind (total capacity o 15 GW in 2020) 18

Figure 7 Projected load actors or UK offshore wind in 2020 20

Figure 8 Box plots o load actors by age or UK onshore wind arms 24

Figure 9A Additive age-perormance curves or UK onshore wind arms 32

Figure 9B Multiplicative age-perormance curves or UK onshore wind arms 32

Figure 10 UK onshore age-perormance curves using equal and capacity weights 33

Figure 11A Residuals by age or perormance curves using equal weights 34

Figure 11B Residuals by age or perormance curves using capacity weights 34

Figure 12 Unit fixed effects by year o commissioning or England and Scotland 35

Figure 13A Age-perormance curves or Danish onshore wind arms 37

Figure 13B Age-perormance curves or Danish offshore wind arms 37

Figure 14 Danish age-perormance curves using equal and capacity weights 38

able 1 Average load actors by year and country () 40

able 2 Estimation results or UK onshore wind arms 41

able 3 Equations or trends in unit fixed effects 43

able 4 Estimation results or Danish wind arms 44

able 5 Perormance by year o commissioning or Danish wind arms 46

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 752

About the author

Dr Gordon Hughes is a Proessor o Economics at the University o Edinburgh where he teaches courses in

the Economics o Natural Resources and Public Economics He was a senior adviser on energy and environ-

mental policy at the World Bank until 2001 He has advised governments on the design and implementationo environmental policies and was responsible or some o the World Bankrsquos most important environmental

guidelines

About the Renewable Energy Foundation

he Renewable Energy Foundation is a registered charity promoting sustainable development or the benefit

o the public by means o energy conservation and the use o renewable energy

REF is supported by private donation and has no political affiliation or corporate membership In pursuit o

its principal goals REF highlights the need or an overall energy policy that is balanced ecologically sensi-

tive and effective

We aim to raise public awareness o the issues and encourage inormed debate regarding a structured energy

policy that is both ecologically sensitive and practical he issues o climate change and security o energy

supply are complex and closely intertwined REF contributes to the debate surrounding these issues by

commissioning reports to provide an independent and authoritative source o inormation

For urther inormation see wwwreorguk

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 852

Authorrsquos Note

Note he views expressed in this paper are strictly personal and do not reflect the position o any organ-

isation with which I am associated I am grateul to John Constable Lee Moroney and a reeree or their

comments on earlier drafs o the paper he results o this study have been presented at seminars andlectures to various groups in Edinburgh and Glasgow I have benefited rom the questions and comments

rom participants in those events

Gordon Hughes

School o Economics

University o Edinburgh

gahughesedacuk

Note on the Data

For the convenience o interested researchers both the raw and cleaned data employed in this analysis will

published alongside the electronic version o the study on the REF website wwwreorguk

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 7

Executive Summary

1 Onshore wind turbines represent a relatively mature technology which ought to have achieved

a satisactory level o reliability in operation as plants age Unortunately detailed analysis o

the relationship between age and perormance gives a rather different picture or both theUnited Kingdom and Denmark with a significant decline in the average load actor o onshore

wind arms adjusted or wind availability as they get older An even more dramatic decline is

observed or offshore wind arms in Denmark but this may be a reflection o the immaturity

o the technology

2 he study has used data on the monthly output o wind arms in the UK and Denmark reported

under regulatory arrangements and schemes or subsidising renewable energy Normalised

age-perormance curves have been estimated using standard statistical techniques which allow

or differences between sites and over time in wind resources and other actors

3 he normalised load actor or UK onshore wind arms declines rom a peak o about 24 at

age 1 to 15 at age 10 and 11 at age 15 he decline in the normalised load actor or Danish

onshore wind arms is slower but still significant with a all rom a peak o 22 to 18 at age 15

On the other hand or offshore wind arms in Denmark the normalised load actor alls rom

39 at age 0 to 15 at age 10 he reasons or the observed declines in normalised load actors

cannot be ully assessed using the data available but outages due to mechanical breakdowns

appear to be a contributory actor

4 Analysis o site-specific perormance reveals that the average normalised load actor o new UK

onshore wind arms at age 1 (the peak year o operation) declined significantly rom 2000 to2011 In addition larger wind arms have systematically worse perormance than smaller wind

arms Adjusted or age and wind availability the overall perormance o wind arms in the UK

has deteriorated markedly since the beginning o the century

5 hese findings have important implications or policy towards wind generation in the UK First

they suggest that the subsidy regime is extremely generous i investment in new wind arms

is profitable despite the decline in perormance due to age and over time Second meeting

the UK Governmentrsquos targets or wind generation will require a much higher level o wind

capacity ndash and thus capital investment ndash than current projections imply hird the structure

o contracts offered to wind generators under the proposed reorm o the electricity market

should be modified since ew wind arms will operate or more than 12ndash15 years

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 9

The Performance of Wind Farms in the United Kingdom

and Denmark

Introduction

1 Any assessment o the costs o wind power must rely heavily upon assumptions about theaverage load actor that will be achieved by new wind installations over their lietime It is

standard practice to calculate average load actors by year and country or onshore and offshore

installations as shown in able 1 (page 40)1 However such estimates do not provide a

reliable statistical basis or assessing the uture perormance o wind arms in aggregate Part

o the reason is that the amount o wind in any month or year is influenced by long term mete-

orological cycles that have periods o many years notably the North Atlantic Oscillation In

addition average load actors do not allow or changes in the composition o wind installations

by location age size and other actors

2 As the number o wind arms operating in developed countries has grown engineers have an

opportunity to analyse operating experience over extended periods o time With any (relatively)

novel technology the incidence o temporary or permanent breakdowns may be expected to all

over time so it is difficult to disentangle the effects o age on operating perormance rom the

technological immaturity o turbines installed 10 or 20 years ago Nonetheless the technology

or onshore wind turbines has been reasonably mature since the early 2000s his is documented

by data produced by the US Department o Energy which shows that the decline in capital costs

that is characteristic o immature technologies slowed afer 2000 and was reversed afer 20042

3 With at least 10 years o operating data since 2000 it should be possible to examine how thetypical operating perormance o wind installations in the United Kingdom and Denmark

varies with the age o the turbines he estimated age-perormance curves should then be

incorporated in estimates o levelised costs which are ofen used to compare the costs o wind

and other orms o generation

4 he Appendix provides a detailed description o the data and methods employed or this study

In both the UK and Denmark wind operators have a large incentive to produce electricity

whenever there is sufficient wind available since marginal operating costs are small while the

operator receives a much higher price In the UK this price is the sum o the market price plus

the value o (a) the Renewable Obligation Certificates (ROCs) awarded or each MWh o elec-

tricity produced and (b) the associated exemption rom the Climate Change Levy In practice

the effective market value o ROCs and associated incentives means that an onshore wind

operator earns roughly double the average market price o electricity per MWh In Denmark

most onshore wind capacity receives a price premium equivalent to about pound28 per MWh on top

1 he load actor is calculated as the ratio o the amount o electricity actually produced by a turbine or wind arm

over a period o a month or a year divided by the amount o output that would have been produced had it operated

at ull nameplate capacity or the entire period his is expressed as a percentage so that reported load actors lie

between 0 and 100

2 R Wiser amp M Bolinger (2012) 2011 Wind Technologies Market Report Lawrence Berkeley National Laboratory

US Department o Energy It should be noted that the installed cost o wind plants has allen since 2010 but these

changes reflect the cyclical orces o demand and supply that are well established across the electricity sector and

which apply to ossil-uel as well as renewable generators

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1252

10 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

o the market price or about 125 years o operation at a typical load actor3 Offshore capacity

is eligible or a similar price premium but most o it is contracted via tenders that offer a guar-

anteed eed-in tariff he tender tariffs have varied rom pound56 per MWh to twice that figure

extending over 16-18 years o operation at a typical load actor

Age-performance curves

5 A standard indicator o the operating perormance o a wind turbine is its normalised load

actor defined as the total output over some period divided by the maximum potential output

adjusted or wind availability Normalisation or wind availability is not straightorward I

detailed data on wind speeds at specific locations is available then the potential output rom

individual turbines or wind arms can be calculated using standard power curves Since wind

speeds can vary substantially over small distances such calculations are both difficult to imple-

ment and prone to errors when applied to large numbers o wind generators using public

sources o data4

6 Many analyses work in the opposite direction For example the Danish Wind Index ndash the

longest running series o data on wind availability in Europe ndash is constructed rom the common

component o monthly variations in wind output rom a large sample o wind turbines5 his

study adopts a similar approach Wind availability is treated as an unobserved actor whose

contribution to monthly variations in wind output is estimated by a series o fixed period

(monthyear) effects in a statistical model o monthly output or all wind arms he reasons

or adopting this approach are discussed in Section D o the Appendix his explains why the

method must perorm better than the easible alternative o using a single index o average

wind speed or the UK or Denmark

7 he analysis relies upon standard techniques that are widely used in the biological and medical

sciences as well economics and engineering Full details are given in the Appendix he idea is

to treat the actual load actor (ie without adjustment or wind conditions) or a specific wind

generator in any time period as being determined by a set o components which reflect specific

site conditions and the age-related perormance o wind plants in general together with the

period fixed effects and an uncorrelated random error he onshore wind datasets or the UK

and Denmark used or the analysis are large with monthly observations on 282 installations in

the UK and 823 installations in Denmark with an age range rom 0 to 19 years he offshore

wind dataset or Denmark is rather smaller covering only 30 installations but it can be used to

obtain reasonable estimates o perormance up to age 10 he results discussed here are based

upon least squares estimation (see page 27) combined with standard errors that are robust to

various departures rom classical assumptions about the nature o the data generation process

8 he results o the statistical analysis demonstrate an unambiguous and statistically significant

decline in the operating perormance o wind arms as they grow older Figure 1 shows the

3 Conversions to GBP are based on an exchange rate o pound1 = DKK 918 (as at 19 October 2012)

4 All wind generators collect their own inormation on wind speeds in order to manage their plants but they do not

publish such inormation and will not have access to inormation collected by other operators

5 Details o the construction o the Danish Wind Index are given at wwwvindstatdk A paper by Boccard ndash N Boccard

(2009) lsquoCapacity actor o wind power realized values vs estimatesrsquo Energy Policy Vol 37 pp 2679-88 ndash includes a

discussion o the construction o similar indices o wind availability in Germany the UK Netherlands and Sweden

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 11

simplest variant o the normalised age-perormance curve or onshore UK wind installations

together with the equivalent curves or onshore and offshore Danish installations he curves

illustrated are calculated rom the multiplicative error components model described in the

Appendix with log(load actor) as the dependent variable and a quadratic in plant age to repre-

sent the variation in plant perormance with age9 he rate o decline in perormance is greatest or offshore installations in Denmark with a all

rom load actors o over 40 at ages 0 and 1 to less than 15 by at ages 9 and 10 Onshore

installations in the UK show a more rapid rate o decline ndash 09 percentage points per year over

the first 10 years o operation ndash than is the case or Denmark though the normalised Danish

perormance curve lies below the UK curve or the first our years For the UK the normalised

load actor alls to just over 15 at age 10 and to 11 at age 15 With such low load actors it

seems likely that many wind arms will be re-powered ndash ie the turbines will be replaced ndash once

they reach the age o 10 or at most 15

Figure 1 Performance degradation due to age using equal weights

0

5

10

15

20

25

30

35

40

45

50

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

L o

a d f a c t o r ( )

Age (years)

UK onshore DK onshore DK offshore

Note Normalised load actors in Source Authorrsquos estimates

10 here are two plausible explanations or the observed decline in average load actors as wind

arms age he first is that the turbines become less efficient over time as a result o mechanical

wear and tear erosion o the turbine blades and related actors he second is that the turbines

experience more requent breakdowns and their operators take more time to bring them back

into service because they are less concerned about the perormance o older plants Both

reasons may be relevant in different circumstances and it is not possible to identiy a primary

explanation rom the data he requency o extended shutdowns does seem to increase with

age but this could be a reflection o the timing o planned maintenance operations rather than

breakdowns Whatever the cause the reduction in perormance with age is much greater than

would be expected or thermal generating plants

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1452

12 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

11 he age-perormance curves in Figure 1 are derived rom equations estimated by giving equal

weight to each wind arm irrespective o its generating capacity Figure 2 illustrates similar

curves but in this case the estimates are constructed by weighting each wind arm by its generat-

ing capacity reerred to as capacity-weighted his gives a better representation o perormance

degradation per MW o generating capacity he striking result is that the rate o decline inthe perormance o UK onshore wind installations is significantly aster when capacity weights

are used which implies that large wind arms in the UK experience a more rapid decline than

smaller ones he differences are smaller in Denmark but the capacity-weighted decline in

perormance is more rapid than the equal weights decline or Danish onshore installations

Figure 2 Performance degradation due to age using capacity weights

0

5

10

15

20

25

30

35

40

45

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

L o a d f a c t o r ( )

Age (years)

UK onshore DK onshore DK offshore

Note Normalised load actors in Source Authorrsquos estimates

Differences between wind farms

12 he specification o the statistical model means that the unit fixed effects identiy the specific

perormance characteristics o individual wind arms afer adjusting or wind availability and

age hese characteristics may reflect the site andor design o the installation as well as the

way in which it is operated A positive value or the unit fixed effect is equivalent to shifing

the perormance curve up by some percentage applied uniormly at each age while a negative

value shifs it down It ollows that the distribution o unit fixed effects across installations can

be used to compare the relative effectiveness o new installations by location date o commis-

sioning etc Care is required when making comparisons between or example UK and Danish

onshore wind installations because these have been normalised separately so the unit fixed

effects reer to different underlying averages6

6 Some care is required in interpreting the unit fixed effects By deault the average value o the unit fixed effects or

each group is zero but these averages are taken over all observations or each installation in the basic sample For the

purpose o the comparisons in Figure 3 the unit fixed effects have been adjusted so that the average values or each

group are zero when using one value or each installation

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 13

13 o give a sense o the numbers unit fixed effects o (a) 02 or (b) -02 mean that the base load

actor given the age o the plant is multiplied by (a) exp(02) = 122 or (b) exp(-02) = 082 Since

the base load actor or a plant o age 0 is 231 these fixed effects translate to load actors in

the first year o operation o (a) 282 and (b) 189 For a wind arm with 50 MW o gener-

ating capacity the difference amounts to an additional 40700 MWh o electricity output peryear Using the standard figures cited by RenewableUK this would be sufficient to supply 12300

homes or a year his range (040) is slightly greater than the interquartile range o unit fixed

effects or UK onshore wind arms (035) Again or comparison operators o thermal power

plants would regard any difference o even 10 up or down in efficiency relative to what is

expected as a striking and perhaps worrying indication o either good or bad perormance

Figure 3 Range of unit fixed effects by category

- 1

- 5

0

5

1

A d j f i x e d e f f e c t ndash q u a d r a t i c

UK onshore DK onshore DK offshore

Source Authorrsquos estimates

14 A notable eature o the distributions o the unit fixed effects that are summarised in the box

plots shown in Figure 3 is the range o variation between the best and worst perormers espe-

cially or UK onshore and Danish offshore wind arms Box plots are constructed so that the

top and bottom o the box correspond to the upper and lower quartiles while the bars at topand bottom correspond to the maximum and minimum values excluding outliers7 A plant at

the upper quartile o the distribution o UK onshore wind arms will generate an annual output

that is about 40 higher than a plant at the lower quartile he equivalent figure or Danish

onshore wind plants is an increase o 30 while or Danish offshore wind arms the difference

between the upper and lower quartiles is equivalent to 25 times the output o a lower quartile

plant By any standard such differences are important

7 he upper quartile is the point in a distribution such that the 25 o values exceed it while the lower quartile is the

point in a distribution such that 75 o values exceed it he line in the middle o the box marks the median the

point exceeded by 50 o values Outliers are defined as values that lie outside the range o ukey adjacent values

which are defined as the upperlower quartiles +- 15 x interquartile range

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1652

14 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

15 Outside the range between the upper and lower quartiles the top and bottom segments o

the distributions extend urther up and down or UK onshore wind arms than or those in

Denmark he rather different pattern or Danish offshore wind installations is partly a conse-

quence o the relatively small size o the sample but the evidence points to considerably greater

variability in the perormance o offshore wind installations than or onshore installations16 It is somewhat surprising to observe the magnitude o the differences in the perormance in

onshore wind arms in the UK Given the nature o the subsidy regimes and the high capital

cost o developing new installations it might be expected that operators have a strong incen-

tive to identiy the best locations and then choose equipment that will deliver the maximum

amount o electricity output at a high level o reliability For UK onshore operators it seems that

good or bad perormance is somewhat o a lottery However i the subsidies provided by ROCs

are sufficient to underwrite investment in inefficient plants ndash as appears to be the case ndash then

those subsidies are extremely generous or plants that operate close to the efficient rontier

As location is likely to be the main actor that determines the perormance o a specific plant

relative to all other plants the inerence must be that many wind plants have been developed on

sites with poor wind characteristics

Figure 4 Initial load factors by year of commissioning for onshore wind

10

15

20

25

30

35

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

I n i t i a l l o a d f a c t o r ( )

Year commissioned

UK onshore DK onshore Linear (UK onshore) Linear (DK onshore)

Source Authorrsquos estimates

17 here is another disturbing characteristic o UK wind arms which is revealed by examination

o the unit fixed effects and illustrated in Figure 4 It might be expected that the average peror-

mance o recently installed units would be higher than that o older units his should reflect

improvements in turbine design and reliability as well as in the identification o good sites or

wind generation However that is not the pattern or onshore wind arms in the UK he figure

shows the average values o the initial load actors calculated using equally weighted data by

year o commissioning or onshore wind arms in the UK and Denmark together with fitted

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 15

trend lines8 here has been a clear decline in the initial load actors in the UK which has not

occurred in Denmark his pattern is confirmed by more detailed statistical results shown in

the Appendix which take account o location and size o installation

18 he decline in initial perormance is particularly obvious afer 2005 when the rate o commis-

sioning new capacity increased sharply rom less than 500 MW in the 5 years 2000ndash04 to morethan 500 MW per year rom 2005 onwards In contrast the peak years or commissioning new

onshore wind arms in Denmark were 2000ndash02 when 920 MW was installed he rate o new

building (or replacement) picked up again in 2008 but it is still much lower than at the turn o

century Not only is the perormance o the onshore wind plants in the UK commissioned in

recent years significantly worse than that or wind arms commissioned beore 2005 but it is

possible that this is a direct consequence o an overly rapid expansion o capacity In essence

the evidence suggests that the industry does not have the capability to identiy develop and

operate new onshore wind arms at the rate envisaged by UK government targets while main-

taining a satisactory average level o perormance

Implications for future policy and performance

19 he pattern o perormance degradation with age and over time identified in this analysis has

implications or the assessment o new investment in wind arms and or the design o the

proposed Electricity Market Reorm (EMR) here are two related issues that are affected by

the results

a) What are the implications or the design o policies intended to promote the adoption

and expansion o wind generation in the UK his includes the structure o current

subsidies as well as the specification o the contracts that are proposed under the EMR A

related question concerns the expected lie o new wind arms

b) How might the average perormance o wind generation develop over the next decade

his is absolutely critical to any assessment o the amount o wind capacity that will be

required to meet the UK governmentrsquos targets or the share o electricity ndash and more

generally energy ndash rom renewable sources he government has assumed that the

average load actors or both onshore and offshore wind arms will either remain stable

or increase in uture he prospects will be very different i these assumptions are not

borne out

20 he first question can be addressed by considering the discounted sum o cumulative net

output rom a new wind arm on the assumption that its perormance matches the averages

identified in the analysis his takes account o both perormance degradation over time plus

the annual cost o maintaining the plant which may be expressed as a deduction rom gross

production so as to calculate the contribution o expected output at age s to recovering the

original capital cost o building the plant hese figures have to be discounted back to the date

o commissioning in order to assess the net present value o the initial investment in the plant

he curves or constant and observed perormance using a discount rate o 9 (a typical cost o

capital or wind generation) are shown in Figure 5 he black line is based on the ull set o age

8 he initial load actors are calculated as the normalised load actor or UKDanish wind arms at age 0 multiplied by

where is the unit fixed effect or installation i

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1852

16 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

effects while the red dashed line is based on the smoothed quadratic perormance curve and

the blue dashed line shows what the pattern would be i operating perormance was not affected

by age he calculation is extended out to age 25 since DECCrsquos standard comparisons assume

that wind turbines have a lie o 25 years

Figure 5 Impact of performance degradation on discounted cumulative output

0

10

20

30

40

50

60

70

80

90

100

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

o

f d i s c o u n t e d c u m u l a t i v e o u t p u t

Age of plant (years)

Constant Quadratic Full age effects

Source Authorrsquos estimates

21 he key points in the figure are that 80 o the discounted cumulative output o a new wind

plant is likely to be produced in the first 10 years o its lie and 90 in the first 14 years his

is consistent with the structure o Danish subsidies or onshore wind arms which extend over

a typical period o 125 years Since sites or wind arms are scarce and involve the payment o

significant rents that may be linked to output it is very unlikely that any new installation o

wind turbines will have an expected lie o more than 15 years Instead as has happened in the

past wind operators will have a strong incentive to decommission plants afer no more than 15

years and replace the turbines with newer equipment

22 As a consequence any economic assessment o wind generation should not be based on an

expected lie which is longer than 15 years In recent work reported in evidence to the House

o Commons Select Committee on Energy and Climate Change I assumed that wind plants

would have a residual value equal to 20 o their initial cost in real terms at the end o 15 years

he analysis in this paper suggests that this is too avourable an assumption Given the costs o

decommissioning old turbines the residual value is likely to be well below 10 o their initial

cost and the decision point may be at 10 rather than 15 years

23 he corollary o this observation is that it makes little sense to offer long term contracts or 20

years or more that guarantee prices or eed-in tariffs (Fis) to wind operators A contract length

o 10 to 12 years would be sufficient to remove most o the market risk associated with invest-

ment in wind generation In this respect the subsidy arrangements implemented in Denmark

are better designed

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 17

24 At the same time the offer o subsidies andor guaranteed prices may have serious adverse

consequences or the efficiency o wind generation Returning to the unit fixed effects measur-

ing the perormance o wind plants commissioned afer 2005 only 28 out o 159 units have an

operating perormance that exceeds the average or the period hose 28 units account or 360

MW o capacity whereas the remaining 131 units account or 2810 MW o capacity Not onlyare recent plants less efficient than the average or the whole period but the plants which are

below-average in efficiency are typically larger than the more efficient ones and account or a

disproportionate share o the recent additions to generating capacity

25 In mid-2012 there was approximately 4500 MW o onshore wind capacity operating in the

UK By 2020 it is expected that there will be at least 10000 MW (based on proposed wind

arms with planning consent) and perhaps as much 15000 MW o wind capacity in operation

he government is relying upon a substantial contribution rom onshore wind arms towards

meeting its targets or renewable energy in 2020 o provide context the governmentrsquos targets

or renewable energy in 2020 speciy a total o 234 Wh o energy rom all renewable sources

o which 90ndash140 Wh is expected to come rom wind and biomass electricity generation9

he projections or biomass electricity are extremely vague and have to be reconciled with

an assumed increase o 3ndash4 times in the amount o biomass used separately or heat So in

concrete terms it is reasonable to assume that wind generation will account or at least 90 Wh

with 24ndash32 Wh rom onshore generation and the remainder rom offshore

26 he average load actor or onshore wind over the 10 year period rom April 2002 to March

2012 was 256 which is influenced by the large amounts o new capacity added rom 2005

onwards I this load actor were achieved in 2020 the total amount o onshore wind capacityrequired to meet the onshore generation target would be 107 to 143 GW he average load

actor or offshore wind installations (on an unchanged configuration basis) or the five year

period 2007ndash11 was 32010 I this load actor were achieved in 2020 the total amount o

offshore wind capacity required to meet the remainder o the 90 Wh target would be 206

to 235 GW though these estimates would all to 180 to 205 GW i the average load actor

matched the historic average value o 367 or Danish offshore wind installations

27 he question that has to be addressed is whether these load actors are consistent with (a) the

investment programmes that would be required to achieve the total amounts o generation

capacity implied by the governmentrsquos targets and (b) the age profiles o the perormance o

wind arms discussed in this paper he short answer is that this is simply not possible i the

initial load actor or new wind arms continues to decline at the rate observed the past decade

So even as a starting point it is necessary to assume that this underlying deterioration in wind

arm perormance ceases his would be a major change in itsel and it is not obvious what

actors might lead to such a result

28 he results illustrated below are based on a vintage model o perormance that takes account

o the actual expansion o capacity up to the end o 2011 plus whatever investment in new or

replacement turbines is required to achieve the target levels o wind capacity in 2020 o high-light the ull range o potential outcomes these targets are either 10 GW or 15 GW o onshore

9 Department o Energy and Climate Change ndash UK Renewable Energy Road Map July 2011 Figure 2

10 he equivalent average load actor or onshore wind arms was 264

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2052

18 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

wind and either 18 or 24 GW o offshore wind he levels o investment required to meet these

targets depend upon the expected lie o wind turbines For onshore turbines an expected lie

o 10 years is consistent with re-powering decisions over the last decade while an expected lie

o 15 years seems to be the maximum consistent with the age-perormance profile For offshore

turbines the analysis uses expected lives o 15 and 20 years

Figure 6A Projected load factors for UK onshore wind (total capacity of 10 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

Figure 6B Projected load factors for UK onshore wind (total capacity of 15 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

29 Figure 6A shows projected values o the average normalised load actors or UK onshore wind

arms in 2015 and 2020 using a target capacity in 2020 equal to 10 GW plus the age-peror-

mance profiles or the UK and Denmark (DK) Since vintage effects are important in this

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 19

calculation the profile o new or replacement investment affects the results In this case the

level o gross investment is assumed to be constant rom 2013 onwards at either 930 or 1220

MW per year ndash the range is defined by the assumed lie o turbines Achieving a target o 15

GW in 2020 (Figure 6B) would require a level o investment rom 2013 onwards o either 1540

or 1930 MW per year Such figures are 3ndash4 times the average level o investment over theperiod 2005ndash11 so the lower target may provide a better indication o what will happen

30 All o the estimates are calibrated to match the observed average load actor in 2007ndash11 so that

they should be interpreted as relative rather than absolute values Using the age-perormance

profile or UK onshore wind installations the average normalised load actor is likely to all by

about 19 rom 2007ndash11 to 2015 and by 23ndash28 to 2020 I the age-perormance curves or

Danish onshore installations were to apply the reduction in the average normalised load actor

would be considerably less but there would still be a all o up to 5 by 2020

31 In the medium term the all is greater i turbines operate or 15 years because the average age

o all turbines will be higher and thus the degradation in perormance due to age will have a

larger effect By 2020 this effect reduces the average normalised load actor rom 203 or a 10

year lie to 195 or a 15 year lie here is a complex economic and commercial trade-off here

o meet a target o 10 GW in 2020 it is necessary to increase the level o investment in onshore

capacity by 30 rom 2013 Almost all o this will take the orm o re-powering existing wind

arms11 he costs are airly high because this will usually involve new development consents

decommissioning the original turbines and replacing them with a smaller number o larger and

more powerul turbines

32 Estimates o re-powering costs are generally treated as being commercially sensitive but broadproject costs cited in specialist magazines suggest that the unit costs are likely to be at least

pound1000 per kW or more than pound300 million per year in aggregate On the other hand the increase

in the amount o energy generated would be about 12 Wh or a 10 year turbine lie rather

than a 15 year turbine lie though this does not allow or the loss o production during re-pow-

ering Under the best circumstances the payback period may be quite short but the best option

rom a broader economic and social perspective is not obvious

33 Even under the best scenario 10 GW o onshore wind capacity will only generate 178 Wh o

electricity in 2020 well below the lower end o the target range or onshore wind generation

specified in DECCrsquos road map In act allowing or interactions between the age-perormance

profile and investment the amount o onshore wind capacity required in 2020 to generate 24

Wh would be a minimum o 134 GW At the top end o DECCrsquos target range the capacity

required in 2020 to generate 32 Wh would be a minimum o 178 GW Since these estimates

are based on a 10 year turbine lie the level o investment required rom 2013 onwards would

be 1650 MW per year or the bottom o the range and 2350 MW or the top end o the range

In the last 8 years total investment in onshore wind has only ever exceeded 600 MW in one

11 Re-powering is usually defined as the ull replacement o existing turbines whereas retrofitting covers the replace-

ment o various mechanical or electrical components In the data analysis re-powering involves the creation o a

new unit whereas retrofitting affects the perormance o a continuing unit Hence the potential benefits o retrofit-

ting are taken into account in constructing the age-perormance curves

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2252

20 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

year (883 MW in 2008) so it seems rather unlikely that even the lower end o DECCrsquos range or

onshore wind generation will actually be realised

34 O course this assessment would be different i UK onshore wind operators were able to match

the Danish age-perormance curve here are many actors which differentiate the UK and

Danish wind industries he average scale o wind arms in the UK is larger they are moreremote and there appears to be significantly greater variation between operating and potential

new sites in achieved perormance It would thereore require a very optimistic observer to

conclude that the difference in age-perormance curves between the two countries will disap-

pear in the short or medium term

Figure 7 Projected load factors for UK offshore wind in 2020

20

22

24

26

28

30

32

34

18 GW ndash life 15 yrs 18 GW ndash life 20 yrs 24 GW ndash life 15 yrs 24 GW ndash life 20 yrs

L o a d f a c t o r ( )

2007ndash11 2015 2020

Note Based on Danish age-perormance curves Source Authorrsquos estimates

35 Figure 7 shows the results rom carrying out a similar exercise or offshore wind arms in the

UK he only evidence on the age-perormance curves or offshore installations is based upon

the Danish experience so these have been used to construct the projections Further the profile

o investment assumes a progressive increase in additions to capacity rom 2013 to 2016 and

then a stable level o investment rom 2017 to 2020 he DECC road map involves a very large

commitment to offshore wind investment ndash in the range 25-35 GW o new capacity annually

rom 2016 onwards As a consequence the average age o offshore capacity in 2020 will be quite

low and the ull effects o the age-perormance curves will only be elt during the decade ollow-

ing 2020 Even so the average normalised load actor in 2020 will be about 15 lower than or

2007ndash11 while by 2025 it will be about 30 below the base level

36 On these projections 232 GW o offshore capacity is the minimum necessary to generate 58

Wh o electricity rom offshore wind in 2020 his requirement will increase to 298 GW by

2025 simply to offset the effect o the ageing o offshore generation capacity on total output

37 Combining the estimates or onshore and offshore wind generation the total amount o wind

capacity required in 2020 to produce 90 Wh o electricity rom wind will be in the range 39

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 21

to 41 GW once allowance is made or the observed age-perormance curves or both onshore

and offshore installations hese estimates are roughly a third higher than official predictions o

28ndash31 GW o wind capacity in 2020

38 urning the figures round gives a slightly different picture o the reality o the governmentrsquos

strategy on renewable electricity generation A total o 30 GW o wind capacity in 2020 (12GW onshore 18 GW offshore) will generate about 64 Wh o electricity in a normal year

Since electricity generation rom wind and biomass are expected to account or 50 o renew-

able energy in 2020 the consequence is that biomass is expected to account or 53 Wh In

2011 landfill gas and sewage sludge digestion accounted or 44 o electricity rom bioenergy

(DUKES ables 64) here are limited opportunities to expand production rom these sources

so a continuation o historic rates o growth would mean that renewable generation rom other

sources o bioenergy would have to increase rom 72 Wh in 2011 to 46 Wh in 2020 here

are severe constraints on increasing the amounts o electricity generated rom animal biomass

biodegradable municipal solid waste and anaerobic digestion so it would be optimistic to

assume that these will contribute more than 6-8 Wh in 2020 (up rom 26 Wh in 2011)

39 In practice i the projections or renewable energy are to be believed they rest upon an assump-

tion that there will be a large increase in the amount o electricity that is generated rom plant

biomass ndash either on its own or through co-firing with coal his increase will involve increas-

ing the amount o electricity generated in these ways rom 46 Wh in 2011 to 38-40 Wh in

2020 Since the amounts o straw and UK-grown timber available or this purpose are strictly

limited this will come down to the amount o wood chips that can be imported and the cost o

doing this40 he historic average load actor or biomass plants is just over 60 so about 7500 MW o

biomass capacity will be required to supply 40 Wh o electricity he weight o wood chips per

MWh o electricity depends upon moisture content storage boiler design and other actors but

an indicative range is 085 to 115 tonnes per MWh o wood chips with a moisture content o

30 According to DECC and Forestry Commission estimates the UK imported 265 million

tonnes o steam coal and 26 million tonnes o other wood products (including wood chips)

in 2011 It seems that the net effect o the governmentrsquos strategy will be to replace 15 million

tonnes o coal imported in 2011 by 32 to 44 million tonnes o imported wood chips his is a

somewhat strange way o increasing energy security

Conclusion

41 Wind power is a highly capital-intensive technology or generating electricity Its merits rely

entirely upon a substitution o capital or uel inputs he same is true or hydro or tidal or wave

power In comparison with hydro power wind is a low quality resource because o its variability

and because it cannot be stored So the economic case or wind power must rest on obtaining

the most out o the wind that is available

42 While the decline in the achieved perormance o onshore wind turbines in Denmark is

much less than that or the UK or offshore nonetheless the decline in expected output under

standardised wind conditions over 10 years is 10 unweighted and 13 capacity weighted

hese declines accelerate afer age 10 so that the reductions in perormance are 17 and 20

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2452

22 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

respectively afer 15 years For UK onshore wind arms the reduction in perormance due to

age is much worse at 27 unweighted and 69 capacity weighted by age 10

43 Evidence on the perormance o Danish offshore installations is both restricted and so poor

that there may be concern that the results are affected by a small number o outliers Still the

sample contains a reasonable number o sites with at least 5 years o operating experience andthe decline in perormance by age 5 is 38 unweighted and 26 capacity weighted

44 In addition to these results there is strong evidence that the average normalised load actor or

new onshore wind installations in the UK has allen significantly over the period rom 2000 to

2011 his is consistent with a pattern in which the most avourable sites are developed first

Equally it could mean that wind developers have been unable to keep up with the rate o new

investment while maintaining the quality o development and operations For example the site

design or selection o turbine characteristics may make less effective use o the available wind

resources or the sites available than was the case in the past

45 Whatever the reasons the deterioration in initial perormance means that the expected returns

rom the expansion in wind capacity both or investors and in terms o the reduction in CO983090

emissions have been alling without a concomitant decrease in the private and social costs

that are borne by customers and the general public Clearly this is unsatisactory at best and it

suggests that the benefits claimed or current policies cannot be taken at ace value

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 23

Appendix Data and Methods

A Data for the United Kingdom

he raw data used or this study was extracted by the Renewable Energy Foundation rom the

Renewables and CHP Register database compiled by Ogem12 his inormation is used in the

administration o the market in Renewable Obligation Certificates (ROCs) he Renewables

Obligation is the primary mechanism by which large scale generators o renewable electric-

ity receive subsidies so the operators o wind arms have a very strong incentive to submit

complete and up-to-date inormation to the database he data extracted covered all onshore

wind generators with at least 05 MW o generating capacity which are eligible or ROCs or the

10 year period April 2002 to March 2012 he key variables are the monthly output o electric-

ity (used or the purpose o allocating ROCs) and nominal generating capacity here are 282

reporting units in the final dataset A reporting unit may correspond to an entire wind arm orto separate phases o development Complications arise when an existing wind arm is re-pow-

ered ie when old wind turbines are replaced by newer and usually more powerul turbines

he database reports this as a change in the generating capacity o the recording unit but in

this analysis repowering is treated as the termination o the old recording unit and the creation

o a new recording unit

A considerable amount o data checking and cleaning was required beore the data could

be analysed he first step was to add the age o each reporting unit his was calculated by

reerence to the month in which the turbines were commissioned For wind arms in operation

in or beore June 2002 the commissioning date was obtained rom various online sources ndash

notably the thewindpowernet database o UK wind arms13 which was cross-checked against

the inormation provided on the websites o wind operators and other sources For recording

units whose first data relates to dates afer June 2002 the commissioning date was initially

assumed to be the month or which data is first reported Cross-checks were carried out or

all recording units which commenced reporting in the period July 2002 to December 2003 In

more than 80 o cases the externally-reported commissioning date is within one month o the

first month or which data is reported he exceptions ndash Bu Farm and Mablethorpe ndash appear to

be due to delays by small operators in reporting data to Ogem

In a number o cases it is clear rom the data that plants were not in ull operation at the

time when data was first reported to the database here are two indicators o partial or delayed

commissioning

bull For an initial series o months the calculated load actor is very low ndash usually below 5 ndash

and then increases abruptly In such cases the data or the early months is dropped and the

commissioning date is treated as the month immediately preceding the first month with a

lsquonormalrsquo load actor14

12 httpswwwrenewablesandchpogemgovuk

13 See httpwwwthewindpowernetwindarms_list_enphp

14 he month in which a plant is reported as having been commissioned is not included in the analysis

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2652

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 25

the decline in the plantrsquos load actor as it ages As an example consider a plant whose reported

capacity was increased rom 99 MW to 10 MW at age 3 I this was simply a change in report-

ing practice this would reduce the estimated load actor rom age 3 onwards By increasing

the initial capacity rom 99 MW to 10 MW ndash or by reducing the capacity rom 10 MW to 99

MW rom age 3 onwards ndash this bias is removed It is not important which adjustment is madebecause the difference between the alternative assumptions is captured by the unit fixed effect

he numbers o separate sites and o observations per site mean that it is not possible to

provide a simple visual representation o the data Figure 8 provides a summary by illustrating

the distributions o load actors by the age o the wind arm pooled across all years he ranges

o the load actors or each age are wide while the medians show no clear trend by age hus a

naiumlve analysis might conclude that there is nothing to investigate However as will be explained

below the distributions mask crucial differences between the perormance o wind arms as

they age because they do not control or the differences in location and wind availability over

time

B Data for Denmark

he data or Danish wind arms used in this study comes rom a database compiled by the

Danish Energy Agency covering the characteristics and perormance o all wind turbines rom

2002 up to the end o August 201216 he basic recording unit or the register is the individual

wind turbine but where a number o turbines are connected to a single meter an identical

average output per turbine is recorded or all turbines in the group he presence o what are in

effect repeated observations increases the noise in the data In addition the model that is used

or the analysis assumes that site operator and turbine characteristics are common or all o theunits that make up a single wind arm o address this point separate wind arms have been

identified by grouping all turbines which have identical values or turbine manuacturer and

type local authority date o commissioning and date o decommissioning (where applicable)

he register contains inormation on 7569 turbines commissioned rom 1977 onwards his

includes a large number o small turbines which would not all within the scope o the ROC

scheme in the UK Hence or comparability with the UK data all turbines commissioned

prior to 1992 were dropped as well as all wind arms with a generating capacity o less than 1

MW he resulting sample covered 823 onshore wind arms and 30 offshore wind arms with

monthly load actors or various periods rom January 2002 to August 2012 Given the cut-off

or wind arms commissioned prior to 1992 the maximum age or an onshore wind arm was 20

years and the maximum age or an offshore wind arm was 17 years

he total capacity o the onshore wind arms recorded in the Danish database was 2757 MW

he average size o an onshore wind arm in Denmark is small ndash only 33 MW as compared

with 161 MW or the UK ndash because 84 o the sample consists o installations with between

1 and 4 turbines he largest onshore installation has 39 turbines with a capacity o 234 MW

while the onshore installation with the greatest capacity has 19 turbines with a capacity o 584

MW Overall the sample o Danish onshore wind arms is older and smaller than the sample

o UK onshore wind arms reflecting the much longer history o wind generation in Denmark

16 httpwwwensdkEN-USINFOFACSANDFIGURESENERGY_SAISICS_AND_INDICAORS

OVERVIEWOFHEENERGYSECORREGISEROFWINDURBINESSiderForsideaspx

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2852

26 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

he larger scale o UK wind arms may account or some o the differences in perormance

curves or the two countries

he sample o offshore wind arms in Denmark is quite small with 30 installations in

total Still it is useul to analyse the perormance o these installations since there is very little

comparable data or the UK he sample covers 394 turbines with an average o just over 13turbines per installation and an average capacity o 288 MW he largest installation which

was commissioned in 2003 has 72 turbines with a total capacity o 1656 MW Again offshore

installations are smaller and older in Denmark than is the current pattern in the UK he first

two wind arms in the sample were commissioned in the years 1995 and 2000 but the sample

increases significantly rom 2002 onwards o avoid reliance upon a sample o 1 or 2 wind

arms to determine specific coefficients or ages gt 10 years all observations with an age greater

than 10 were omitted rom the analysis17

A somewhat different problem with the offshore wind arm data arose because the coeffi-

cients estimated using capacity weights (see below) were heavily influenced by the largest windarm his has had an extremely erratic profile o perormance over time with a very low average

load actor he treatment o this wind arm is discussed in more detail in Section F below

C Specication and estimation methods

As Figure 8 illustrates it is necessary to go beyond simple summary statistics in order to assess

whether there is any systematic relationship between age and the average perormance o wind

arms he perormance o wind arms varies over both time and space some months and years

have more or less wind than the long term average while specific characteristics o wind arms

ndash including location type o turbine and operating regime ndash will influence the perormance oeach plant under identical wind conditions

he combination o (a) site-specific characteristics which are constant over time and (b)

period-specific characteristics which are constant across wind arms can be represented by

what statisticians call an error components model with fixed effects or each wind arm and

each time period he dependent variable in the model consists o a sequence o load actors ndash

denoted by or unit i in month t ndash or each wind arm (the panel unit) over time he load

actors are calculated by dividing the total output in MWh by (24 x number o days in month x

reported generating capacity in MW) and are multiplied by 100 to convert to percentages

he additive version o the error components model may be written as

(1)

in which denotes the age o plant i in period t and is some unction that is either a

representation o or an approximation to the normalised age-perormance relationship he

multiplicative version is similar except that is replaced by where ln() denotes the

natural logarithm he terms are wind arm or unit fixed effects which capture actors such

as location type o turbine operating regimes etc he terms are period fixed effects which

capture the influence o wind conditions over the UK or Denmark as a whole as well as seasonal

maintenance and demand Without loss o generality it may be assumed that the fixed effects are

17 For the avoidance o doubt this means that no attempt was made to estimate the effect o age on the perormance o

offshore wind arms beyond age 10 while the data or the oldest wind arms was included in the analysis up to and

including age 10

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 27

normalised so that ie all o the error components have a zero mean while the

mean o the random error is zero by definition he terms are random errors which capture

random variations in wind conditions that are not specific to the site turbine breakdowns and

other actors that are uncorrelated with either the site or the time period It is assumed that the

mean o the random error is zero while additional assumptions depend upon the method o

estimation that is adopted

he model specified in equation (1) can be estimated in a variety o ways18 A crucial issue is

whether the are or may be significantly correlated with the age terms or any other independent

regressors I it is assumed that there is no such correlation the model is usually reerred to as

a random effects model In that case the equation can be estimated by a variant o least squares

(LS) generalised least squares (GLS) generalised methods o moments (GMM) or maximum

likelihood (ML) ndash all o which should be yield parameter estimates that will converge in prob-

ability to the true parameter values as the sample size increases An alternative but less restric-

tive specification is known as the fixed effects model in which no assumption is made about

the correlation between the unit fixed effects and other regressors he limitation o the fixed

effects model is that it cannot be used to estimate parameters or regressors that vary across

panel units but are constant over time or a panel unit (time-invariant regressors) ndash eg the

year in which a wind arm was commissioned its location or its rated generating capacity he

influence o such variables is captured by the unit fixed effects

In order to avoid the risk o obtaining biased estimates o the age effects the results reported

in this note are based upon estimates using the fixed effects model he influence o time-invar-

iant regressors is examined in a second stage by regressing the estimates o the unit fixed effectson these variables he simplest method o estimating the fixed effects model is known as the

lsquowithinrsquo estimator in which the mean value or each unit is subtracted rom all observations or

that unit For example i is a general polynomial o order M in A equation (1) becomes

(2)

so that the model can be estimated by least square with the inclusion o dummy variables or

each period t he variance-covariance matrix o the coefficients have been estimated by using

(a) a robust sandwich estimator adjusted or clustering by panel unit and (b) a bootstrap esti-

mator with 400 repetitions he estimates o the standard errors will be consistent even i the

random error component is serially correlated within panel units andor has different variances

across panel units19 Since the number o panel units and the average number o time periods

per panel units are both large (other than or Danish offshore wind arms) the least square esti-

mates o the parameters will be unbiased under these assumptions

he bootstrap method is expensive to carry out but it provides a useul cross-check on

the standard errors generated by other methods Unortunately the bootstrap method o

18 See Chapter 21 o AC Cameron amp PK rivedi ndash Microeconometrics Methods and Applications (Cambridge

Cambridge University Press 2005) or a standard textbook treatment o estimation methods or panel data models

19 he estimation was carried out using the xtreg procedure in Stata Version 12 he methods o constructing the

sandwich and bootstrap estimates o the variance-covariance matrix are described in Chapters 8 amp 13 o AC

Cameron amp PK rivedi ndash Microeconometrics using Stata (Revised Edition College Station exas Stata Press 2010)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 6: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 652

List of Figures and TablesList of Figures and Tables

Figure 1 Perormance degradation due to age using equal weights 11

Figure 2 Perormance degradation due to age using capacity weights 12

Figure 3 Range o unit fixed effects by category 13

Figure 4 Initial load actors by year o commissioning or onshore wind 14

Figure 5 Impact o perormance degradation on discounted cumulative output 16

Figure 6A Projected load actors or UK onshore wind (total capacity o 10 GW in 2020) 18

Figure 6B Projected load actors or UK onshore wind (total capacity o 15 GW in 2020) 18

Figure 7 Projected load actors or UK offshore wind in 2020 20

Figure 8 Box plots o load actors by age or UK onshore wind arms 24

Figure 9A Additive age-perormance curves or UK onshore wind arms 32

Figure 9B Multiplicative age-perormance curves or UK onshore wind arms 32

Figure 10 UK onshore age-perormance curves using equal and capacity weights 33

Figure 11A Residuals by age or perormance curves using equal weights 34

Figure 11B Residuals by age or perormance curves using capacity weights 34

Figure 12 Unit fixed effects by year o commissioning or England and Scotland 35

Figure 13A Age-perormance curves or Danish onshore wind arms 37

Figure 13B Age-perormance curves or Danish offshore wind arms 37

Figure 14 Danish age-perormance curves using equal and capacity weights 38

able 1 Average load actors by year and country () 40

able 2 Estimation results or UK onshore wind arms 41

able 3 Equations or trends in unit fixed effects 43

able 4 Estimation results or Danish wind arms 44

able 5 Perormance by year o commissioning or Danish wind arms 46

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 752

About the author

Dr Gordon Hughes is a Proessor o Economics at the University o Edinburgh where he teaches courses in

the Economics o Natural Resources and Public Economics He was a senior adviser on energy and environ-

mental policy at the World Bank until 2001 He has advised governments on the design and implementationo environmental policies and was responsible or some o the World Bankrsquos most important environmental

guidelines

About the Renewable Energy Foundation

he Renewable Energy Foundation is a registered charity promoting sustainable development or the benefit

o the public by means o energy conservation and the use o renewable energy

REF is supported by private donation and has no political affiliation or corporate membership In pursuit o

its principal goals REF highlights the need or an overall energy policy that is balanced ecologically sensi-

tive and effective

We aim to raise public awareness o the issues and encourage inormed debate regarding a structured energy

policy that is both ecologically sensitive and practical he issues o climate change and security o energy

supply are complex and closely intertwined REF contributes to the debate surrounding these issues by

commissioning reports to provide an independent and authoritative source o inormation

For urther inormation see wwwreorguk

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 852

Authorrsquos Note

Note he views expressed in this paper are strictly personal and do not reflect the position o any organ-

isation with which I am associated I am grateul to John Constable Lee Moroney and a reeree or their

comments on earlier drafs o the paper he results o this study have been presented at seminars andlectures to various groups in Edinburgh and Glasgow I have benefited rom the questions and comments

rom participants in those events

Gordon Hughes

School o Economics

University o Edinburgh

gahughesedacuk

Note on the Data

For the convenience o interested researchers both the raw and cleaned data employed in this analysis will

published alongside the electronic version o the study on the REF website wwwreorguk

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 7

Executive Summary

1 Onshore wind turbines represent a relatively mature technology which ought to have achieved

a satisactory level o reliability in operation as plants age Unortunately detailed analysis o

the relationship between age and perormance gives a rather different picture or both theUnited Kingdom and Denmark with a significant decline in the average load actor o onshore

wind arms adjusted or wind availability as they get older An even more dramatic decline is

observed or offshore wind arms in Denmark but this may be a reflection o the immaturity

o the technology

2 he study has used data on the monthly output o wind arms in the UK and Denmark reported

under regulatory arrangements and schemes or subsidising renewable energy Normalised

age-perormance curves have been estimated using standard statistical techniques which allow

or differences between sites and over time in wind resources and other actors

3 he normalised load actor or UK onshore wind arms declines rom a peak o about 24 at

age 1 to 15 at age 10 and 11 at age 15 he decline in the normalised load actor or Danish

onshore wind arms is slower but still significant with a all rom a peak o 22 to 18 at age 15

On the other hand or offshore wind arms in Denmark the normalised load actor alls rom

39 at age 0 to 15 at age 10 he reasons or the observed declines in normalised load actors

cannot be ully assessed using the data available but outages due to mechanical breakdowns

appear to be a contributory actor

4 Analysis o site-specific perormance reveals that the average normalised load actor o new UK

onshore wind arms at age 1 (the peak year o operation) declined significantly rom 2000 to2011 In addition larger wind arms have systematically worse perormance than smaller wind

arms Adjusted or age and wind availability the overall perormance o wind arms in the UK

has deteriorated markedly since the beginning o the century

5 hese findings have important implications or policy towards wind generation in the UK First

they suggest that the subsidy regime is extremely generous i investment in new wind arms

is profitable despite the decline in perormance due to age and over time Second meeting

the UK Governmentrsquos targets or wind generation will require a much higher level o wind

capacity ndash and thus capital investment ndash than current projections imply hird the structure

o contracts offered to wind generators under the proposed reorm o the electricity market

should be modified since ew wind arms will operate or more than 12ndash15 years

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 9

The Performance of Wind Farms in the United Kingdom

and Denmark

Introduction

1 Any assessment o the costs o wind power must rely heavily upon assumptions about theaverage load actor that will be achieved by new wind installations over their lietime It is

standard practice to calculate average load actors by year and country or onshore and offshore

installations as shown in able 1 (page 40)1 However such estimates do not provide a

reliable statistical basis or assessing the uture perormance o wind arms in aggregate Part

o the reason is that the amount o wind in any month or year is influenced by long term mete-

orological cycles that have periods o many years notably the North Atlantic Oscillation In

addition average load actors do not allow or changes in the composition o wind installations

by location age size and other actors

2 As the number o wind arms operating in developed countries has grown engineers have an

opportunity to analyse operating experience over extended periods o time With any (relatively)

novel technology the incidence o temporary or permanent breakdowns may be expected to all

over time so it is difficult to disentangle the effects o age on operating perormance rom the

technological immaturity o turbines installed 10 or 20 years ago Nonetheless the technology

or onshore wind turbines has been reasonably mature since the early 2000s his is documented

by data produced by the US Department o Energy which shows that the decline in capital costs

that is characteristic o immature technologies slowed afer 2000 and was reversed afer 20042

3 With at least 10 years o operating data since 2000 it should be possible to examine how thetypical operating perormance o wind installations in the United Kingdom and Denmark

varies with the age o the turbines he estimated age-perormance curves should then be

incorporated in estimates o levelised costs which are ofen used to compare the costs o wind

and other orms o generation

4 he Appendix provides a detailed description o the data and methods employed or this study

In both the UK and Denmark wind operators have a large incentive to produce electricity

whenever there is sufficient wind available since marginal operating costs are small while the

operator receives a much higher price In the UK this price is the sum o the market price plus

the value o (a) the Renewable Obligation Certificates (ROCs) awarded or each MWh o elec-

tricity produced and (b) the associated exemption rom the Climate Change Levy In practice

the effective market value o ROCs and associated incentives means that an onshore wind

operator earns roughly double the average market price o electricity per MWh In Denmark

most onshore wind capacity receives a price premium equivalent to about pound28 per MWh on top

1 he load actor is calculated as the ratio o the amount o electricity actually produced by a turbine or wind arm

over a period o a month or a year divided by the amount o output that would have been produced had it operated

at ull nameplate capacity or the entire period his is expressed as a percentage so that reported load actors lie

between 0 and 100

2 R Wiser amp M Bolinger (2012) 2011 Wind Technologies Market Report Lawrence Berkeley National Laboratory

US Department o Energy It should be noted that the installed cost o wind plants has allen since 2010 but these

changes reflect the cyclical orces o demand and supply that are well established across the electricity sector and

which apply to ossil-uel as well as renewable generators

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1252

10 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

o the market price or about 125 years o operation at a typical load actor3 Offshore capacity

is eligible or a similar price premium but most o it is contracted via tenders that offer a guar-

anteed eed-in tariff he tender tariffs have varied rom pound56 per MWh to twice that figure

extending over 16-18 years o operation at a typical load actor

Age-performance curves

5 A standard indicator o the operating perormance o a wind turbine is its normalised load

actor defined as the total output over some period divided by the maximum potential output

adjusted or wind availability Normalisation or wind availability is not straightorward I

detailed data on wind speeds at specific locations is available then the potential output rom

individual turbines or wind arms can be calculated using standard power curves Since wind

speeds can vary substantially over small distances such calculations are both difficult to imple-

ment and prone to errors when applied to large numbers o wind generators using public

sources o data4

6 Many analyses work in the opposite direction For example the Danish Wind Index ndash the

longest running series o data on wind availability in Europe ndash is constructed rom the common

component o monthly variations in wind output rom a large sample o wind turbines5 his

study adopts a similar approach Wind availability is treated as an unobserved actor whose

contribution to monthly variations in wind output is estimated by a series o fixed period

(monthyear) effects in a statistical model o monthly output or all wind arms he reasons

or adopting this approach are discussed in Section D o the Appendix his explains why the

method must perorm better than the easible alternative o using a single index o average

wind speed or the UK or Denmark

7 he analysis relies upon standard techniques that are widely used in the biological and medical

sciences as well economics and engineering Full details are given in the Appendix he idea is

to treat the actual load actor (ie without adjustment or wind conditions) or a specific wind

generator in any time period as being determined by a set o components which reflect specific

site conditions and the age-related perormance o wind plants in general together with the

period fixed effects and an uncorrelated random error he onshore wind datasets or the UK

and Denmark used or the analysis are large with monthly observations on 282 installations in

the UK and 823 installations in Denmark with an age range rom 0 to 19 years he offshore

wind dataset or Denmark is rather smaller covering only 30 installations but it can be used to

obtain reasonable estimates o perormance up to age 10 he results discussed here are based

upon least squares estimation (see page 27) combined with standard errors that are robust to

various departures rom classical assumptions about the nature o the data generation process

8 he results o the statistical analysis demonstrate an unambiguous and statistically significant

decline in the operating perormance o wind arms as they grow older Figure 1 shows the

3 Conversions to GBP are based on an exchange rate o pound1 = DKK 918 (as at 19 October 2012)

4 All wind generators collect their own inormation on wind speeds in order to manage their plants but they do not

publish such inormation and will not have access to inormation collected by other operators

5 Details o the construction o the Danish Wind Index are given at wwwvindstatdk A paper by Boccard ndash N Boccard

(2009) lsquoCapacity actor o wind power realized values vs estimatesrsquo Energy Policy Vol 37 pp 2679-88 ndash includes a

discussion o the construction o similar indices o wind availability in Germany the UK Netherlands and Sweden

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 11

simplest variant o the normalised age-perormance curve or onshore UK wind installations

together with the equivalent curves or onshore and offshore Danish installations he curves

illustrated are calculated rom the multiplicative error components model described in the

Appendix with log(load actor) as the dependent variable and a quadratic in plant age to repre-

sent the variation in plant perormance with age9 he rate o decline in perormance is greatest or offshore installations in Denmark with a all

rom load actors o over 40 at ages 0 and 1 to less than 15 by at ages 9 and 10 Onshore

installations in the UK show a more rapid rate o decline ndash 09 percentage points per year over

the first 10 years o operation ndash than is the case or Denmark though the normalised Danish

perormance curve lies below the UK curve or the first our years For the UK the normalised

load actor alls to just over 15 at age 10 and to 11 at age 15 With such low load actors it

seems likely that many wind arms will be re-powered ndash ie the turbines will be replaced ndash once

they reach the age o 10 or at most 15

Figure 1 Performance degradation due to age using equal weights

0

5

10

15

20

25

30

35

40

45

50

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

L o

a d f a c t o r ( )

Age (years)

UK onshore DK onshore DK offshore

Note Normalised load actors in Source Authorrsquos estimates

10 here are two plausible explanations or the observed decline in average load actors as wind

arms age he first is that the turbines become less efficient over time as a result o mechanical

wear and tear erosion o the turbine blades and related actors he second is that the turbines

experience more requent breakdowns and their operators take more time to bring them back

into service because they are less concerned about the perormance o older plants Both

reasons may be relevant in different circumstances and it is not possible to identiy a primary

explanation rom the data he requency o extended shutdowns does seem to increase with

age but this could be a reflection o the timing o planned maintenance operations rather than

breakdowns Whatever the cause the reduction in perormance with age is much greater than

would be expected or thermal generating plants

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1452

12 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

11 he age-perormance curves in Figure 1 are derived rom equations estimated by giving equal

weight to each wind arm irrespective o its generating capacity Figure 2 illustrates similar

curves but in this case the estimates are constructed by weighting each wind arm by its generat-

ing capacity reerred to as capacity-weighted his gives a better representation o perormance

degradation per MW o generating capacity he striking result is that the rate o decline inthe perormance o UK onshore wind installations is significantly aster when capacity weights

are used which implies that large wind arms in the UK experience a more rapid decline than

smaller ones he differences are smaller in Denmark but the capacity-weighted decline in

perormance is more rapid than the equal weights decline or Danish onshore installations

Figure 2 Performance degradation due to age using capacity weights

0

5

10

15

20

25

30

35

40

45

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

L o a d f a c t o r ( )

Age (years)

UK onshore DK onshore DK offshore

Note Normalised load actors in Source Authorrsquos estimates

Differences between wind farms

12 he specification o the statistical model means that the unit fixed effects identiy the specific

perormance characteristics o individual wind arms afer adjusting or wind availability and

age hese characteristics may reflect the site andor design o the installation as well as the

way in which it is operated A positive value or the unit fixed effect is equivalent to shifing

the perormance curve up by some percentage applied uniormly at each age while a negative

value shifs it down It ollows that the distribution o unit fixed effects across installations can

be used to compare the relative effectiveness o new installations by location date o commis-

sioning etc Care is required when making comparisons between or example UK and Danish

onshore wind installations because these have been normalised separately so the unit fixed

effects reer to different underlying averages6

6 Some care is required in interpreting the unit fixed effects By deault the average value o the unit fixed effects or

each group is zero but these averages are taken over all observations or each installation in the basic sample For the

purpose o the comparisons in Figure 3 the unit fixed effects have been adjusted so that the average values or each

group are zero when using one value or each installation

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 13

13 o give a sense o the numbers unit fixed effects o (a) 02 or (b) -02 mean that the base load

actor given the age o the plant is multiplied by (a) exp(02) = 122 or (b) exp(-02) = 082 Since

the base load actor or a plant o age 0 is 231 these fixed effects translate to load actors in

the first year o operation o (a) 282 and (b) 189 For a wind arm with 50 MW o gener-

ating capacity the difference amounts to an additional 40700 MWh o electricity output peryear Using the standard figures cited by RenewableUK this would be sufficient to supply 12300

homes or a year his range (040) is slightly greater than the interquartile range o unit fixed

effects or UK onshore wind arms (035) Again or comparison operators o thermal power

plants would regard any difference o even 10 up or down in efficiency relative to what is

expected as a striking and perhaps worrying indication o either good or bad perormance

Figure 3 Range of unit fixed effects by category

- 1

- 5

0

5

1

A d j f i x e d e f f e c t ndash q u a d r a t i c

UK onshore DK onshore DK offshore

Source Authorrsquos estimates

14 A notable eature o the distributions o the unit fixed effects that are summarised in the box

plots shown in Figure 3 is the range o variation between the best and worst perormers espe-

cially or UK onshore and Danish offshore wind arms Box plots are constructed so that the

top and bottom o the box correspond to the upper and lower quartiles while the bars at topand bottom correspond to the maximum and minimum values excluding outliers7 A plant at

the upper quartile o the distribution o UK onshore wind arms will generate an annual output

that is about 40 higher than a plant at the lower quartile he equivalent figure or Danish

onshore wind plants is an increase o 30 while or Danish offshore wind arms the difference

between the upper and lower quartiles is equivalent to 25 times the output o a lower quartile

plant By any standard such differences are important

7 he upper quartile is the point in a distribution such that the 25 o values exceed it while the lower quartile is the

point in a distribution such that 75 o values exceed it he line in the middle o the box marks the median the

point exceeded by 50 o values Outliers are defined as values that lie outside the range o ukey adjacent values

which are defined as the upperlower quartiles +- 15 x interquartile range

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1652

14 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

15 Outside the range between the upper and lower quartiles the top and bottom segments o

the distributions extend urther up and down or UK onshore wind arms than or those in

Denmark he rather different pattern or Danish offshore wind installations is partly a conse-

quence o the relatively small size o the sample but the evidence points to considerably greater

variability in the perormance o offshore wind installations than or onshore installations16 It is somewhat surprising to observe the magnitude o the differences in the perormance in

onshore wind arms in the UK Given the nature o the subsidy regimes and the high capital

cost o developing new installations it might be expected that operators have a strong incen-

tive to identiy the best locations and then choose equipment that will deliver the maximum

amount o electricity output at a high level o reliability For UK onshore operators it seems that

good or bad perormance is somewhat o a lottery However i the subsidies provided by ROCs

are sufficient to underwrite investment in inefficient plants ndash as appears to be the case ndash then

those subsidies are extremely generous or plants that operate close to the efficient rontier

As location is likely to be the main actor that determines the perormance o a specific plant

relative to all other plants the inerence must be that many wind plants have been developed on

sites with poor wind characteristics

Figure 4 Initial load factors by year of commissioning for onshore wind

10

15

20

25

30

35

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

I n i t i a l l o a d f a c t o r ( )

Year commissioned

UK onshore DK onshore Linear (UK onshore) Linear (DK onshore)

Source Authorrsquos estimates

17 here is another disturbing characteristic o UK wind arms which is revealed by examination

o the unit fixed effects and illustrated in Figure 4 It might be expected that the average peror-

mance o recently installed units would be higher than that o older units his should reflect

improvements in turbine design and reliability as well as in the identification o good sites or

wind generation However that is not the pattern or onshore wind arms in the UK he figure

shows the average values o the initial load actors calculated using equally weighted data by

year o commissioning or onshore wind arms in the UK and Denmark together with fitted

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 15

trend lines8 here has been a clear decline in the initial load actors in the UK which has not

occurred in Denmark his pattern is confirmed by more detailed statistical results shown in

the Appendix which take account o location and size o installation

18 he decline in initial perormance is particularly obvious afer 2005 when the rate o commis-

sioning new capacity increased sharply rom less than 500 MW in the 5 years 2000ndash04 to morethan 500 MW per year rom 2005 onwards In contrast the peak years or commissioning new

onshore wind arms in Denmark were 2000ndash02 when 920 MW was installed he rate o new

building (or replacement) picked up again in 2008 but it is still much lower than at the turn o

century Not only is the perormance o the onshore wind plants in the UK commissioned in

recent years significantly worse than that or wind arms commissioned beore 2005 but it is

possible that this is a direct consequence o an overly rapid expansion o capacity In essence

the evidence suggests that the industry does not have the capability to identiy develop and

operate new onshore wind arms at the rate envisaged by UK government targets while main-

taining a satisactory average level o perormance

Implications for future policy and performance

19 he pattern o perormance degradation with age and over time identified in this analysis has

implications or the assessment o new investment in wind arms and or the design o the

proposed Electricity Market Reorm (EMR) here are two related issues that are affected by

the results

a) What are the implications or the design o policies intended to promote the adoption

and expansion o wind generation in the UK his includes the structure o current

subsidies as well as the specification o the contracts that are proposed under the EMR A

related question concerns the expected lie o new wind arms

b) How might the average perormance o wind generation develop over the next decade

his is absolutely critical to any assessment o the amount o wind capacity that will be

required to meet the UK governmentrsquos targets or the share o electricity ndash and more

generally energy ndash rom renewable sources he government has assumed that the

average load actors or both onshore and offshore wind arms will either remain stable

or increase in uture he prospects will be very different i these assumptions are not

borne out

20 he first question can be addressed by considering the discounted sum o cumulative net

output rom a new wind arm on the assumption that its perormance matches the averages

identified in the analysis his takes account o both perormance degradation over time plus

the annual cost o maintaining the plant which may be expressed as a deduction rom gross

production so as to calculate the contribution o expected output at age s to recovering the

original capital cost o building the plant hese figures have to be discounted back to the date

o commissioning in order to assess the net present value o the initial investment in the plant

he curves or constant and observed perormance using a discount rate o 9 (a typical cost o

capital or wind generation) are shown in Figure 5 he black line is based on the ull set o age

8 he initial load actors are calculated as the normalised load actor or UKDanish wind arms at age 0 multiplied by

where is the unit fixed effect or installation i

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1852

16 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

effects while the red dashed line is based on the smoothed quadratic perormance curve and

the blue dashed line shows what the pattern would be i operating perormance was not affected

by age he calculation is extended out to age 25 since DECCrsquos standard comparisons assume

that wind turbines have a lie o 25 years

Figure 5 Impact of performance degradation on discounted cumulative output

0

10

20

30

40

50

60

70

80

90

100

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

o

f d i s c o u n t e d c u m u l a t i v e o u t p u t

Age of plant (years)

Constant Quadratic Full age effects

Source Authorrsquos estimates

21 he key points in the figure are that 80 o the discounted cumulative output o a new wind

plant is likely to be produced in the first 10 years o its lie and 90 in the first 14 years his

is consistent with the structure o Danish subsidies or onshore wind arms which extend over

a typical period o 125 years Since sites or wind arms are scarce and involve the payment o

significant rents that may be linked to output it is very unlikely that any new installation o

wind turbines will have an expected lie o more than 15 years Instead as has happened in the

past wind operators will have a strong incentive to decommission plants afer no more than 15

years and replace the turbines with newer equipment

22 As a consequence any economic assessment o wind generation should not be based on an

expected lie which is longer than 15 years In recent work reported in evidence to the House

o Commons Select Committee on Energy and Climate Change I assumed that wind plants

would have a residual value equal to 20 o their initial cost in real terms at the end o 15 years

he analysis in this paper suggests that this is too avourable an assumption Given the costs o

decommissioning old turbines the residual value is likely to be well below 10 o their initial

cost and the decision point may be at 10 rather than 15 years

23 he corollary o this observation is that it makes little sense to offer long term contracts or 20

years or more that guarantee prices or eed-in tariffs (Fis) to wind operators A contract length

o 10 to 12 years would be sufficient to remove most o the market risk associated with invest-

ment in wind generation In this respect the subsidy arrangements implemented in Denmark

are better designed

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 17

24 At the same time the offer o subsidies andor guaranteed prices may have serious adverse

consequences or the efficiency o wind generation Returning to the unit fixed effects measur-

ing the perormance o wind plants commissioned afer 2005 only 28 out o 159 units have an

operating perormance that exceeds the average or the period hose 28 units account or 360

MW o capacity whereas the remaining 131 units account or 2810 MW o capacity Not onlyare recent plants less efficient than the average or the whole period but the plants which are

below-average in efficiency are typically larger than the more efficient ones and account or a

disproportionate share o the recent additions to generating capacity

25 In mid-2012 there was approximately 4500 MW o onshore wind capacity operating in the

UK By 2020 it is expected that there will be at least 10000 MW (based on proposed wind

arms with planning consent) and perhaps as much 15000 MW o wind capacity in operation

he government is relying upon a substantial contribution rom onshore wind arms towards

meeting its targets or renewable energy in 2020 o provide context the governmentrsquos targets

or renewable energy in 2020 speciy a total o 234 Wh o energy rom all renewable sources

o which 90ndash140 Wh is expected to come rom wind and biomass electricity generation9

he projections or biomass electricity are extremely vague and have to be reconciled with

an assumed increase o 3ndash4 times in the amount o biomass used separately or heat So in

concrete terms it is reasonable to assume that wind generation will account or at least 90 Wh

with 24ndash32 Wh rom onshore generation and the remainder rom offshore

26 he average load actor or onshore wind over the 10 year period rom April 2002 to March

2012 was 256 which is influenced by the large amounts o new capacity added rom 2005

onwards I this load actor were achieved in 2020 the total amount o onshore wind capacityrequired to meet the onshore generation target would be 107 to 143 GW he average load

actor or offshore wind installations (on an unchanged configuration basis) or the five year

period 2007ndash11 was 32010 I this load actor were achieved in 2020 the total amount o

offshore wind capacity required to meet the remainder o the 90 Wh target would be 206

to 235 GW though these estimates would all to 180 to 205 GW i the average load actor

matched the historic average value o 367 or Danish offshore wind installations

27 he question that has to be addressed is whether these load actors are consistent with (a) the

investment programmes that would be required to achieve the total amounts o generation

capacity implied by the governmentrsquos targets and (b) the age profiles o the perormance o

wind arms discussed in this paper he short answer is that this is simply not possible i the

initial load actor or new wind arms continues to decline at the rate observed the past decade

So even as a starting point it is necessary to assume that this underlying deterioration in wind

arm perormance ceases his would be a major change in itsel and it is not obvious what

actors might lead to such a result

28 he results illustrated below are based on a vintage model o perormance that takes account

o the actual expansion o capacity up to the end o 2011 plus whatever investment in new or

replacement turbines is required to achieve the target levels o wind capacity in 2020 o high-light the ull range o potential outcomes these targets are either 10 GW or 15 GW o onshore

9 Department o Energy and Climate Change ndash UK Renewable Energy Road Map July 2011 Figure 2

10 he equivalent average load actor or onshore wind arms was 264

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2052

18 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

wind and either 18 or 24 GW o offshore wind he levels o investment required to meet these

targets depend upon the expected lie o wind turbines For onshore turbines an expected lie

o 10 years is consistent with re-powering decisions over the last decade while an expected lie

o 15 years seems to be the maximum consistent with the age-perormance profile For offshore

turbines the analysis uses expected lives o 15 and 20 years

Figure 6A Projected load factors for UK onshore wind (total capacity of 10 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

Figure 6B Projected load factors for UK onshore wind (total capacity of 15 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

29 Figure 6A shows projected values o the average normalised load actors or UK onshore wind

arms in 2015 and 2020 using a target capacity in 2020 equal to 10 GW plus the age-peror-

mance profiles or the UK and Denmark (DK) Since vintage effects are important in this

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 19

calculation the profile o new or replacement investment affects the results In this case the

level o gross investment is assumed to be constant rom 2013 onwards at either 930 or 1220

MW per year ndash the range is defined by the assumed lie o turbines Achieving a target o 15

GW in 2020 (Figure 6B) would require a level o investment rom 2013 onwards o either 1540

or 1930 MW per year Such figures are 3ndash4 times the average level o investment over theperiod 2005ndash11 so the lower target may provide a better indication o what will happen

30 All o the estimates are calibrated to match the observed average load actor in 2007ndash11 so that

they should be interpreted as relative rather than absolute values Using the age-perormance

profile or UK onshore wind installations the average normalised load actor is likely to all by

about 19 rom 2007ndash11 to 2015 and by 23ndash28 to 2020 I the age-perormance curves or

Danish onshore installations were to apply the reduction in the average normalised load actor

would be considerably less but there would still be a all o up to 5 by 2020

31 In the medium term the all is greater i turbines operate or 15 years because the average age

o all turbines will be higher and thus the degradation in perormance due to age will have a

larger effect By 2020 this effect reduces the average normalised load actor rom 203 or a 10

year lie to 195 or a 15 year lie here is a complex economic and commercial trade-off here

o meet a target o 10 GW in 2020 it is necessary to increase the level o investment in onshore

capacity by 30 rom 2013 Almost all o this will take the orm o re-powering existing wind

arms11 he costs are airly high because this will usually involve new development consents

decommissioning the original turbines and replacing them with a smaller number o larger and

more powerul turbines

32 Estimates o re-powering costs are generally treated as being commercially sensitive but broadproject costs cited in specialist magazines suggest that the unit costs are likely to be at least

pound1000 per kW or more than pound300 million per year in aggregate On the other hand the increase

in the amount o energy generated would be about 12 Wh or a 10 year turbine lie rather

than a 15 year turbine lie though this does not allow or the loss o production during re-pow-

ering Under the best circumstances the payback period may be quite short but the best option

rom a broader economic and social perspective is not obvious

33 Even under the best scenario 10 GW o onshore wind capacity will only generate 178 Wh o

electricity in 2020 well below the lower end o the target range or onshore wind generation

specified in DECCrsquos road map In act allowing or interactions between the age-perormance

profile and investment the amount o onshore wind capacity required in 2020 to generate 24

Wh would be a minimum o 134 GW At the top end o DECCrsquos target range the capacity

required in 2020 to generate 32 Wh would be a minimum o 178 GW Since these estimates

are based on a 10 year turbine lie the level o investment required rom 2013 onwards would

be 1650 MW per year or the bottom o the range and 2350 MW or the top end o the range

In the last 8 years total investment in onshore wind has only ever exceeded 600 MW in one

11 Re-powering is usually defined as the ull replacement o existing turbines whereas retrofitting covers the replace-

ment o various mechanical or electrical components In the data analysis re-powering involves the creation o a

new unit whereas retrofitting affects the perormance o a continuing unit Hence the potential benefits o retrofit-

ting are taken into account in constructing the age-perormance curves

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2252

20 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

year (883 MW in 2008) so it seems rather unlikely that even the lower end o DECCrsquos range or

onshore wind generation will actually be realised

34 O course this assessment would be different i UK onshore wind operators were able to match

the Danish age-perormance curve here are many actors which differentiate the UK and

Danish wind industries he average scale o wind arms in the UK is larger they are moreremote and there appears to be significantly greater variation between operating and potential

new sites in achieved perormance It would thereore require a very optimistic observer to

conclude that the difference in age-perormance curves between the two countries will disap-

pear in the short or medium term

Figure 7 Projected load factors for UK offshore wind in 2020

20

22

24

26

28

30

32

34

18 GW ndash life 15 yrs 18 GW ndash life 20 yrs 24 GW ndash life 15 yrs 24 GW ndash life 20 yrs

L o a d f a c t o r ( )

2007ndash11 2015 2020

Note Based on Danish age-perormance curves Source Authorrsquos estimates

35 Figure 7 shows the results rom carrying out a similar exercise or offshore wind arms in the

UK he only evidence on the age-perormance curves or offshore installations is based upon

the Danish experience so these have been used to construct the projections Further the profile

o investment assumes a progressive increase in additions to capacity rom 2013 to 2016 and

then a stable level o investment rom 2017 to 2020 he DECC road map involves a very large

commitment to offshore wind investment ndash in the range 25-35 GW o new capacity annually

rom 2016 onwards As a consequence the average age o offshore capacity in 2020 will be quite

low and the ull effects o the age-perormance curves will only be elt during the decade ollow-

ing 2020 Even so the average normalised load actor in 2020 will be about 15 lower than or

2007ndash11 while by 2025 it will be about 30 below the base level

36 On these projections 232 GW o offshore capacity is the minimum necessary to generate 58

Wh o electricity rom offshore wind in 2020 his requirement will increase to 298 GW by

2025 simply to offset the effect o the ageing o offshore generation capacity on total output

37 Combining the estimates or onshore and offshore wind generation the total amount o wind

capacity required in 2020 to produce 90 Wh o electricity rom wind will be in the range 39

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 21

to 41 GW once allowance is made or the observed age-perormance curves or both onshore

and offshore installations hese estimates are roughly a third higher than official predictions o

28ndash31 GW o wind capacity in 2020

38 urning the figures round gives a slightly different picture o the reality o the governmentrsquos

strategy on renewable electricity generation A total o 30 GW o wind capacity in 2020 (12GW onshore 18 GW offshore) will generate about 64 Wh o electricity in a normal year

Since electricity generation rom wind and biomass are expected to account or 50 o renew-

able energy in 2020 the consequence is that biomass is expected to account or 53 Wh In

2011 landfill gas and sewage sludge digestion accounted or 44 o electricity rom bioenergy

(DUKES ables 64) here are limited opportunities to expand production rom these sources

so a continuation o historic rates o growth would mean that renewable generation rom other

sources o bioenergy would have to increase rom 72 Wh in 2011 to 46 Wh in 2020 here

are severe constraints on increasing the amounts o electricity generated rom animal biomass

biodegradable municipal solid waste and anaerobic digestion so it would be optimistic to

assume that these will contribute more than 6-8 Wh in 2020 (up rom 26 Wh in 2011)

39 In practice i the projections or renewable energy are to be believed they rest upon an assump-

tion that there will be a large increase in the amount o electricity that is generated rom plant

biomass ndash either on its own or through co-firing with coal his increase will involve increas-

ing the amount o electricity generated in these ways rom 46 Wh in 2011 to 38-40 Wh in

2020 Since the amounts o straw and UK-grown timber available or this purpose are strictly

limited this will come down to the amount o wood chips that can be imported and the cost o

doing this40 he historic average load actor or biomass plants is just over 60 so about 7500 MW o

biomass capacity will be required to supply 40 Wh o electricity he weight o wood chips per

MWh o electricity depends upon moisture content storage boiler design and other actors but

an indicative range is 085 to 115 tonnes per MWh o wood chips with a moisture content o

30 According to DECC and Forestry Commission estimates the UK imported 265 million

tonnes o steam coal and 26 million tonnes o other wood products (including wood chips)

in 2011 It seems that the net effect o the governmentrsquos strategy will be to replace 15 million

tonnes o coal imported in 2011 by 32 to 44 million tonnes o imported wood chips his is a

somewhat strange way o increasing energy security

Conclusion

41 Wind power is a highly capital-intensive technology or generating electricity Its merits rely

entirely upon a substitution o capital or uel inputs he same is true or hydro or tidal or wave

power In comparison with hydro power wind is a low quality resource because o its variability

and because it cannot be stored So the economic case or wind power must rest on obtaining

the most out o the wind that is available

42 While the decline in the achieved perormance o onshore wind turbines in Denmark is

much less than that or the UK or offshore nonetheless the decline in expected output under

standardised wind conditions over 10 years is 10 unweighted and 13 capacity weighted

hese declines accelerate afer age 10 so that the reductions in perormance are 17 and 20

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2452

22 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

respectively afer 15 years For UK onshore wind arms the reduction in perormance due to

age is much worse at 27 unweighted and 69 capacity weighted by age 10

43 Evidence on the perormance o Danish offshore installations is both restricted and so poor

that there may be concern that the results are affected by a small number o outliers Still the

sample contains a reasonable number o sites with at least 5 years o operating experience andthe decline in perormance by age 5 is 38 unweighted and 26 capacity weighted

44 In addition to these results there is strong evidence that the average normalised load actor or

new onshore wind installations in the UK has allen significantly over the period rom 2000 to

2011 his is consistent with a pattern in which the most avourable sites are developed first

Equally it could mean that wind developers have been unable to keep up with the rate o new

investment while maintaining the quality o development and operations For example the site

design or selection o turbine characteristics may make less effective use o the available wind

resources or the sites available than was the case in the past

45 Whatever the reasons the deterioration in initial perormance means that the expected returns

rom the expansion in wind capacity both or investors and in terms o the reduction in CO983090

emissions have been alling without a concomitant decrease in the private and social costs

that are borne by customers and the general public Clearly this is unsatisactory at best and it

suggests that the benefits claimed or current policies cannot be taken at ace value

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 23

Appendix Data and Methods

A Data for the United Kingdom

he raw data used or this study was extracted by the Renewable Energy Foundation rom the

Renewables and CHP Register database compiled by Ogem12 his inormation is used in the

administration o the market in Renewable Obligation Certificates (ROCs) he Renewables

Obligation is the primary mechanism by which large scale generators o renewable electric-

ity receive subsidies so the operators o wind arms have a very strong incentive to submit

complete and up-to-date inormation to the database he data extracted covered all onshore

wind generators with at least 05 MW o generating capacity which are eligible or ROCs or the

10 year period April 2002 to March 2012 he key variables are the monthly output o electric-

ity (used or the purpose o allocating ROCs) and nominal generating capacity here are 282

reporting units in the final dataset A reporting unit may correspond to an entire wind arm orto separate phases o development Complications arise when an existing wind arm is re-pow-

ered ie when old wind turbines are replaced by newer and usually more powerul turbines

he database reports this as a change in the generating capacity o the recording unit but in

this analysis repowering is treated as the termination o the old recording unit and the creation

o a new recording unit

A considerable amount o data checking and cleaning was required beore the data could

be analysed he first step was to add the age o each reporting unit his was calculated by

reerence to the month in which the turbines were commissioned For wind arms in operation

in or beore June 2002 the commissioning date was obtained rom various online sources ndash

notably the thewindpowernet database o UK wind arms13 which was cross-checked against

the inormation provided on the websites o wind operators and other sources For recording

units whose first data relates to dates afer June 2002 the commissioning date was initially

assumed to be the month or which data is first reported Cross-checks were carried out or

all recording units which commenced reporting in the period July 2002 to December 2003 In

more than 80 o cases the externally-reported commissioning date is within one month o the

first month or which data is reported he exceptions ndash Bu Farm and Mablethorpe ndash appear to

be due to delays by small operators in reporting data to Ogem

In a number o cases it is clear rom the data that plants were not in ull operation at the

time when data was first reported to the database here are two indicators o partial or delayed

commissioning

bull For an initial series o months the calculated load actor is very low ndash usually below 5 ndash

and then increases abruptly In such cases the data or the early months is dropped and the

commissioning date is treated as the month immediately preceding the first month with a

lsquonormalrsquo load actor14

12 httpswwwrenewablesandchpogemgovuk

13 See httpwwwthewindpowernetwindarms_list_enphp

14 he month in which a plant is reported as having been commissioned is not included in the analysis

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2652

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 25

the decline in the plantrsquos load actor as it ages As an example consider a plant whose reported

capacity was increased rom 99 MW to 10 MW at age 3 I this was simply a change in report-

ing practice this would reduce the estimated load actor rom age 3 onwards By increasing

the initial capacity rom 99 MW to 10 MW ndash or by reducing the capacity rom 10 MW to 99

MW rom age 3 onwards ndash this bias is removed It is not important which adjustment is madebecause the difference between the alternative assumptions is captured by the unit fixed effect

he numbers o separate sites and o observations per site mean that it is not possible to

provide a simple visual representation o the data Figure 8 provides a summary by illustrating

the distributions o load actors by the age o the wind arm pooled across all years he ranges

o the load actors or each age are wide while the medians show no clear trend by age hus a

naiumlve analysis might conclude that there is nothing to investigate However as will be explained

below the distributions mask crucial differences between the perormance o wind arms as

they age because they do not control or the differences in location and wind availability over

time

B Data for Denmark

he data or Danish wind arms used in this study comes rom a database compiled by the

Danish Energy Agency covering the characteristics and perormance o all wind turbines rom

2002 up to the end o August 201216 he basic recording unit or the register is the individual

wind turbine but where a number o turbines are connected to a single meter an identical

average output per turbine is recorded or all turbines in the group he presence o what are in

effect repeated observations increases the noise in the data In addition the model that is used

or the analysis assumes that site operator and turbine characteristics are common or all o theunits that make up a single wind arm o address this point separate wind arms have been

identified by grouping all turbines which have identical values or turbine manuacturer and

type local authority date o commissioning and date o decommissioning (where applicable)

he register contains inormation on 7569 turbines commissioned rom 1977 onwards his

includes a large number o small turbines which would not all within the scope o the ROC

scheme in the UK Hence or comparability with the UK data all turbines commissioned

prior to 1992 were dropped as well as all wind arms with a generating capacity o less than 1

MW he resulting sample covered 823 onshore wind arms and 30 offshore wind arms with

monthly load actors or various periods rom January 2002 to August 2012 Given the cut-off

or wind arms commissioned prior to 1992 the maximum age or an onshore wind arm was 20

years and the maximum age or an offshore wind arm was 17 years

he total capacity o the onshore wind arms recorded in the Danish database was 2757 MW

he average size o an onshore wind arm in Denmark is small ndash only 33 MW as compared

with 161 MW or the UK ndash because 84 o the sample consists o installations with between

1 and 4 turbines he largest onshore installation has 39 turbines with a capacity o 234 MW

while the onshore installation with the greatest capacity has 19 turbines with a capacity o 584

MW Overall the sample o Danish onshore wind arms is older and smaller than the sample

o UK onshore wind arms reflecting the much longer history o wind generation in Denmark

16 httpwwwensdkEN-USINFOFACSANDFIGURESENERGY_SAISICS_AND_INDICAORS

OVERVIEWOFHEENERGYSECORREGISEROFWINDURBINESSiderForsideaspx

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2852

26 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

he larger scale o UK wind arms may account or some o the differences in perormance

curves or the two countries

he sample o offshore wind arms in Denmark is quite small with 30 installations in

total Still it is useul to analyse the perormance o these installations since there is very little

comparable data or the UK he sample covers 394 turbines with an average o just over 13turbines per installation and an average capacity o 288 MW he largest installation which

was commissioned in 2003 has 72 turbines with a total capacity o 1656 MW Again offshore

installations are smaller and older in Denmark than is the current pattern in the UK he first

two wind arms in the sample were commissioned in the years 1995 and 2000 but the sample

increases significantly rom 2002 onwards o avoid reliance upon a sample o 1 or 2 wind

arms to determine specific coefficients or ages gt 10 years all observations with an age greater

than 10 were omitted rom the analysis17

A somewhat different problem with the offshore wind arm data arose because the coeffi-

cients estimated using capacity weights (see below) were heavily influenced by the largest windarm his has had an extremely erratic profile o perormance over time with a very low average

load actor he treatment o this wind arm is discussed in more detail in Section F below

C Specication and estimation methods

As Figure 8 illustrates it is necessary to go beyond simple summary statistics in order to assess

whether there is any systematic relationship between age and the average perormance o wind

arms he perormance o wind arms varies over both time and space some months and years

have more or less wind than the long term average while specific characteristics o wind arms

ndash including location type o turbine and operating regime ndash will influence the perormance oeach plant under identical wind conditions

he combination o (a) site-specific characteristics which are constant over time and (b)

period-specific characteristics which are constant across wind arms can be represented by

what statisticians call an error components model with fixed effects or each wind arm and

each time period he dependent variable in the model consists o a sequence o load actors ndash

denoted by or unit i in month t ndash or each wind arm (the panel unit) over time he load

actors are calculated by dividing the total output in MWh by (24 x number o days in month x

reported generating capacity in MW) and are multiplied by 100 to convert to percentages

he additive version o the error components model may be written as

(1)

in which denotes the age o plant i in period t and is some unction that is either a

representation o or an approximation to the normalised age-perormance relationship he

multiplicative version is similar except that is replaced by where ln() denotes the

natural logarithm he terms are wind arm or unit fixed effects which capture actors such

as location type o turbine operating regimes etc he terms are period fixed effects which

capture the influence o wind conditions over the UK or Denmark as a whole as well as seasonal

maintenance and demand Without loss o generality it may be assumed that the fixed effects are

17 For the avoidance o doubt this means that no attempt was made to estimate the effect o age on the perormance o

offshore wind arms beyond age 10 while the data or the oldest wind arms was included in the analysis up to and

including age 10

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 27

normalised so that ie all o the error components have a zero mean while the

mean o the random error is zero by definition he terms are random errors which capture

random variations in wind conditions that are not specific to the site turbine breakdowns and

other actors that are uncorrelated with either the site or the time period It is assumed that the

mean o the random error is zero while additional assumptions depend upon the method o

estimation that is adopted

he model specified in equation (1) can be estimated in a variety o ways18 A crucial issue is

whether the are or may be significantly correlated with the age terms or any other independent

regressors I it is assumed that there is no such correlation the model is usually reerred to as

a random effects model In that case the equation can be estimated by a variant o least squares

(LS) generalised least squares (GLS) generalised methods o moments (GMM) or maximum

likelihood (ML) ndash all o which should be yield parameter estimates that will converge in prob-

ability to the true parameter values as the sample size increases An alternative but less restric-

tive specification is known as the fixed effects model in which no assumption is made about

the correlation between the unit fixed effects and other regressors he limitation o the fixed

effects model is that it cannot be used to estimate parameters or regressors that vary across

panel units but are constant over time or a panel unit (time-invariant regressors) ndash eg the

year in which a wind arm was commissioned its location or its rated generating capacity he

influence o such variables is captured by the unit fixed effects

In order to avoid the risk o obtaining biased estimates o the age effects the results reported

in this note are based upon estimates using the fixed effects model he influence o time-invar-

iant regressors is examined in a second stage by regressing the estimates o the unit fixed effectson these variables he simplest method o estimating the fixed effects model is known as the

lsquowithinrsquo estimator in which the mean value or each unit is subtracted rom all observations or

that unit For example i is a general polynomial o order M in A equation (1) becomes

(2)

so that the model can be estimated by least square with the inclusion o dummy variables or

each period t he variance-covariance matrix o the coefficients have been estimated by using

(a) a robust sandwich estimator adjusted or clustering by panel unit and (b) a bootstrap esti-

mator with 400 repetitions he estimates o the standard errors will be consistent even i the

random error component is serially correlated within panel units andor has different variances

across panel units19 Since the number o panel units and the average number o time periods

per panel units are both large (other than or Danish offshore wind arms) the least square esti-

mates o the parameters will be unbiased under these assumptions

he bootstrap method is expensive to carry out but it provides a useul cross-check on

the standard errors generated by other methods Unortunately the bootstrap method o

18 See Chapter 21 o AC Cameron amp PK rivedi ndash Microeconometrics Methods and Applications (Cambridge

Cambridge University Press 2005) or a standard textbook treatment o estimation methods or panel data models

19 he estimation was carried out using the xtreg procedure in Stata Version 12 he methods o constructing the

sandwich and bootstrap estimates o the variance-covariance matrix are described in Chapters 8 amp 13 o AC

Cameron amp PK rivedi ndash Microeconometrics using Stata (Revised Edition College Station exas Stata Press 2010)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 7: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 752

About the author

Dr Gordon Hughes is a Proessor o Economics at the University o Edinburgh where he teaches courses in

the Economics o Natural Resources and Public Economics He was a senior adviser on energy and environ-

mental policy at the World Bank until 2001 He has advised governments on the design and implementationo environmental policies and was responsible or some o the World Bankrsquos most important environmental

guidelines

About the Renewable Energy Foundation

he Renewable Energy Foundation is a registered charity promoting sustainable development or the benefit

o the public by means o energy conservation and the use o renewable energy

REF is supported by private donation and has no political affiliation or corporate membership In pursuit o

its principal goals REF highlights the need or an overall energy policy that is balanced ecologically sensi-

tive and effective

We aim to raise public awareness o the issues and encourage inormed debate regarding a structured energy

policy that is both ecologically sensitive and practical he issues o climate change and security o energy

supply are complex and closely intertwined REF contributes to the debate surrounding these issues by

commissioning reports to provide an independent and authoritative source o inormation

For urther inormation see wwwreorguk

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 852

Authorrsquos Note

Note he views expressed in this paper are strictly personal and do not reflect the position o any organ-

isation with which I am associated I am grateul to John Constable Lee Moroney and a reeree or their

comments on earlier drafs o the paper he results o this study have been presented at seminars andlectures to various groups in Edinburgh and Glasgow I have benefited rom the questions and comments

rom participants in those events

Gordon Hughes

School o Economics

University o Edinburgh

gahughesedacuk

Note on the Data

For the convenience o interested researchers both the raw and cleaned data employed in this analysis will

published alongside the electronic version o the study on the REF website wwwreorguk

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 7

Executive Summary

1 Onshore wind turbines represent a relatively mature technology which ought to have achieved

a satisactory level o reliability in operation as plants age Unortunately detailed analysis o

the relationship between age and perormance gives a rather different picture or both theUnited Kingdom and Denmark with a significant decline in the average load actor o onshore

wind arms adjusted or wind availability as they get older An even more dramatic decline is

observed or offshore wind arms in Denmark but this may be a reflection o the immaturity

o the technology

2 he study has used data on the monthly output o wind arms in the UK and Denmark reported

under regulatory arrangements and schemes or subsidising renewable energy Normalised

age-perormance curves have been estimated using standard statistical techniques which allow

or differences between sites and over time in wind resources and other actors

3 he normalised load actor or UK onshore wind arms declines rom a peak o about 24 at

age 1 to 15 at age 10 and 11 at age 15 he decline in the normalised load actor or Danish

onshore wind arms is slower but still significant with a all rom a peak o 22 to 18 at age 15

On the other hand or offshore wind arms in Denmark the normalised load actor alls rom

39 at age 0 to 15 at age 10 he reasons or the observed declines in normalised load actors

cannot be ully assessed using the data available but outages due to mechanical breakdowns

appear to be a contributory actor

4 Analysis o site-specific perormance reveals that the average normalised load actor o new UK

onshore wind arms at age 1 (the peak year o operation) declined significantly rom 2000 to2011 In addition larger wind arms have systematically worse perormance than smaller wind

arms Adjusted or age and wind availability the overall perormance o wind arms in the UK

has deteriorated markedly since the beginning o the century

5 hese findings have important implications or policy towards wind generation in the UK First

they suggest that the subsidy regime is extremely generous i investment in new wind arms

is profitable despite the decline in perormance due to age and over time Second meeting

the UK Governmentrsquos targets or wind generation will require a much higher level o wind

capacity ndash and thus capital investment ndash than current projections imply hird the structure

o contracts offered to wind generators under the proposed reorm o the electricity market

should be modified since ew wind arms will operate or more than 12ndash15 years

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 9

The Performance of Wind Farms in the United Kingdom

and Denmark

Introduction

1 Any assessment o the costs o wind power must rely heavily upon assumptions about theaverage load actor that will be achieved by new wind installations over their lietime It is

standard practice to calculate average load actors by year and country or onshore and offshore

installations as shown in able 1 (page 40)1 However such estimates do not provide a

reliable statistical basis or assessing the uture perormance o wind arms in aggregate Part

o the reason is that the amount o wind in any month or year is influenced by long term mete-

orological cycles that have periods o many years notably the North Atlantic Oscillation In

addition average load actors do not allow or changes in the composition o wind installations

by location age size and other actors

2 As the number o wind arms operating in developed countries has grown engineers have an

opportunity to analyse operating experience over extended periods o time With any (relatively)

novel technology the incidence o temporary or permanent breakdowns may be expected to all

over time so it is difficult to disentangle the effects o age on operating perormance rom the

technological immaturity o turbines installed 10 or 20 years ago Nonetheless the technology

or onshore wind turbines has been reasonably mature since the early 2000s his is documented

by data produced by the US Department o Energy which shows that the decline in capital costs

that is characteristic o immature technologies slowed afer 2000 and was reversed afer 20042

3 With at least 10 years o operating data since 2000 it should be possible to examine how thetypical operating perormance o wind installations in the United Kingdom and Denmark

varies with the age o the turbines he estimated age-perormance curves should then be

incorporated in estimates o levelised costs which are ofen used to compare the costs o wind

and other orms o generation

4 he Appendix provides a detailed description o the data and methods employed or this study

In both the UK and Denmark wind operators have a large incentive to produce electricity

whenever there is sufficient wind available since marginal operating costs are small while the

operator receives a much higher price In the UK this price is the sum o the market price plus

the value o (a) the Renewable Obligation Certificates (ROCs) awarded or each MWh o elec-

tricity produced and (b) the associated exemption rom the Climate Change Levy In practice

the effective market value o ROCs and associated incentives means that an onshore wind

operator earns roughly double the average market price o electricity per MWh In Denmark

most onshore wind capacity receives a price premium equivalent to about pound28 per MWh on top

1 he load actor is calculated as the ratio o the amount o electricity actually produced by a turbine or wind arm

over a period o a month or a year divided by the amount o output that would have been produced had it operated

at ull nameplate capacity or the entire period his is expressed as a percentage so that reported load actors lie

between 0 and 100

2 R Wiser amp M Bolinger (2012) 2011 Wind Technologies Market Report Lawrence Berkeley National Laboratory

US Department o Energy It should be noted that the installed cost o wind plants has allen since 2010 but these

changes reflect the cyclical orces o demand and supply that are well established across the electricity sector and

which apply to ossil-uel as well as renewable generators

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1252

10 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

o the market price or about 125 years o operation at a typical load actor3 Offshore capacity

is eligible or a similar price premium but most o it is contracted via tenders that offer a guar-

anteed eed-in tariff he tender tariffs have varied rom pound56 per MWh to twice that figure

extending over 16-18 years o operation at a typical load actor

Age-performance curves

5 A standard indicator o the operating perormance o a wind turbine is its normalised load

actor defined as the total output over some period divided by the maximum potential output

adjusted or wind availability Normalisation or wind availability is not straightorward I

detailed data on wind speeds at specific locations is available then the potential output rom

individual turbines or wind arms can be calculated using standard power curves Since wind

speeds can vary substantially over small distances such calculations are both difficult to imple-

ment and prone to errors when applied to large numbers o wind generators using public

sources o data4

6 Many analyses work in the opposite direction For example the Danish Wind Index ndash the

longest running series o data on wind availability in Europe ndash is constructed rom the common

component o monthly variations in wind output rom a large sample o wind turbines5 his

study adopts a similar approach Wind availability is treated as an unobserved actor whose

contribution to monthly variations in wind output is estimated by a series o fixed period

(monthyear) effects in a statistical model o monthly output or all wind arms he reasons

or adopting this approach are discussed in Section D o the Appendix his explains why the

method must perorm better than the easible alternative o using a single index o average

wind speed or the UK or Denmark

7 he analysis relies upon standard techniques that are widely used in the biological and medical

sciences as well economics and engineering Full details are given in the Appendix he idea is

to treat the actual load actor (ie without adjustment or wind conditions) or a specific wind

generator in any time period as being determined by a set o components which reflect specific

site conditions and the age-related perormance o wind plants in general together with the

period fixed effects and an uncorrelated random error he onshore wind datasets or the UK

and Denmark used or the analysis are large with monthly observations on 282 installations in

the UK and 823 installations in Denmark with an age range rom 0 to 19 years he offshore

wind dataset or Denmark is rather smaller covering only 30 installations but it can be used to

obtain reasonable estimates o perormance up to age 10 he results discussed here are based

upon least squares estimation (see page 27) combined with standard errors that are robust to

various departures rom classical assumptions about the nature o the data generation process

8 he results o the statistical analysis demonstrate an unambiguous and statistically significant

decline in the operating perormance o wind arms as they grow older Figure 1 shows the

3 Conversions to GBP are based on an exchange rate o pound1 = DKK 918 (as at 19 October 2012)

4 All wind generators collect their own inormation on wind speeds in order to manage their plants but they do not

publish such inormation and will not have access to inormation collected by other operators

5 Details o the construction o the Danish Wind Index are given at wwwvindstatdk A paper by Boccard ndash N Boccard

(2009) lsquoCapacity actor o wind power realized values vs estimatesrsquo Energy Policy Vol 37 pp 2679-88 ndash includes a

discussion o the construction o similar indices o wind availability in Germany the UK Netherlands and Sweden

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 11

simplest variant o the normalised age-perormance curve or onshore UK wind installations

together with the equivalent curves or onshore and offshore Danish installations he curves

illustrated are calculated rom the multiplicative error components model described in the

Appendix with log(load actor) as the dependent variable and a quadratic in plant age to repre-

sent the variation in plant perormance with age9 he rate o decline in perormance is greatest or offshore installations in Denmark with a all

rom load actors o over 40 at ages 0 and 1 to less than 15 by at ages 9 and 10 Onshore

installations in the UK show a more rapid rate o decline ndash 09 percentage points per year over

the first 10 years o operation ndash than is the case or Denmark though the normalised Danish

perormance curve lies below the UK curve or the first our years For the UK the normalised

load actor alls to just over 15 at age 10 and to 11 at age 15 With such low load actors it

seems likely that many wind arms will be re-powered ndash ie the turbines will be replaced ndash once

they reach the age o 10 or at most 15

Figure 1 Performance degradation due to age using equal weights

0

5

10

15

20

25

30

35

40

45

50

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

L o

a d f a c t o r ( )

Age (years)

UK onshore DK onshore DK offshore

Note Normalised load actors in Source Authorrsquos estimates

10 here are two plausible explanations or the observed decline in average load actors as wind

arms age he first is that the turbines become less efficient over time as a result o mechanical

wear and tear erosion o the turbine blades and related actors he second is that the turbines

experience more requent breakdowns and their operators take more time to bring them back

into service because they are less concerned about the perormance o older plants Both

reasons may be relevant in different circumstances and it is not possible to identiy a primary

explanation rom the data he requency o extended shutdowns does seem to increase with

age but this could be a reflection o the timing o planned maintenance operations rather than

breakdowns Whatever the cause the reduction in perormance with age is much greater than

would be expected or thermal generating plants

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1452

12 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

11 he age-perormance curves in Figure 1 are derived rom equations estimated by giving equal

weight to each wind arm irrespective o its generating capacity Figure 2 illustrates similar

curves but in this case the estimates are constructed by weighting each wind arm by its generat-

ing capacity reerred to as capacity-weighted his gives a better representation o perormance

degradation per MW o generating capacity he striking result is that the rate o decline inthe perormance o UK onshore wind installations is significantly aster when capacity weights

are used which implies that large wind arms in the UK experience a more rapid decline than

smaller ones he differences are smaller in Denmark but the capacity-weighted decline in

perormance is more rapid than the equal weights decline or Danish onshore installations

Figure 2 Performance degradation due to age using capacity weights

0

5

10

15

20

25

30

35

40

45

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

L o a d f a c t o r ( )

Age (years)

UK onshore DK onshore DK offshore

Note Normalised load actors in Source Authorrsquos estimates

Differences between wind farms

12 he specification o the statistical model means that the unit fixed effects identiy the specific

perormance characteristics o individual wind arms afer adjusting or wind availability and

age hese characteristics may reflect the site andor design o the installation as well as the

way in which it is operated A positive value or the unit fixed effect is equivalent to shifing

the perormance curve up by some percentage applied uniormly at each age while a negative

value shifs it down It ollows that the distribution o unit fixed effects across installations can

be used to compare the relative effectiveness o new installations by location date o commis-

sioning etc Care is required when making comparisons between or example UK and Danish

onshore wind installations because these have been normalised separately so the unit fixed

effects reer to different underlying averages6

6 Some care is required in interpreting the unit fixed effects By deault the average value o the unit fixed effects or

each group is zero but these averages are taken over all observations or each installation in the basic sample For the

purpose o the comparisons in Figure 3 the unit fixed effects have been adjusted so that the average values or each

group are zero when using one value or each installation

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 13

13 o give a sense o the numbers unit fixed effects o (a) 02 or (b) -02 mean that the base load

actor given the age o the plant is multiplied by (a) exp(02) = 122 or (b) exp(-02) = 082 Since

the base load actor or a plant o age 0 is 231 these fixed effects translate to load actors in

the first year o operation o (a) 282 and (b) 189 For a wind arm with 50 MW o gener-

ating capacity the difference amounts to an additional 40700 MWh o electricity output peryear Using the standard figures cited by RenewableUK this would be sufficient to supply 12300

homes or a year his range (040) is slightly greater than the interquartile range o unit fixed

effects or UK onshore wind arms (035) Again or comparison operators o thermal power

plants would regard any difference o even 10 up or down in efficiency relative to what is

expected as a striking and perhaps worrying indication o either good or bad perormance

Figure 3 Range of unit fixed effects by category

- 1

- 5

0

5

1

A d j f i x e d e f f e c t ndash q u a d r a t i c

UK onshore DK onshore DK offshore

Source Authorrsquos estimates

14 A notable eature o the distributions o the unit fixed effects that are summarised in the box

plots shown in Figure 3 is the range o variation between the best and worst perormers espe-

cially or UK onshore and Danish offshore wind arms Box plots are constructed so that the

top and bottom o the box correspond to the upper and lower quartiles while the bars at topand bottom correspond to the maximum and minimum values excluding outliers7 A plant at

the upper quartile o the distribution o UK onshore wind arms will generate an annual output

that is about 40 higher than a plant at the lower quartile he equivalent figure or Danish

onshore wind plants is an increase o 30 while or Danish offshore wind arms the difference

between the upper and lower quartiles is equivalent to 25 times the output o a lower quartile

plant By any standard such differences are important

7 he upper quartile is the point in a distribution such that the 25 o values exceed it while the lower quartile is the

point in a distribution such that 75 o values exceed it he line in the middle o the box marks the median the

point exceeded by 50 o values Outliers are defined as values that lie outside the range o ukey adjacent values

which are defined as the upperlower quartiles +- 15 x interquartile range

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1652

14 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

15 Outside the range between the upper and lower quartiles the top and bottom segments o

the distributions extend urther up and down or UK onshore wind arms than or those in

Denmark he rather different pattern or Danish offshore wind installations is partly a conse-

quence o the relatively small size o the sample but the evidence points to considerably greater

variability in the perormance o offshore wind installations than or onshore installations16 It is somewhat surprising to observe the magnitude o the differences in the perormance in

onshore wind arms in the UK Given the nature o the subsidy regimes and the high capital

cost o developing new installations it might be expected that operators have a strong incen-

tive to identiy the best locations and then choose equipment that will deliver the maximum

amount o electricity output at a high level o reliability For UK onshore operators it seems that

good or bad perormance is somewhat o a lottery However i the subsidies provided by ROCs

are sufficient to underwrite investment in inefficient plants ndash as appears to be the case ndash then

those subsidies are extremely generous or plants that operate close to the efficient rontier

As location is likely to be the main actor that determines the perormance o a specific plant

relative to all other plants the inerence must be that many wind plants have been developed on

sites with poor wind characteristics

Figure 4 Initial load factors by year of commissioning for onshore wind

10

15

20

25

30

35

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

I n i t i a l l o a d f a c t o r ( )

Year commissioned

UK onshore DK onshore Linear (UK onshore) Linear (DK onshore)

Source Authorrsquos estimates

17 here is another disturbing characteristic o UK wind arms which is revealed by examination

o the unit fixed effects and illustrated in Figure 4 It might be expected that the average peror-

mance o recently installed units would be higher than that o older units his should reflect

improvements in turbine design and reliability as well as in the identification o good sites or

wind generation However that is not the pattern or onshore wind arms in the UK he figure

shows the average values o the initial load actors calculated using equally weighted data by

year o commissioning or onshore wind arms in the UK and Denmark together with fitted

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 15

trend lines8 here has been a clear decline in the initial load actors in the UK which has not

occurred in Denmark his pattern is confirmed by more detailed statistical results shown in

the Appendix which take account o location and size o installation

18 he decline in initial perormance is particularly obvious afer 2005 when the rate o commis-

sioning new capacity increased sharply rom less than 500 MW in the 5 years 2000ndash04 to morethan 500 MW per year rom 2005 onwards In contrast the peak years or commissioning new

onshore wind arms in Denmark were 2000ndash02 when 920 MW was installed he rate o new

building (or replacement) picked up again in 2008 but it is still much lower than at the turn o

century Not only is the perormance o the onshore wind plants in the UK commissioned in

recent years significantly worse than that or wind arms commissioned beore 2005 but it is

possible that this is a direct consequence o an overly rapid expansion o capacity In essence

the evidence suggests that the industry does not have the capability to identiy develop and

operate new onshore wind arms at the rate envisaged by UK government targets while main-

taining a satisactory average level o perormance

Implications for future policy and performance

19 he pattern o perormance degradation with age and over time identified in this analysis has

implications or the assessment o new investment in wind arms and or the design o the

proposed Electricity Market Reorm (EMR) here are two related issues that are affected by

the results

a) What are the implications or the design o policies intended to promote the adoption

and expansion o wind generation in the UK his includes the structure o current

subsidies as well as the specification o the contracts that are proposed under the EMR A

related question concerns the expected lie o new wind arms

b) How might the average perormance o wind generation develop over the next decade

his is absolutely critical to any assessment o the amount o wind capacity that will be

required to meet the UK governmentrsquos targets or the share o electricity ndash and more

generally energy ndash rom renewable sources he government has assumed that the

average load actors or both onshore and offshore wind arms will either remain stable

or increase in uture he prospects will be very different i these assumptions are not

borne out

20 he first question can be addressed by considering the discounted sum o cumulative net

output rom a new wind arm on the assumption that its perormance matches the averages

identified in the analysis his takes account o both perormance degradation over time plus

the annual cost o maintaining the plant which may be expressed as a deduction rom gross

production so as to calculate the contribution o expected output at age s to recovering the

original capital cost o building the plant hese figures have to be discounted back to the date

o commissioning in order to assess the net present value o the initial investment in the plant

he curves or constant and observed perormance using a discount rate o 9 (a typical cost o

capital or wind generation) are shown in Figure 5 he black line is based on the ull set o age

8 he initial load actors are calculated as the normalised load actor or UKDanish wind arms at age 0 multiplied by

where is the unit fixed effect or installation i

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1852

16 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

effects while the red dashed line is based on the smoothed quadratic perormance curve and

the blue dashed line shows what the pattern would be i operating perormance was not affected

by age he calculation is extended out to age 25 since DECCrsquos standard comparisons assume

that wind turbines have a lie o 25 years

Figure 5 Impact of performance degradation on discounted cumulative output

0

10

20

30

40

50

60

70

80

90

100

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

o

f d i s c o u n t e d c u m u l a t i v e o u t p u t

Age of plant (years)

Constant Quadratic Full age effects

Source Authorrsquos estimates

21 he key points in the figure are that 80 o the discounted cumulative output o a new wind

plant is likely to be produced in the first 10 years o its lie and 90 in the first 14 years his

is consistent with the structure o Danish subsidies or onshore wind arms which extend over

a typical period o 125 years Since sites or wind arms are scarce and involve the payment o

significant rents that may be linked to output it is very unlikely that any new installation o

wind turbines will have an expected lie o more than 15 years Instead as has happened in the

past wind operators will have a strong incentive to decommission plants afer no more than 15

years and replace the turbines with newer equipment

22 As a consequence any economic assessment o wind generation should not be based on an

expected lie which is longer than 15 years In recent work reported in evidence to the House

o Commons Select Committee on Energy and Climate Change I assumed that wind plants

would have a residual value equal to 20 o their initial cost in real terms at the end o 15 years

he analysis in this paper suggests that this is too avourable an assumption Given the costs o

decommissioning old turbines the residual value is likely to be well below 10 o their initial

cost and the decision point may be at 10 rather than 15 years

23 he corollary o this observation is that it makes little sense to offer long term contracts or 20

years or more that guarantee prices or eed-in tariffs (Fis) to wind operators A contract length

o 10 to 12 years would be sufficient to remove most o the market risk associated with invest-

ment in wind generation In this respect the subsidy arrangements implemented in Denmark

are better designed

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 17

24 At the same time the offer o subsidies andor guaranteed prices may have serious adverse

consequences or the efficiency o wind generation Returning to the unit fixed effects measur-

ing the perormance o wind plants commissioned afer 2005 only 28 out o 159 units have an

operating perormance that exceeds the average or the period hose 28 units account or 360

MW o capacity whereas the remaining 131 units account or 2810 MW o capacity Not onlyare recent plants less efficient than the average or the whole period but the plants which are

below-average in efficiency are typically larger than the more efficient ones and account or a

disproportionate share o the recent additions to generating capacity

25 In mid-2012 there was approximately 4500 MW o onshore wind capacity operating in the

UK By 2020 it is expected that there will be at least 10000 MW (based on proposed wind

arms with planning consent) and perhaps as much 15000 MW o wind capacity in operation

he government is relying upon a substantial contribution rom onshore wind arms towards

meeting its targets or renewable energy in 2020 o provide context the governmentrsquos targets

or renewable energy in 2020 speciy a total o 234 Wh o energy rom all renewable sources

o which 90ndash140 Wh is expected to come rom wind and biomass electricity generation9

he projections or biomass electricity are extremely vague and have to be reconciled with

an assumed increase o 3ndash4 times in the amount o biomass used separately or heat So in

concrete terms it is reasonable to assume that wind generation will account or at least 90 Wh

with 24ndash32 Wh rom onshore generation and the remainder rom offshore

26 he average load actor or onshore wind over the 10 year period rom April 2002 to March

2012 was 256 which is influenced by the large amounts o new capacity added rom 2005

onwards I this load actor were achieved in 2020 the total amount o onshore wind capacityrequired to meet the onshore generation target would be 107 to 143 GW he average load

actor or offshore wind installations (on an unchanged configuration basis) or the five year

period 2007ndash11 was 32010 I this load actor were achieved in 2020 the total amount o

offshore wind capacity required to meet the remainder o the 90 Wh target would be 206

to 235 GW though these estimates would all to 180 to 205 GW i the average load actor

matched the historic average value o 367 or Danish offshore wind installations

27 he question that has to be addressed is whether these load actors are consistent with (a) the

investment programmes that would be required to achieve the total amounts o generation

capacity implied by the governmentrsquos targets and (b) the age profiles o the perormance o

wind arms discussed in this paper he short answer is that this is simply not possible i the

initial load actor or new wind arms continues to decline at the rate observed the past decade

So even as a starting point it is necessary to assume that this underlying deterioration in wind

arm perormance ceases his would be a major change in itsel and it is not obvious what

actors might lead to such a result

28 he results illustrated below are based on a vintage model o perormance that takes account

o the actual expansion o capacity up to the end o 2011 plus whatever investment in new or

replacement turbines is required to achieve the target levels o wind capacity in 2020 o high-light the ull range o potential outcomes these targets are either 10 GW or 15 GW o onshore

9 Department o Energy and Climate Change ndash UK Renewable Energy Road Map July 2011 Figure 2

10 he equivalent average load actor or onshore wind arms was 264

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2052

18 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

wind and either 18 or 24 GW o offshore wind he levels o investment required to meet these

targets depend upon the expected lie o wind turbines For onshore turbines an expected lie

o 10 years is consistent with re-powering decisions over the last decade while an expected lie

o 15 years seems to be the maximum consistent with the age-perormance profile For offshore

turbines the analysis uses expected lives o 15 and 20 years

Figure 6A Projected load factors for UK onshore wind (total capacity of 10 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

Figure 6B Projected load factors for UK onshore wind (total capacity of 15 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

29 Figure 6A shows projected values o the average normalised load actors or UK onshore wind

arms in 2015 and 2020 using a target capacity in 2020 equal to 10 GW plus the age-peror-

mance profiles or the UK and Denmark (DK) Since vintage effects are important in this

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 19

calculation the profile o new or replacement investment affects the results In this case the

level o gross investment is assumed to be constant rom 2013 onwards at either 930 or 1220

MW per year ndash the range is defined by the assumed lie o turbines Achieving a target o 15

GW in 2020 (Figure 6B) would require a level o investment rom 2013 onwards o either 1540

or 1930 MW per year Such figures are 3ndash4 times the average level o investment over theperiod 2005ndash11 so the lower target may provide a better indication o what will happen

30 All o the estimates are calibrated to match the observed average load actor in 2007ndash11 so that

they should be interpreted as relative rather than absolute values Using the age-perormance

profile or UK onshore wind installations the average normalised load actor is likely to all by

about 19 rom 2007ndash11 to 2015 and by 23ndash28 to 2020 I the age-perormance curves or

Danish onshore installations were to apply the reduction in the average normalised load actor

would be considerably less but there would still be a all o up to 5 by 2020

31 In the medium term the all is greater i turbines operate or 15 years because the average age

o all turbines will be higher and thus the degradation in perormance due to age will have a

larger effect By 2020 this effect reduces the average normalised load actor rom 203 or a 10

year lie to 195 or a 15 year lie here is a complex economic and commercial trade-off here

o meet a target o 10 GW in 2020 it is necessary to increase the level o investment in onshore

capacity by 30 rom 2013 Almost all o this will take the orm o re-powering existing wind

arms11 he costs are airly high because this will usually involve new development consents

decommissioning the original turbines and replacing them with a smaller number o larger and

more powerul turbines

32 Estimates o re-powering costs are generally treated as being commercially sensitive but broadproject costs cited in specialist magazines suggest that the unit costs are likely to be at least

pound1000 per kW or more than pound300 million per year in aggregate On the other hand the increase

in the amount o energy generated would be about 12 Wh or a 10 year turbine lie rather

than a 15 year turbine lie though this does not allow or the loss o production during re-pow-

ering Under the best circumstances the payback period may be quite short but the best option

rom a broader economic and social perspective is not obvious

33 Even under the best scenario 10 GW o onshore wind capacity will only generate 178 Wh o

electricity in 2020 well below the lower end o the target range or onshore wind generation

specified in DECCrsquos road map In act allowing or interactions between the age-perormance

profile and investment the amount o onshore wind capacity required in 2020 to generate 24

Wh would be a minimum o 134 GW At the top end o DECCrsquos target range the capacity

required in 2020 to generate 32 Wh would be a minimum o 178 GW Since these estimates

are based on a 10 year turbine lie the level o investment required rom 2013 onwards would

be 1650 MW per year or the bottom o the range and 2350 MW or the top end o the range

In the last 8 years total investment in onshore wind has only ever exceeded 600 MW in one

11 Re-powering is usually defined as the ull replacement o existing turbines whereas retrofitting covers the replace-

ment o various mechanical or electrical components In the data analysis re-powering involves the creation o a

new unit whereas retrofitting affects the perormance o a continuing unit Hence the potential benefits o retrofit-

ting are taken into account in constructing the age-perormance curves

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2252

20 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

year (883 MW in 2008) so it seems rather unlikely that even the lower end o DECCrsquos range or

onshore wind generation will actually be realised

34 O course this assessment would be different i UK onshore wind operators were able to match

the Danish age-perormance curve here are many actors which differentiate the UK and

Danish wind industries he average scale o wind arms in the UK is larger they are moreremote and there appears to be significantly greater variation between operating and potential

new sites in achieved perormance It would thereore require a very optimistic observer to

conclude that the difference in age-perormance curves between the two countries will disap-

pear in the short or medium term

Figure 7 Projected load factors for UK offshore wind in 2020

20

22

24

26

28

30

32

34

18 GW ndash life 15 yrs 18 GW ndash life 20 yrs 24 GW ndash life 15 yrs 24 GW ndash life 20 yrs

L o a d f a c t o r ( )

2007ndash11 2015 2020

Note Based on Danish age-perormance curves Source Authorrsquos estimates

35 Figure 7 shows the results rom carrying out a similar exercise or offshore wind arms in the

UK he only evidence on the age-perormance curves or offshore installations is based upon

the Danish experience so these have been used to construct the projections Further the profile

o investment assumes a progressive increase in additions to capacity rom 2013 to 2016 and

then a stable level o investment rom 2017 to 2020 he DECC road map involves a very large

commitment to offshore wind investment ndash in the range 25-35 GW o new capacity annually

rom 2016 onwards As a consequence the average age o offshore capacity in 2020 will be quite

low and the ull effects o the age-perormance curves will only be elt during the decade ollow-

ing 2020 Even so the average normalised load actor in 2020 will be about 15 lower than or

2007ndash11 while by 2025 it will be about 30 below the base level

36 On these projections 232 GW o offshore capacity is the minimum necessary to generate 58

Wh o electricity rom offshore wind in 2020 his requirement will increase to 298 GW by

2025 simply to offset the effect o the ageing o offshore generation capacity on total output

37 Combining the estimates or onshore and offshore wind generation the total amount o wind

capacity required in 2020 to produce 90 Wh o electricity rom wind will be in the range 39

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 21

to 41 GW once allowance is made or the observed age-perormance curves or both onshore

and offshore installations hese estimates are roughly a third higher than official predictions o

28ndash31 GW o wind capacity in 2020

38 urning the figures round gives a slightly different picture o the reality o the governmentrsquos

strategy on renewable electricity generation A total o 30 GW o wind capacity in 2020 (12GW onshore 18 GW offshore) will generate about 64 Wh o electricity in a normal year

Since electricity generation rom wind and biomass are expected to account or 50 o renew-

able energy in 2020 the consequence is that biomass is expected to account or 53 Wh In

2011 landfill gas and sewage sludge digestion accounted or 44 o electricity rom bioenergy

(DUKES ables 64) here are limited opportunities to expand production rom these sources

so a continuation o historic rates o growth would mean that renewable generation rom other

sources o bioenergy would have to increase rom 72 Wh in 2011 to 46 Wh in 2020 here

are severe constraints on increasing the amounts o electricity generated rom animal biomass

biodegradable municipal solid waste and anaerobic digestion so it would be optimistic to

assume that these will contribute more than 6-8 Wh in 2020 (up rom 26 Wh in 2011)

39 In practice i the projections or renewable energy are to be believed they rest upon an assump-

tion that there will be a large increase in the amount o electricity that is generated rom plant

biomass ndash either on its own or through co-firing with coal his increase will involve increas-

ing the amount o electricity generated in these ways rom 46 Wh in 2011 to 38-40 Wh in

2020 Since the amounts o straw and UK-grown timber available or this purpose are strictly

limited this will come down to the amount o wood chips that can be imported and the cost o

doing this40 he historic average load actor or biomass plants is just over 60 so about 7500 MW o

biomass capacity will be required to supply 40 Wh o electricity he weight o wood chips per

MWh o electricity depends upon moisture content storage boiler design and other actors but

an indicative range is 085 to 115 tonnes per MWh o wood chips with a moisture content o

30 According to DECC and Forestry Commission estimates the UK imported 265 million

tonnes o steam coal and 26 million tonnes o other wood products (including wood chips)

in 2011 It seems that the net effect o the governmentrsquos strategy will be to replace 15 million

tonnes o coal imported in 2011 by 32 to 44 million tonnes o imported wood chips his is a

somewhat strange way o increasing energy security

Conclusion

41 Wind power is a highly capital-intensive technology or generating electricity Its merits rely

entirely upon a substitution o capital or uel inputs he same is true or hydro or tidal or wave

power In comparison with hydro power wind is a low quality resource because o its variability

and because it cannot be stored So the economic case or wind power must rest on obtaining

the most out o the wind that is available

42 While the decline in the achieved perormance o onshore wind turbines in Denmark is

much less than that or the UK or offshore nonetheless the decline in expected output under

standardised wind conditions over 10 years is 10 unweighted and 13 capacity weighted

hese declines accelerate afer age 10 so that the reductions in perormance are 17 and 20

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2452

22 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

respectively afer 15 years For UK onshore wind arms the reduction in perormance due to

age is much worse at 27 unweighted and 69 capacity weighted by age 10

43 Evidence on the perormance o Danish offshore installations is both restricted and so poor

that there may be concern that the results are affected by a small number o outliers Still the

sample contains a reasonable number o sites with at least 5 years o operating experience andthe decline in perormance by age 5 is 38 unweighted and 26 capacity weighted

44 In addition to these results there is strong evidence that the average normalised load actor or

new onshore wind installations in the UK has allen significantly over the period rom 2000 to

2011 his is consistent with a pattern in which the most avourable sites are developed first

Equally it could mean that wind developers have been unable to keep up with the rate o new

investment while maintaining the quality o development and operations For example the site

design or selection o turbine characteristics may make less effective use o the available wind

resources or the sites available than was the case in the past

45 Whatever the reasons the deterioration in initial perormance means that the expected returns

rom the expansion in wind capacity both or investors and in terms o the reduction in CO983090

emissions have been alling without a concomitant decrease in the private and social costs

that are borne by customers and the general public Clearly this is unsatisactory at best and it

suggests that the benefits claimed or current policies cannot be taken at ace value

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 23

Appendix Data and Methods

A Data for the United Kingdom

he raw data used or this study was extracted by the Renewable Energy Foundation rom the

Renewables and CHP Register database compiled by Ogem12 his inormation is used in the

administration o the market in Renewable Obligation Certificates (ROCs) he Renewables

Obligation is the primary mechanism by which large scale generators o renewable electric-

ity receive subsidies so the operators o wind arms have a very strong incentive to submit

complete and up-to-date inormation to the database he data extracted covered all onshore

wind generators with at least 05 MW o generating capacity which are eligible or ROCs or the

10 year period April 2002 to March 2012 he key variables are the monthly output o electric-

ity (used or the purpose o allocating ROCs) and nominal generating capacity here are 282

reporting units in the final dataset A reporting unit may correspond to an entire wind arm orto separate phases o development Complications arise when an existing wind arm is re-pow-

ered ie when old wind turbines are replaced by newer and usually more powerul turbines

he database reports this as a change in the generating capacity o the recording unit but in

this analysis repowering is treated as the termination o the old recording unit and the creation

o a new recording unit

A considerable amount o data checking and cleaning was required beore the data could

be analysed he first step was to add the age o each reporting unit his was calculated by

reerence to the month in which the turbines were commissioned For wind arms in operation

in or beore June 2002 the commissioning date was obtained rom various online sources ndash

notably the thewindpowernet database o UK wind arms13 which was cross-checked against

the inormation provided on the websites o wind operators and other sources For recording

units whose first data relates to dates afer June 2002 the commissioning date was initially

assumed to be the month or which data is first reported Cross-checks were carried out or

all recording units which commenced reporting in the period July 2002 to December 2003 In

more than 80 o cases the externally-reported commissioning date is within one month o the

first month or which data is reported he exceptions ndash Bu Farm and Mablethorpe ndash appear to

be due to delays by small operators in reporting data to Ogem

In a number o cases it is clear rom the data that plants were not in ull operation at the

time when data was first reported to the database here are two indicators o partial or delayed

commissioning

bull For an initial series o months the calculated load actor is very low ndash usually below 5 ndash

and then increases abruptly In such cases the data or the early months is dropped and the

commissioning date is treated as the month immediately preceding the first month with a

lsquonormalrsquo load actor14

12 httpswwwrenewablesandchpogemgovuk

13 See httpwwwthewindpowernetwindarms_list_enphp

14 he month in which a plant is reported as having been commissioned is not included in the analysis

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2652

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 25

the decline in the plantrsquos load actor as it ages As an example consider a plant whose reported

capacity was increased rom 99 MW to 10 MW at age 3 I this was simply a change in report-

ing practice this would reduce the estimated load actor rom age 3 onwards By increasing

the initial capacity rom 99 MW to 10 MW ndash or by reducing the capacity rom 10 MW to 99

MW rom age 3 onwards ndash this bias is removed It is not important which adjustment is madebecause the difference between the alternative assumptions is captured by the unit fixed effect

he numbers o separate sites and o observations per site mean that it is not possible to

provide a simple visual representation o the data Figure 8 provides a summary by illustrating

the distributions o load actors by the age o the wind arm pooled across all years he ranges

o the load actors or each age are wide while the medians show no clear trend by age hus a

naiumlve analysis might conclude that there is nothing to investigate However as will be explained

below the distributions mask crucial differences between the perormance o wind arms as

they age because they do not control or the differences in location and wind availability over

time

B Data for Denmark

he data or Danish wind arms used in this study comes rom a database compiled by the

Danish Energy Agency covering the characteristics and perormance o all wind turbines rom

2002 up to the end o August 201216 he basic recording unit or the register is the individual

wind turbine but where a number o turbines are connected to a single meter an identical

average output per turbine is recorded or all turbines in the group he presence o what are in

effect repeated observations increases the noise in the data In addition the model that is used

or the analysis assumes that site operator and turbine characteristics are common or all o theunits that make up a single wind arm o address this point separate wind arms have been

identified by grouping all turbines which have identical values or turbine manuacturer and

type local authority date o commissioning and date o decommissioning (where applicable)

he register contains inormation on 7569 turbines commissioned rom 1977 onwards his

includes a large number o small turbines which would not all within the scope o the ROC

scheme in the UK Hence or comparability with the UK data all turbines commissioned

prior to 1992 were dropped as well as all wind arms with a generating capacity o less than 1

MW he resulting sample covered 823 onshore wind arms and 30 offshore wind arms with

monthly load actors or various periods rom January 2002 to August 2012 Given the cut-off

or wind arms commissioned prior to 1992 the maximum age or an onshore wind arm was 20

years and the maximum age or an offshore wind arm was 17 years

he total capacity o the onshore wind arms recorded in the Danish database was 2757 MW

he average size o an onshore wind arm in Denmark is small ndash only 33 MW as compared

with 161 MW or the UK ndash because 84 o the sample consists o installations with between

1 and 4 turbines he largest onshore installation has 39 turbines with a capacity o 234 MW

while the onshore installation with the greatest capacity has 19 turbines with a capacity o 584

MW Overall the sample o Danish onshore wind arms is older and smaller than the sample

o UK onshore wind arms reflecting the much longer history o wind generation in Denmark

16 httpwwwensdkEN-USINFOFACSANDFIGURESENERGY_SAISICS_AND_INDICAORS

OVERVIEWOFHEENERGYSECORREGISEROFWINDURBINESSiderForsideaspx

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2852

26 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

he larger scale o UK wind arms may account or some o the differences in perormance

curves or the two countries

he sample o offshore wind arms in Denmark is quite small with 30 installations in

total Still it is useul to analyse the perormance o these installations since there is very little

comparable data or the UK he sample covers 394 turbines with an average o just over 13turbines per installation and an average capacity o 288 MW he largest installation which

was commissioned in 2003 has 72 turbines with a total capacity o 1656 MW Again offshore

installations are smaller and older in Denmark than is the current pattern in the UK he first

two wind arms in the sample were commissioned in the years 1995 and 2000 but the sample

increases significantly rom 2002 onwards o avoid reliance upon a sample o 1 or 2 wind

arms to determine specific coefficients or ages gt 10 years all observations with an age greater

than 10 were omitted rom the analysis17

A somewhat different problem with the offshore wind arm data arose because the coeffi-

cients estimated using capacity weights (see below) were heavily influenced by the largest windarm his has had an extremely erratic profile o perormance over time with a very low average

load actor he treatment o this wind arm is discussed in more detail in Section F below

C Specication and estimation methods

As Figure 8 illustrates it is necessary to go beyond simple summary statistics in order to assess

whether there is any systematic relationship between age and the average perormance o wind

arms he perormance o wind arms varies over both time and space some months and years

have more or less wind than the long term average while specific characteristics o wind arms

ndash including location type o turbine and operating regime ndash will influence the perormance oeach plant under identical wind conditions

he combination o (a) site-specific characteristics which are constant over time and (b)

period-specific characteristics which are constant across wind arms can be represented by

what statisticians call an error components model with fixed effects or each wind arm and

each time period he dependent variable in the model consists o a sequence o load actors ndash

denoted by or unit i in month t ndash or each wind arm (the panel unit) over time he load

actors are calculated by dividing the total output in MWh by (24 x number o days in month x

reported generating capacity in MW) and are multiplied by 100 to convert to percentages

he additive version o the error components model may be written as

(1)

in which denotes the age o plant i in period t and is some unction that is either a

representation o or an approximation to the normalised age-perormance relationship he

multiplicative version is similar except that is replaced by where ln() denotes the

natural logarithm he terms are wind arm or unit fixed effects which capture actors such

as location type o turbine operating regimes etc he terms are period fixed effects which

capture the influence o wind conditions over the UK or Denmark as a whole as well as seasonal

maintenance and demand Without loss o generality it may be assumed that the fixed effects are

17 For the avoidance o doubt this means that no attempt was made to estimate the effect o age on the perormance o

offshore wind arms beyond age 10 while the data or the oldest wind arms was included in the analysis up to and

including age 10

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 27

normalised so that ie all o the error components have a zero mean while the

mean o the random error is zero by definition he terms are random errors which capture

random variations in wind conditions that are not specific to the site turbine breakdowns and

other actors that are uncorrelated with either the site or the time period It is assumed that the

mean o the random error is zero while additional assumptions depend upon the method o

estimation that is adopted

he model specified in equation (1) can be estimated in a variety o ways18 A crucial issue is

whether the are or may be significantly correlated with the age terms or any other independent

regressors I it is assumed that there is no such correlation the model is usually reerred to as

a random effects model In that case the equation can be estimated by a variant o least squares

(LS) generalised least squares (GLS) generalised methods o moments (GMM) or maximum

likelihood (ML) ndash all o which should be yield parameter estimates that will converge in prob-

ability to the true parameter values as the sample size increases An alternative but less restric-

tive specification is known as the fixed effects model in which no assumption is made about

the correlation between the unit fixed effects and other regressors he limitation o the fixed

effects model is that it cannot be used to estimate parameters or regressors that vary across

panel units but are constant over time or a panel unit (time-invariant regressors) ndash eg the

year in which a wind arm was commissioned its location or its rated generating capacity he

influence o such variables is captured by the unit fixed effects

In order to avoid the risk o obtaining biased estimates o the age effects the results reported

in this note are based upon estimates using the fixed effects model he influence o time-invar-

iant regressors is examined in a second stage by regressing the estimates o the unit fixed effectson these variables he simplest method o estimating the fixed effects model is known as the

lsquowithinrsquo estimator in which the mean value or each unit is subtracted rom all observations or

that unit For example i is a general polynomial o order M in A equation (1) becomes

(2)

so that the model can be estimated by least square with the inclusion o dummy variables or

each period t he variance-covariance matrix o the coefficients have been estimated by using

(a) a robust sandwich estimator adjusted or clustering by panel unit and (b) a bootstrap esti-

mator with 400 repetitions he estimates o the standard errors will be consistent even i the

random error component is serially correlated within panel units andor has different variances

across panel units19 Since the number o panel units and the average number o time periods

per panel units are both large (other than or Danish offshore wind arms) the least square esti-

mates o the parameters will be unbiased under these assumptions

he bootstrap method is expensive to carry out but it provides a useul cross-check on

the standard errors generated by other methods Unortunately the bootstrap method o

18 See Chapter 21 o AC Cameron amp PK rivedi ndash Microeconometrics Methods and Applications (Cambridge

Cambridge University Press 2005) or a standard textbook treatment o estimation methods or panel data models

19 he estimation was carried out using the xtreg procedure in Stata Version 12 he methods o constructing the

sandwich and bootstrap estimates o the variance-covariance matrix are described in Chapters 8 amp 13 o AC

Cameron amp PK rivedi ndash Microeconometrics using Stata (Revised Edition College Station exas Stata Press 2010)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 8: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 852

Authorrsquos Note

Note he views expressed in this paper are strictly personal and do not reflect the position o any organ-

isation with which I am associated I am grateul to John Constable Lee Moroney and a reeree or their

comments on earlier drafs o the paper he results o this study have been presented at seminars andlectures to various groups in Edinburgh and Glasgow I have benefited rom the questions and comments

rom participants in those events

Gordon Hughes

School o Economics

University o Edinburgh

gahughesedacuk

Note on the Data

For the convenience o interested researchers both the raw and cleaned data employed in this analysis will

published alongside the electronic version o the study on the REF website wwwreorguk

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 7

Executive Summary

1 Onshore wind turbines represent a relatively mature technology which ought to have achieved

a satisactory level o reliability in operation as plants age Unortunately detailed analysis o

the relationship between age and perormance gives a rather different picture or both theUnited Kingdom and Denmark with a significant decline in the average load actor o onshore

wind arms adjusted or wind availability as they get older An even more dramatic decline is

observed or offshore wind arms in Denmark but this may be a reflection o the immaturity

o the technology

2 he study has used data on the monthly output o wind arms in the UK and Denmark reported

under regulatory arrangements and schemes or subsidising renewable energy Normalised

age-perormance curves have been estimated using standard statistical techniques which allow

or differences between sites and over time in wind resources and other actors

3 he normalised load actor or UK onshore wind arms declines rom a peak o about 24 at

age 1 to 15 at age 10 and 11 at age 15 he decline in the normalised load actor or Danish

onshore wind arms is slower but still significant with a all rom a peak o 22 to 18 at age 15

On the other hand or offshore wind arms in Denmark the normalised load actor alls rom

39 at age 0 to 15 at age 10 he reasons or the observed declines in normalised load actors

cannot be ully assessed using the data available but outages due to mechanical breakdowns

appear to be a contributory actor

4 Analysis o site-specific perormance reveals that the average normalised load actor o new UK

onshore wind arms at age 1 (the peak year o operation) declined significantly rom 2000 to2011 In addition larger wind arms have systematically worse perormance than smaller wind

arms Adjusted or age and wind availability the overall perormance o wind arms in the UK

has deteriorated markedly since the beginning o the century

5 hese findings have important implications or policy towards wind generation in the UK First

they suggest that the subsidy regime is extremely generous i investment in new wind arms

is profitable despite the decline in perormance due to age and over time Second meeting

the UK Governmentrsquos targets or wind generation will require a much higher level o wind

capacity ndash and thus capital investment ndash than current projections imply hird the structure

o contracts offered to wind generators under the proposed reorm o the electricity market

should be modified since ew wind arms will operate or more than 12ndash15 years

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 9

The Performance of Wind Farms in the United Kingdom

and Denmark

Introduction

1 Any assessment o the costs o wind power must rely heavily upon assumptions about theaverage load actor that will be achieved by new wind installations over their lietime It is

standard practice to calculate average load actors by year and country or onshore and offshore

installations as shown in able 1 (page 40)1 However such estimates do not provide a

reliable statistical basis or assessing the uture perormance o wind arms in aggregate Part

o the reason is that the amount o wind in any month or year is influenced by long term mete-

orological cycles that have periods o many years notably the North Atlantic Oscillation In

addition average load actors do not allow or changes in the composition o wind installations

by location age size and other actors

2 As the number o wind arms operating in developed countries has grown engineers have an

opportunity to analyse operating experience over extended periods o time With any (relatively)

novel technology the incidence o temporary or permanent breakdowns may be expected to all

over time so it is difficult to disentangle the effects o age on operating perormance rom the

technological immaturity o turbines installed 10 or 20 years ago Nonetheless the technology

or onshore wind turbines has been reasonably mature since the early 2000s his is documented

by data produced by the US Department o Energy which shows that the decline in capital costs

that is characteristic o immature technologies slowed afer 2000 and was reversed afer 20042

3 With at least 10 years o operating data since 2000 it should be possible to examine how thetypical operating perormance o wind installations in the United Kingdom and Denmark

varies with the age o the turbines he estimated age-perormance curves should then be

incorporated in estimates o levelised costs which are ofen used to compare the costs o wind

and other orms o generation

4 he Appendix provides a detailed description o the data and methods employed or this study

In both the UK and Denmark wind operators have a large incentive to produce electricity

whenever there is sufficient wind available since marginal operating costs are small while the

operator receives a much higher price In the UK this price is the sum o the market price plus

the value o (a) the Renewable Obligation Certificates (ROCs) awarded or each MWh o elec-

tricity produced and (b) the associated exemption rom the Climate Change Levy In practice

the effective market value o ROCs and associated incentives means that an onshore wind

operator earns roughly double the average market price o electricity per MWh In Denmark

most onshore wind capacity receives a price premium equivalent to about pound28 per MWh on top

1 he load actor is calculated as the ratio o the amount o electricity actually produced by a turbine or wind arm

over a period o a month or a year divided by the amount o output that would have been produced had it operated

at ull nameplate capacity or the entire period his is expressed as a percentage so that reported load actors lie

between 0 and 100

2 R Wiser amp M Bolinger (2012) 2011 Wind Technologies Market Report Lawrence Berkeley National Laboratory

US Department o Energy It should be noted that the installed cost o wind plants has allen since 2010 but these

changes reflect the cyclical orces o demand and supply that are well established across the electricity sector and

which apply to ossil-uel as well as renewable generators

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1252

10 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

o the market price or about 125 years o operation at a typical load actor3 Offshore capacity

is eligible or a similar price premium but most o it is contracted via tenders that offer a guar-

anteed eed-in tariff he tender tariffs have varied rom pound56 per MWh to twice that figure

extending over 16-18 years o operation at a typical load actor

Age-performance curves

5 A standard indicator o the operating perormance o a wind turbine is its normalised load

actor defined as the total output over some period divided by the maximum potential output

adjusted or wind availability Normalisation or wind availability is not straightorward I

detailed data on wind speeds at specific locations is available then the potential output rom

individual turbines or wind arms can be calculated using standard power curves Since wind

speeds can vary substantially over small distances such calculations are both difficult to imple-

ment and prone to errors when applied to large numbers o wind generators using public

sources o data4

6 Many analyses work in the opposite direction For example the Danish Wind Index ndash the

longest running series o data on wind availability in Europe ndash is constructed rom the common

component o monthly variations in wind output rom a large sample o wind turbines5 his

study adopts a similar approach Wind availability is treated as an unobserved actor whose

contribution to monthly variations in wind output is estimated by a series o fixed period

(monthyear) effects in a statistical model o monthly output or all wind arms he reasons

or adopting this approach are discussed in Section D o the Appendix his explains why the

method must perorm better than the easible alternative o using a single index o average

wind speed or the UK or Denmark

7 he analysis relies upon standard techniques that are widely used in the biological and medical

sciences as well economics and engineering Full details are given in the Appendix he idea is

to treat the actual load actor (ie without adjustment or wind conditions) or a specific wind

generator in any time period as being determined by a set o components which reflect specific

site conditions and the age-related perormance o wind plants in general together with the

period fixed effects and an uncorrelated random error he onshore wind datasets or the UK

and Denmark used or the analysis are large with monthly observations on 282 installations in

the UK and 823 installations in Denmark with an age range rom 0 to 19 years he offshore

wind dataset or Denmark is rather smaller covering only 30 installations but it can be used to

obtain reasonable estimates o perormance up to age 10 he results discussed here are based

upon least squares estimation (see page 27) combined with standard errors that are robust to

various departures rom classical assumptions about the nature o the data generation process

8 he results o the statistical analysis demonstrate an unambiguous and statistically significant

decline in the operating perormance o wind arms as they grow older Figure 1 shows the

3 Conversions to GBP are based on an exchange rate o pound1 = DKK 918 (as at 19 October 2012)

4 All wind generators collect their own inormation on wind speeds in order to manage their plants but they do not

publish such inormation and will not have access to inormation collected by other operators

5 Details o the construction o the Danish Wind Index are given at wwwvindstatdk A paper by Boccard ndash N Boccard

(2009) lsquoCapacity actor o wind power realized values vs estimatesrsquo Energy Policy Vol 37 pp 2679-88 ndash includes a

discussion o the construction o similar indices o wind availability in Germany the UK Netherlands and Sweden

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 11

simplest variant o the normalised age-perormance curve or onshore UK wind installations

together with the equivalent curves or onshore and offshore Danish installations he curves

illustrated are calculated rom the multiplicative error components model described in the

Appendix with log(load actor) as the dependent variable and a quadratic in plant age to repre-

sent the variation in plant perormance with age9 he rate o decline in perormance is greatest or offshore installations in Denmark with a all

rom load actors o over 40 at ages 0 and 1 to less than 15 by at ages 9 and 10 Onshore

installations in the UK show a more rapid rate o decline ndash 09 percentage points per year over

the first 10 years o operation ndash than is the case or Denmark though the normalised Danish

perormance curve lies below the UK curve or the first our years For the UK the normalised

load actor alls to just over 15 at age 10 and to 11 at age 15 With such low load actors it

seems likely that many wind arms will be re-powered ndash ie the turbines will be replaced ndash once

they reach the age o 10 or at most 15

Figure 1 Performance degradation due to age using equal weights

0

5

10

15

20

25

30

35

40

45

50

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

L o

a d f a c t o r ( )

Age (years)

UK onshore DK onshore DK offshore

Note Normalised load actors in Source Authorrsquos estimates

10 here are two plausible explanations or the observed decline in average load actors as wind

arms age he first is that the turbines become less efficient over time as a result o mechanical

wear and tear erosion o the turbine blades and related actors he second is that the turbines

experience more requent breakdowns and their operators take more time to bring them back

into service because they are less concerned about the perormance o older plants Both

reasons may be relevant in different circumstances and it is not possible to identiy a primary

explanation rom the data he requency o extended shutdowns does seem to increase with

age but this could be a reflection o the timing o planned maintenance operations rather than

breakdowns Whatever the cause the reduction in perormance with age is much greater than

would be expected or thermal generating plants

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1452

12 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

11 he age-perormance curves in Figure 1 are derived rom equations estimated by giving equal

weight to each wind arm irrespective o its generating capacity Figure 2 illustrates similar

curves but in this case the estimates are constructed by weighting each wind arm by its generat-

ing capacity reerred to as capacity-weighted his gives a better representation o perormance

degradation per MW o generating capacity he striking result is that the rate o decline inthe perormance o UK onshore wind installations is significantly aster when capacity weights

are used which implies that large wind arms in the UK experience a more rapid decline than

smaller ones he differences are smaller in Denmark but the capacity-weighted decline in

perormance is more rapid than the equal weights decline or Danish onshore installations

Figure 2 Performance degradation due to age using capacity weights

0

5

10

15

20

25

30

35

40

45

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

L o a d f a c t o r ( )

Age (years)

UK onshore DK onshore DK offshore

Note Normalised load actors in Source Authorrsquos estimates

Differences between wind farms

12 he specification o the statistical model means that the unit fixed effects identiy the specific

perormance characteristics o individual wind arms afer adjusting or wind availability and

age hese characteristics may reflect the site andor design o the installation as well as the

way in which it is operated A positive value or the unit fixed effect is equivalent to shifing

the perormance curve up by some percentage applied uniormly at each age while a negative

value shifs it down It ollows that the distribution o unit fixed effects across installations can

be used to compare the relative effectiveness o new installations by location date o commis-

sioning etc Care is required when making comparisons between or example UK and Danish

onshore wind installations because these have been normalised separately so the unit fixed

effects reer to different underlying averages6

6 Some care is required in interpreting the unit fixed effects By deault the average value o the unit fixed effects or

each group is zero but these averages are taken over all observations or each installation in the basic sample For the

purpose o the comparisons in Figure 3 the unit fixed effects have been adjusted so that the average values or each

group are zero when using one value or each installation

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 13

13 o give a sense o the numbers unit fixed effects o (a) 02 or (b) -02 mean that the base load

actor given the age o the plant is multiplied by (a) exp(02) = 122 or (b) exp(-02) = 082 Since

the base load actor or a plant o age 0 is 231 these fixed effects translate to load actors in

the first year o operation o (a) 282 and (b) 189 For a wind arm with 50 MW o gener-

ating capacity the difference amounts to an additional 40700 MWh o electricity output peryear Using the standard figures cited by RenewableUK this would be sufficient to supply 12300

homes or a year his range (040) is slightly greater than the interquartile range o unit fixed

effects or UK onshore wind arms (035) Again or comparison operators o thermal power

plants would regard any difference o even 10 up or down in efficiency relative to what is

expected as a striking and perhaps worrying indication o either good or bad perormance

Figure 3 Range of unit fixed effects by category

- 1

- 5

0

5

1

A d j f i x e d e f f e c t ndash q u a d r a t i c

UK onshore DK onshore DK offshore

Source Authorrsquos estimates

14 A notable eature o the distributions o the unit fixed effects that are summarised in the box

plots shown in Figure 3 is the range o variation between the best and worst perormers espe-

cially or UK onshore and Danish offshore wind arms Box plots are constructed so that the

top and bottom o the box correspond to the upper and lower quartiles while the bars at topand bottom correspond to the maximum and minimum values excluding outliers7 A plant at

the upper quartile o the distribution o UK onshore wind arms will generate an annual output

that is about 40 higher than a plant at the lower quartile he equivalent figure or Danish

onshore wind plants is an increase o 30 while or Danish offshore wind arms the difference

between the upper and lower quartiles is equivalent to 25 times the output o a lower quartile

plant By any standard such differences are important

7 he upper quartile is the point in a distribution such that the 25 o values exceed it while the lower quartile is the

point in a distribution such that 75 o values exceed it he line in the middle o the box marks the median the

point exceeded by 50 o values Outliers are defined as values that lie outside the range o ukey adjacent values

which are defined as the upperlower quartiles +- 15 x interquartile range

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1652

14 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

15 Outside the range between the upper and lower quartiles the top and bottom segments o

the distributions extend urther up and down or UK onshore wind arms than or those in

Denmark he rather different pattern or Danish offshore wind installations is partly a conse-

quence o the relatively small size o the sample but the evidence points to considerably greater

variability in the perormance o offshore wind installations than or onshore installations16 It is somewhat surprising to observe the magnitude o the differences in the perormance in

onshore wind arms in the UK Given the nature o the subsidy regimes and the high capital

cost o developing new installations it might be expected that operators have a strong incen-

tive to identiy the best locations and then choose equipment that will deliver the maximum

amount o electricity output at a high level o reliability For UK onshore operators it seems that

good or bad perormance is somewhat o a lottery However i the subsidies provided by ROCs

are sufficient to underwrite investment in inefficient plants ndash as appears to be the case ndash then

those subsidies are extremely generous or plants that operate close to the efficient rontier

As location is likely to be the main actor that determines the perormance o a specific plant

relative to all other plants the inerence must be that many wind plants have been developed on

sites with poor wind characteristics

Figure 4 Initial load factors by year of commissioning for onshore wind

10

15

20

25

30

35

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

I n i t i a l l o a d f a c t o r ( )

Year commissioned

UK onshore DK onshore Linear (UK onshore) Linear (DK onshore)

Source Authorrsquos estimates

17 here is another disturbing characteristic o UK wind arms which is revealed by examination

o the unit fixed effects and illustrated in Figure 4 It might be expected that the average peror-

mance o recently installed units would be higher than that o older units his should reflect

improvements in turbine design and reliability as well as in the identification o good sites or

wind generation However that is not the pattern or onshore wind arms in the UK he figure

shows the average values o the initial load actors calculated using equally weighted data by

year o commissioning or onshore wind arms in the UK and Denmark together with fitted

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 15

trend lines8 here has been a clear decline in the initial load actors in the UK which has not

occurred in Denmark his pattern is confirmed by more detailed statistical results shown in

the Appendix which take account o location and size o installation

18 he decline in initial perormance is particularly obvious afer 2005 when the rate o commis-

sioning new capacity increased sharply rom less than 500 MW in the 5 years 2000ndash04 to morethan 500 MW per year rom 2005 onwards In contrast the peak years or commissioning new

onshore wind arms in Denmark were 2000ndash02 when 920 MW was installed he rate o new

building (or replacement) picked up again in 2008 but it is still much lower than at the turn o

century Not only is the perormance o the onshore wind plants in the UK commissioned in

recent years significantly worse than that or wind arms commissioned beore 2005 but it is

possible that this is a direct consequence o an overly rapid expansion o capacity In essence

the evidence suggests that the industry does not have the capability to identiy develop and

operate new onshore wind arms at the rate envisaged by UK government targets while main-

taining a satisactory average level o perormance

Implications for future policy and performance

19 he pattern o perormance degradation with age and over time identified in this analysis has

implications or the assessment o new investment in wind arms and or the design o the

proposed Electricity Market Reorm (EMR) here are two related issues that are affected by

the results

a) What are the implications or the design o policies intended to promote the adoption

and expansion o wind generation in the UK his includes the structure o current

subsidies as well as the specification o the contracts that are proposed under the EMR A

related question concerns the expected lie o new wind arms

b) How might the average perormance o wind generation develop over the next decade

his is absolutely critical to any assessment o the amount o wind capacity that will be

required to meet the UK governmentrsquos targets or the share o electricity ndash and more

generally energy ndash rom renewable sources he government has assumed that the

average load actors or both onshore and offshore wind arms will either remain stable

or increase in uture he prospects will be very different i these assumptions are not

borne out

20 he first question can be addressed by considering the discounted sum o cumulative net

output rom a new wind arm on the assumption that its perormance matches the averages

identified in the analysis his takes account o both perormance degradation over time plus

the annual cost o maintaining the plant which may be expressed as a deduction rom gross

production so as to calculate the contribution o expected output at age s to recovering the

original capital cost o building the plant hese figures have to be discounted back to the date

o commissioning in order to assess the net present value o the initial investment in the plant

he curves or constant and observed perormance using a discount rate o 9 (a typical cost o

capital or wind generation) are shown in Figure 5 he black line is based on the ull set o age

8 he initial load actors are calculated as the normalised load actor or UKDanish wind arms at age 0 multiplied by

where is the unit fixed effect or installation i

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1852

16 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

effects while the red dashed line is based on the smoothed quadratic perormance curve and

the blue dashed line shows what the pattern would be i operating perormance was not affected

by age he calculation is extended out to age 25 since DECCrsquos standard comparisons assume

that wind turbines have a lie o 25 years

Figure 5 Impact of performance degradation on discounted cumulative output

0

10

20

30

40

50

60

70

80

90

100

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

o

f d i s c o u n t e d c u m u l a t i v e o u t p u t

Age of plant (years)

Constant Quadratic Full age effects

Source Authorrsquos estimates

21 he key points in the figure are that 80 o the discounted cumulative output o a new wind

plant is likely to be produced in the first 10 years o its lie and 90 in the first 14 years his

is consistent with the structure o Danish subsidies or onshore wind arms which extend over

a typical period o 125 years Since sites or wind arms are scarce and involve the payment o

significant rents that may be linked to output it is very unlikely that any new installation o

wind turbines will have an expected lie o more than 15 years Instead as has happened in the

past wind operators will have a strong incentive to decommission plants afer no more than 15

years and replace the turbines with newer equipment

22 As a consequence any economic assessment o wind generation should not be based on an

expected lie which is longer than 15 years In recent work reported in evidence to the House

o Commons Select Committee on Energy and Climate Change I assumed that wind plants

would have a residual value equal to 20 o their initial cost in real terms at the end o 15 years

he analysis in this paper suggests that this is too avourable an assumption Given the costs o

decommissioning old turbines the residual value is likely to be well below 10 o their initial

cost and the decision point may be at 10 rather than 15 years

23 he corollary o this observation is that it makes little sense to offer long term contracts or 20

years or more that guarantee prices or eed-in tariffs (Fis) to wind operators A contract length

o 10 to 12 years would be sufficient to remove most o the market risk associated with invest-

ment in wind generation In this respect the subsidy arrangements implemented in Denmark

are better designed

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 17

24 At the same time the offer o subsidies andor guaranteed prices may have serious adverse

consequences or the efficiency o wind generation Returning to the unit fixed effects measur-

ing the perormance o wind plants commissioned afer 2005 only 28 out o 159 units have an

operating perormance that exceeds the average or the period hose 28 units account or 360

MW o capacity whereas the remaining 131 units account or 2810 MW o capacity Not onlyare recent plants less efficient than the average or the whole period but the plants which are

below-average in efficiency are typically larger than the more efficient ones and account or a

disproportionate share o the recent additions to generating capacity

25 In mid-2012 there was approximately 4500 MW o onshore wind capacity operating in the

UK By 2020 it is expected that there will be at least 10000 MW (based on proposed wind

arms with planning consent) and perhaps as much 15000 MW o wind capacity in operation

he government is relying upon a substantial contribution rom onshore wind arms towards

meeting its targets or renewable energy in 2020 o provide context the governmentrsquos targets

or renewable energy in 2020 speciy a total o 234 Wh o energy rom all renewable sources

o which 90ndash140 Wh is expected to come rom wind and biomass electricity generation9

he projections or biomass electricity are extremely vague and have to be reconciled with

an assumed increase o 3ndash4 times in the amount o biomass used separately or heat So in

concrete terms it is reasonable to assume that wind generation will account or at least 90 Wh

with 24ndash32 Wh rom onshore generation and the remainder rom offshore

26 he average load actor or onshore wind over the 10 year period rom April 2002 to March

2012 was 256 which is influenced by the large amounts o new capacity added rom 2005

onwards I this load actor were achieved in 2020 the total amount o onshore wind capacityrequired to meet the onshore generation target would be 107 to 143 GW he average load

actor or offshore wind installations (on an unchanged configuration basis) or the five year

period 2007ndash11 was 32010 I this load actor were achieved in 2020 the total amount o

offshore wind capacity required to meet the remainder o the 90 Wh target would be 206

to 235 GW though these estimates would all to 180 to 205 GW i the average load actor

matched the historic average value o 367 or Danish offshore wind installations

27 he question that has to be addressed is whether these load actors are consistent with (a) the

investment programmes that would be required to achieve the total amounts o generation

capacity implied by the governmentrsquos targets and (b) the age profiles o the perormance o

wind arms discussed in this paper he short answer is that this is simply not possible i the

initial load actor or new wind arms continues to decline at the rate observed the past decade

So even as a starting point it is necessary to assume that this underlying deterioration in wind

arm perormance ceases his would be a major change in itsel and it is not obvious what

actors might lead to such a result

28 he results illustrated below are based on a vintage model o perormance that takes account

o the actual expansion o capacity up to the end o 2011 plus whatever investment in new or

replacement turbines is required to achieve the target levels o wind capacity in 2020 o high-light the ull range o potential outcomes these targets are either 10 GW or 15 GW o onshore

9 Department o Energy and Climate Change ndash UK Renewable Energy Road Map July 2011 Figure 2

10 he equivalent average load actor or onshore wind arms was 264

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2052

18 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

wind and either 18 or 24 GW o offshore wind he levels o investment required to meet these

targets depend upon the expected lie o wind turbines For onshore turbines an expected lie

o 10 years is consistent with re-powering decisions over the last decade while an expected lie

o 15 years seems to be the maximum consistent with the age-perormance profile For offshore

turbines the analysis uses expected lives o 15 and 20 years

Figure 6A Projected load factors for UK onshore wind (total capacity of 10 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

Figure 6B Projected load factors for UK onshore wind (total capacity of 15 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

29 Figure 6A shows projected values o the average normalised load actors or UK onshore wind

arms in 2015 and 2020 using a target capacity in 2020 equal to 10 GW plus the age-peror-

mance profiles or the UK and Denmark (DK) Since vintage effects are important in this

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 19

calculation the profile o new or replacement investment affects the results In this case the

level o gross investment is assumed to be constant rom 2013 onwards at either 930 or 1220

MW per year ndash the range is defined by the assumed lie o turbines Achieving a target o 15

GW in 2020 (Figure 6B) would require a level o investment rom 2013 onwards o either 1540

or 1930 MW per year Such figures are 3ndash4 times the average level o investment over theperiod 2005ndash11 so the lower target may provide a better indication o what will happen

30 All o the estimates are calibrated to match the observed average load actor in 2007ndash11 so that

they should be interpreted as relative rather than absolute values Using the age-perormance

profile or UK onshore wind installations the average normalised load actor is likely to all by

about 19 rom 2007ndash11 to 2015 and by 23ndash28 to 2020 I the age-perormance curves or

Danish onshore installations were to apply the reduction in the average normalised load actor

would be considerably less but there would still be a all o up to 5 by 2020

31 In the medium term the all is greater i turbines operate or 15 years because the average age

o all turbines will be higher and thus the degradation in perormance due to age will have a

larger effect By 2020 this effect reduces the average normalised load actor rom 203 or a 10

year lie to 195 or a 15 year lie here is a complex economic and commercial trade-off here

o meet a target o 10 GW in 2020 it is necessary to increase the level o investment in onshore

capacity by 30 rom 2013 Almost all o this will take the orm o re-powering existing wind

arms11 he costs are airly high because this will usually involve new development consents

decommissioning the original turbines and replacing them with a smaller number o larger and

more powerul turbines

32 Estimates o re-powering costs are generally treated as being commercially sensitive but broadproject costs cited in specialist magazines suggest that the unit costs are likely to be at least

pound1000 per kW or more than pound300 million per year in aggregate On the other hand the increase

in the amount o energy generated would be about 12 Wh or a 10 year turbine lie rather

than a 15 year turbine lie though this does not allow or the loss o production during re-pow-

ering Under the best circumstances the payback period may be quite short but the best option

rom a broader economic and social perspective is not obvious

33 Even under the best scenario 10 GW o onshore wind capacity will only generate 178 Wh o

electricity in 2020 well below the lower end o the target range or onshore wind generation

specified in DECCrsquos road map In act allowing or interactions between the age-perormance

profile and investment the amount o onshore wind capacity required in 2020 to generate 24

Wh would be a minimum o 134 GW At the top end o DECCrsquos target range the capacity

required in 2020 to generate 32 Wh would be a minimum o 178 GW Since these estimates

are based on a 10 year turbine lie the level o investment required rom 2013 onwards would

be 1650 MW per year or the bottom o the range and 2350 MW or the top end o the range

In the last 8 years total investment in onshore wind has only ever exceeded 600 MW in one

11 Re-powering is usually defined as the ull replacement o existing turbines whereas retrofitting covers the replace-

ment o various mechanical or electrical components In the data analysis re-powering involves the creation o a

new unit whereas retrofitting affects the perormance o a continuing unit Hence the potential benefits o retrofit-

ting are taken into account in constructing the age-perormance curves

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2252

20 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

year (883 MW in 2008) so it seems rather unlikely that even the lower end o DECCrsquos range or

onshore wind generation will actually be realised

34 O course this assessment would be different i UK onshore wind operators were able to match

the Danish age-perormance curve here are many actors which differentiate the UK and

Danish wind industries he average scale o wind arms in the UK is larger they are moreremote and there appears to be significantly greater variation between operating and potential

new sites in achieved perormance It would thereore require a very optimistic observer to

conclude that the difference in age-perormance curves between the two countries will disap-

pear in the short or medium term

Figure 7 Projected load factors for UK offshore wind in 2020

20

22

24

26

28

30

32

34

18 GW ndash life 15 yrs 18 GW ndash life 20 yrs 24 GW ndash life 15 yrs 24 GW ndash life 20 yrs

L o a d f a c t o r ( )

2007ndash11 2015 2020

Note Based on Danish age-perormance curves Source Authorrsquos estimates

35 Figure 7 shows the results rom carrying out a similar exercise or offshore wind arms in the

UK he only evidence on the age-perormance curves or offshore installations is based upon

the Danish experience so these have been used to construct the projections Further the profile

o investment assumes a progressive increase in additions to capacity rom 2013 to 2016 and

then a stable level o investment rom 2017 to 2020 he DECC road map involves a very large

commitment to offshore wind investment ndash in the range 25-35 GW o new capacity annually

rom 2016 onwards As a consequence the average age o offshore capacity in 2020 will be quite

low and the ull effects o the age-perormance curves will only be elt during the decade ollow-

ing 2020 Even so the average normalised load actor in 2020 will be about 15 lower than or

2007ndash11 while by 2025 it will be about 30 below the base level

36 On these projections 232 GW o offshore capacity is the minimum necessary to generate 58

Wh o electricity rom offshore wind in 2020 his requirement will increase to 298 GW by

2025 simply to offset the effect o the ageing o offshore generation capacity on total output

37 Combining the estimates or onshore and offshore wind generation the total amount o wind

capacity required in 2020 to produce 90 Wh o electricity rom wind will be in the range 39

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 21

to 41 GW once allowance is made or the observed age-perormance curves or both onshore

and offshore installations hese estimates are roughly a third higher than official predictions o

28ndash31 GW o wind capacity in 2020

38 urning the figures round gives a slightly different picture o the reality o the governmentrsquos

strategy on renewable electricity generation A total o 30 GW o wind capacity in 2020 (12GW onshore 18 GW offshore) will generate about 64 Wh o electricity in a normal year

Since electricity generation rom wind and biomass are expected to account or 50 o renew-

able energy in 2020 the consequence is that biomass is expected to account or 53 Wh In

2011 landfill gas and sewage sludge digestion accounted or 44 o electricity rom bioenergy

(DUKES ables 64) here are limited opportunities to expand production rom these sources

so a continuation o historic rates o growth would mean that renewable generation rom other

sources o bioenergy would have to increase rom 72 Wh in 2011 to 46 Wh in 2020 here

are severe constraints on increasing the amounts o electricity generated rom animal biomass

biodegradable municipal solid waste and anaerobic digestion so it would be optimistic to

assume that these will contribute more than 6-8 Wh in 2020 (up rom 26 Wh in 2011)

39 In practice i the projections or renewable energy are to be believed they rest upon an assump-

tion that there will be a large increase in the amount o electricity that is generated rom plant

biomass ndash either on its own or through co-firing with coal his increase will involve increas-

ing the amount o electricity generated in these ways rom 46 Wh in 2011 to 38-40 Wh in

2020 Since the amounts o straw and UK-grown timber available or this purpose are strictly

limited this will come down to the amount o wood chips that can be imported and the cost o

doing this40 he historic average load actor or biomass plants is just over 60 so about 7500 MW o

biomass capacity will be required to supply 40 Wh o electricity he weight o wood chips per

MWh o electricity depends upon moisture content storage boiler design and other actors but

an indicative range is 085 to 115 tonnes per MWh o wood chips with a moisture content o

30 According to DECC and Forestry Commission estimates the UK imported 265 million

tonnes o steam coal and 26 million tonnes o other wood products (including wood chips)

in 2011 It seems that the net effect o the governmentrsquos strategy will be to replace 15 million

tonnes o coal imported in 2011 by 32 to 44 million tonnes o imported wood chips his is a

somewhat strange way o increasing energy security

Conclusion

41 Wind power is a highly capital-intensive technology or generating electricity Its merits rely

entirely upon a substitution o capital or uel inputs he same is true or hydro or tidal or wave

power In comparison with hydro power wind is a low quality resource because o its variability

and because it cannot be stored So the economic case or wind power must rest on obtaining

the most out o the wind that is available

42 While the decline in the achieved perormance o onshore wind turbines in Denmark is

much less than that or the UK or offshore nonetheless the decline in expected output under

standardised wind conditions over 10 years is 10 unweighted and 13 capacity weighted

hese declines accelerate afer age 10 so that the reductions in perormance are 17 and 20

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2452

22 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

respectively afer 15 years For UK onshore wind arms the reduction in perormance due to

age is much worse at 27 unweighted and 69 capacity weighted by age 10

43 Evidence on the perormance o Danish offshore installations is both restricted and so poor

that there may be concern that the results are affected by a small number o outliers Still the

sample contains a reasonable number o sites with at least 5 years o operating experience andthe decline in perormance by age 5 is 38 unweighted and 26 capacity weighted

44 In addition to these results there is strong evidence that the average normalised load actor or

new onshore wind installations in the UK has allen significantly over the period rom 2000 to

2011 his is consistent with a pattern in which the most avourable sites are developed first

Equally it could mean that wind developers have been unable to keep up with the rate o new

investment while maintaining the quality o development and operations For example the site

design or selection o turbine characteristics may make less effective use o the available wind

resources or the sites available than was the case in the past

45 Whatever the reasons the deterioration in initial perormance means that the expected returns

rom the expansion in wind capacity both or investors and in terms o the reduction in CO983090

emissions have been alling without a concomitant decrease in the private and social costs

that are borne by customers and the general public Clearly this is unsatisactory at best and it

suggests that the benefits claimed or current policies cannot be taken at ace value

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 23

Appendix Data and Methods

A Data for the United Kingdom

he raw data used or this study was extracted by the Renewable Energy Foundation rom the

Renewables and CHP Register database compiled by Ogem12 his inormation is used in the

administration o the market in Renewable Obligation Certificates (ROCs) he Renewables

Obligation is the primary mechanism by which large scale generators o renewable electric-

ity receive subsidies so the operators o wind arms have a very strong incentive to submit

complete and up-to-date inormation to the database he data extracted covered all onshore

wind generators with at least 05 MW o generating capacity which are eligible or ROCs or the

10 year period April 2002 to March 2012 he key variables are the monthly output o electric-

ity (used or the purpose o allocating ROCs) and nominal generating capacity here are 282

reporting units in the final dataset A reporting unit may correspond to an entire wind arm orto separate phases o development Complications arise when an existing wind arm is re-pow-

ered ie when old wind turbines are replaced by newer and usually more powerul turbines

he database reports this as a change in the generating capacity o the recording unit but in

this analysis repowering is treated as the termination o the old recording unit and the creation

o a new recording unit

A considerable amount o data checking and cleaning was required beore the data could

be analysed he first step was to add the age o each reporting unit his was calculated by

reerence to the month in which the turbines were commissioned For wind arms in operation

in or beore June 2002 the commissioning date was obtained rom various online sources ndash

notably the thewindpowernet database o UK wind arms13 which was cross-checked against

the inormation provided on the websites o wind operators and other sources For recording

units whose first data relates to dates afer June 2002 the commissioning date was initially

assumed to be the month or which data is first reported Cross-checks were carried out or

all recording units which commenced reporting in the period July 2002 to December 2003 In

more than 80 o cases the externally-reported commissioning date is within one month o the

first month or which data is reported he exceptions ndash Bu Farm and Mablethorpe ndash appear to

be due to delays by small operators in reporting data to Ogem

In a number o cases it is clear rom the data that plants were not in ull operation at the

time when data was first reported to the database here are two indicators o partial or delayed

commissioning

bull For an initial series o months the calculated load actor is very low ndash usually below 5 ndash

and then increases abruptly In such cases the data or the early months is dropped and the

commissioning date is treated as the month immediately preceding the first month with a

lsquonormalrsquo load actor14

12 httpswwwrenewablesandchpogemgovuk

13 See httpwwwthewindpowernetwindarms_list_enphp

14 he month in which a plant is reported as having been commissioned is not included in the analysis

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2652

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 25

the decline in the plantrsquos load actor as it ages As an example consider a plant whose reported

capacity was increased rom 99 MW to 10 MW at age 3 I this was simply a change in report-

ing practice this would reduce the estimated load actor rom age 3 onwards By increasing

the initial capacity rom 99 MW to 10 MW ndash or by reducing the capacity rom 10 MW to 99

MW rom age 3 onwards ndash this bias is removed It is not important which adjustment is madebecause the difference between the alternative assumptions is captured by the unit fixed effect

he numbers o separate sites and o observations per site mean that it is not possible to

provide a simple visual representation o the data Figure 8 provides a summary by illustrating

the distributions o load actors by the age o the wind arm pooled across all years he ranges

o the load actors or each age are wide while the medians show no clear trend by age hus a

naiumlve analysis might conclude that there is nothing to investigate However as will be explained

below the distributions mask crucial differences between the perormance o wind arms as

they age because they do not control or the differences in location and wind availability over

time

B Data for Denmark

he data or Danish wind arms used in this study comes rom a database compiled by the

Danish Energy Agency covering the characteristics and perormance o all wind turbines rom

2002 up to the end o August 201216 he basic recording unit or the register is the individual

wind turbine but where a number o turbines are connected to a single meter an identical

average output per turbine is recorded or all turbines in the group he presence o what are in

effect repeated observations increases the noise in the data In addition the model that is used

or the analysis assumes that site operator and turbine characteristics are common or all o theunits that make up a single wind arm o address this point separate wind arms have been

identified by grouping all turbines which have identical values or turbine manuacturer and

type local authority date o commissioning and date o decommissioning (where applicable)

he register contains inormation on 7569 turbines commissioned rom 1977 onwards his

includes a large number o small turbines which would not all within the scope o the ROC

scheme in the UK Hence or comparability with the UK data all turbines commissioned

prior to 1992 were dropped as well as all wind arms with a generating capacity o less than 1

MW he resulting sample covered 823 onshore wind arms and 30 offshore wind arms with

monthly load actors or various periods rom January 2002 to August 2012 Given the cut-off

or wind arms commissioned prior to 1992 the maximum age or an onshore wind arm was 20

years and the maximum age or an offshore wind arm was 17 years

he total capacity o the onshore wind arms recorded in the Danish database was 2757 MW

he average size o an onshore wind arm in Denmark is small ndash only 33 MW as compared

with 161 MW or the UK ndash because 84 o the sample consists o installations with between

1 and 4 turbines he largest onshore installation has 39 turbines with a capacity o 234 MW

while the onshore installation with the greatest capacity has 19 turbines with a capacity o 584

MW Overall the sample o Danish onshore wind arms is older and smaller than the sample

o UK onshore wind arms reflecting the much longer history o wind generation in Denmark

16 httpwwwensdkEN-USINFOFACSANDFIGURESENERGY_SAISICS_AND_INDICAORS

OVERVIEWOFHEENERGYSECORREGISEROFWINDURBINESSiderForsideaspx

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2852

26 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

he larger scale o UK wind arms may account or some o the differences in perormance

curves or the two countries

he sample o offshore wind arms in Denmark is quite small with 30 installations in

total Still it is useul to analyse the perormance o these installations since there is very little

comparable data or the UK he sample covers 394 turbines with an average o just over 13turbines per installation and an average capacity o 288 MW he largest installation which

was commissioned in 2003 has 72 turbines with a total capacity o 1656 MW Again offshore

installations are smaller and older in Denmark than is the current pattern in the UK he first

two wind arms in the sample were commissioned in the years 1995 and 2000 but the sample

increases significantly rom 2002 onwards o avoid reliance upon a sample o 1 or 2 wind

arms to determine specific coefficients or ages gt 10 years all observations with an age greater

than 10 were omitted rom the analysis17

A somewhat different problem with the offshore wind arm data arose because the coeffi-

cients estimated using capacity weights (see below) were heavily influenced by the largest windarm his has had an extremely erratic profile o perormance over time with a very low average

load actor he treatment o this wind arm is discussed in more detail in Section F below

C Specication and estimation methods

As Figure 8 illustrates it is necessary to go beyond simple summary statistics in order to assess

whether there is any systematic relationship between age and the average perormance o wind

arms he perormance o wind arms varies over both time and space some months and years

have more or less wind than the long term average while specific characteristics o wind arms

ndash including location type o turbine and operating regime ndash will influence the perormance oeach plant under identical wind conditions

he combination o (a) site-specific characteristics which are constant over time and (b)

period-specific characteristics which are constant across wind arms can be represented by

what statisticians call an error components model with fixed effects or each wind arm and

each time period he dependent variable in the model consists o a sequence o load actors ndash

denoted by or unit i in month t ndash or each wind arm (the panel unit) over time he load

actors are calculated by dividing the total output in MWh by (24 x number o days in month x

reported generating capacity in MW) and are multiplied by 100 to convert to percentages

he additive version o the error components model may be written as

(1)

in which denotes the age o plant i in period t and is some unction that is either a

representation o or an approximation to the normalised age-perormance relationship he

multiplicative version is similar except that is replaced by where ln() denotes the

natural logarithm he terms are wind arm or unit fixed effects which capture actors such

as location type o turbine operating regimes etc he terms are period fixed effects which

capture the influence o wind conditions over the UK or Denmark as a whole as well as seasonal

maintenance and demand Without loss o generality it may be assumed that the fixed effects are

17 For the avoidance o doubt this means that no attempt was made to estimate the effect o age on the perormance o

offshore wind arms beyond age 10 while the data or the oldest wind arms was included in the analysis up to and

including age 10

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 27

normalised so that ie all o the error components have a zero mean while the

mean o the random error is zero by definition he terms are random errors which capture

random variations in wind conditions that are not specific to the site turbine breakdowns and

other actors that are uncorrelated with either the site or the time period It is assumed that the

mean o the random error is zero while additional assumptions depend upon the method o

estimation that is adopted

he model specified in equation (1) can be estimated in a variety o ways18 A crucial issue is

whether the are or may be significantly correlated with the age terms or any other independent

regressors I it is assumed that there is no such correlation the model is usually reerred to as

a random effects model In that case the equation can be estimated by a variant o least squares

(LS) generalised least squares (GLS) generalised methods o moments (GMM) or maximum

likelihood (ML) ndash all o which should be yield parameter estimates that will converge in prob-

ability to the true parameter values as the sample size increases An alternative but less restric-

tive specification is known as the fixed effects model in which no assumption is made about

the correlation between the unit fixed effects and other regressors he limitation o the fixed

effects model is that it cannot be used to estimate parameters or regressors that vary across

panel units but are constant over time or a panel unit (time-invariant regressors) ndash eg the

year in which a wind arm was commissioned its location or its rated generating capacity he

influence o such variables is captured by the unit fixed effects

In order to avoid the risk o obtaining biased estimates o the age effects the results reported

in this note are based upon estimates using the fixed effects model he influence o time-invar-

iant regressors is examined in a second stage by regressing the estimates o the unit fixed effectson these variables he simplest method o estimating the fixed effects model is known as the

lsquowithinrsquo estimator in which the mean value or each unit is subtracted rom all observations or

that unit For example i is a general polynomial o order M in A equation (1) becomes

(2)

so that the model can be estimated by least square with the inclusion o dummy variables or

each period t he variance-covariance matrix o the coefficients have been estimated by using

(a) a robust sandwich estimator adjusted or clustering by panel unit and (b) a bootstrap esti-

mator with 400 repetitions he estimates o the standard errors will be consistent even i the

random error component is serially correlated within panel units andor has different variances

across panel units19 Since the number o panel units and the average number o time periods

per panel units are both large (other than or Danish offshore wind arms) the least square esti-

mates o the parameters will be unbiased under these assumptions

he bootstrap method is expensive to carry out but it provides a useul cross-check on

the standard errors generated by other methods Unortunately the bootstrap method o

18 See Chapter 21 o AC Cameron amp PK rivedi ndash Microeconometrics Methods and Applications (Cambridge

Cambridge University Press 2005) or a standard textbook treatment o estimation methods or panel data models

19 he estimation was carried out using the xtreg procedure in Stata Version 12 he methods o constructing the

sandwich and bootstrap estimates o the variance-covariance matrix are described in Chapters 8 amp 13 o AC

Cameron amp PK rivedi ndash Microeconometrics using Stata (Revised Edition College Station exas Stata Press 2010)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 9: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 7

Executive Summary

1 Onshore wind turbines represent a relatively mature technology which ought to have achieved

a satisactory level o reliability in operation as plants age Unortunately detailed analysis o

the relationship between age and perormance gives a rather different picture or both theUnited Kingdom and Denmark with a significant decline in the average load actor o onshore

wind arms adjusted or wind availability as they get older An even more dramatic decline is

observed or offshore wind arms in Denmark but this may be a reflection o the immaturity

o the technology

2 he study has used data on the monthly output o wind arms in the UK and Denmark reported

under regulatory arrangements and schemes or subsidising renewable energy Normalised

age-perormance curves have been estimated using standard statistical techniques which allow

or differences between sites and over time in wind resources and other actors

3 he normalised load actor or UK onshore wind arms declines rom a peak o about 24 at

age 1 to 15 at age 10 and 11 at age 15 he decline in the normalised load actor or Danish

onshore wind arms is slower but still significant with a all rom a peak o 22 to 18 at age 15

On the other hand or offshore wind arms in Denmark the normalised load actor alls rom

39 at age 0 to 15 at age 10 he reasons or the observed declines in normalised load actors

cannot be ully assessed using the data available but outages due to mechanical breakdowns

appear to be a contributory actor

4 Analysis o site-specific perormance reveals that the average normalised load actor o new UK

onshore wind arms at age 1 (the peak year o operation) declined significantly rom 2000 to2011 In addition larger wind arms have systematically worse perormance than smaller wind

arms Adjusted or age and wind availability the overall perormance o wind arms in the UK

has deteriorated markedly since the beginning o the century

5 hese findings have important implications or policy towards wind generation in the UK First

they suggest that the subsidy regime is extremely generous i investment in new wind arms

is profitable despite the decline in perormance due to age and over time Second meeting

the UK Governmentrsquos targets or wind generation will require a much higher level o wind

capacity ndash and thus capital investment ndash than current projections imply hird the structure

o contracts offered to wind generators under the proposed reorm o the electricity market

should be modified since ew wind arms will operate or more than 12ndash15 years

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 9

The Performance of Wind Farms in the United Kingdom

and Denmark

Introduction

1 Any assessment o the costs o wind power must rely heavily upon assumptions about theaverage load actor that will be achieved by new wind installations over their lietime It is

standard practice to calculate average load actors by year and country or onshore and offshore

installations as shown in able 1 (page 40)1 However such estimates do not provide a

reliable statistical basis or assessing the uture perormance o wind arms in aggregate Part

o the reason is that the amount o wind in any month or year is influenced by long term mete-

orological cycles that have periods o many years notably the North Atlantic Oscillation In

addition average load actors do not allow or changes in the composition o wind installations

by location age size and other actors

2 As the number o wind arms operating in developed countries has grown engineers have an

opportunity to analyse operating experience over extended periods o time With any (relatively)

novel technology the incidence o temporary or permanent breakdowns may be expected to all

over time so it is difficult to disentangle the effects o age on operating perormance rom the

technological immaturity o turbines installed 10 or 20 years ago Nonetheless the technology

or onshore wind turbines has been reasonably mature since the early 2000s his is documented

by data produced by the US Department o Energy which shows that the decline in capital costs

that is characteristic o immature technologies slowed afer 2000 and was reversed afer 20042

3 With at least 10 years o operating data since 2000 it should be possible to examine how thetypical operating perormance o wind installations in the United Kingdom and Denmark

varies with the age o the turbines he estimated age-perormance curves should then be

incorporated in estimates o levelised costs which are ofen used to compare the costs o wind

and other orms o generation

4 he Appendix provides a detailed description o the data and methods employed or this study

In both the UK and Denmark wind operators have a large incentive to produce electricity

whenever there is sufficient wind available since marginal operating costs are small while the

operator receives a much higher price In the UK this price is the sum o the market price plus

the value o (a) the Renewable Obligation Certificates (ROCs) awarded or each MWh o elec-

tricity produced and (b) the associated exemption rom the Climate Change Levy In practice

the effective market value o ROCs and associated incentives means that an onshore wind

operator earns roughly double the average market price o electricity per MWh In Denmark

most onshore wind capacity receives a price premium equivalent to about pound28 per MWh on top

1 he load actor is calculated as the ratio o the amount o electricity actually produced by a turbine or wind arm

over a period o a month or a year divided by the amount o output that would have been produced had it operated

at ull nameplate capacity or the entire period his is expressed as a percentage so that reported load actors lie

between 0 and 100

2 R Wiser amp M Bolinger (2012) 2011 Wind Technologies Market Report Lawrence Berkeley National Laboratory

US Department o Energy It should be noted that the installed cost o wind plants has allen since 2010 but these

changes reflect the cyclical orces o demand and supply that are well established across the electricity sector and

which apply to ossil-uel as well as renewable generators

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1252

10 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

o the market price or about 125 years o operation at a typical load actor3 Offshore capacity

is eligible or a similar price premium but most o it is contracted via tenders that offer a guar-

anteed eed-in tariff he tender tariffs have varied rom pound56 per MWh to twice that figure

extending over 16-18 years o operation at a typical load actor

Age-performance curves

5 A standard indicator o the operating perormance o a wind turbine is its normalised load

actor defined as the total output over some period divided by the maximum potential output

adjusted or wind availability Normalisation or wind availability is not straightorward I

detailed data on wind speeds at specific locations is available then the potential output rom

individual turbines or wind arms can be calculated using standard power curves Since wind

speeds can vary substantially over small distances such calculations are both difficult to imple-

ment and prone to errors when applied to large numbers o wind generators using public

sources o data4

6 Many analyses work in the opposite direction For example the Danish Wind Index ndash the

longest running series o data on wind availability in Europe ndash is constructed rom the common

component o monthly variations in wind output rom a large sample o wind turbines5 his

study adopts a similar approach Wind availability is treated as an unobserved actor whose

contribution to monthly variations in wind output is estimated by a series o fixed period

(monthyear) effects in a statistical model o monthly output or all wind arms he reasons

or adopting this approach are discussed in Section D o the Appendix his explains why the

method must perorm better than the easible alternative o using a single index o average

wind speed or the UK or Denmark

7 he analysis relies upon standard techniques that are widely used in the biological and medical

sciences as well economics and engineering Full details are given in the Appendix he idea is

to treat the actual load actor (ie without adjustment or wind conditions) or a specific wind

generator in any time period as being determined by a set o components which reflect specific

site conditions and the age-related perormance o wind plants in general together with the

period fixed effects and an uncorrelated random error he onshore wind datasets or the UK

and Denmark used or the analysis are large with monthly observations on 282 installations in

the UK and 823 installations in Denmark with an age range rom 0 to 19 years he offshore

wind dataset or Denmark is rather smaller covering only 30 installations but it can be used to

obtain reasonable estimates o perormance up to age 10 he results discussed here are based

upon least squares estimation (see page 27) combined with standard errors that are robust to

various departures rom classical assumptions about the nature o the data generation process

8 he results o the statistical analysis demonstrate an unambiguous and statistically significant

decline in the operating perormance o wind arms as they grow older Figure 1 shows the

3 Conversions to GBP are based on an exchange rate o pound1 = DKK 918 (as at 19 October 2012)

4 All wind generators collect their own inormation on wind speeds in order to manage their plants but they do not

publish such inormation and will not have access to inormation collected by other operators

5 Details o the construction o the Danish Wind Index are given at wwwvindstatdk A paper by Boccard ndash N Boccard

(2009) lsquoCapacity actor o wind power realized values vs estimatesrsquo Energy Policy Vol 37 pp 2679-88 ndash includes a

discussion o the construction o similar indices o wind availability in Germany the UK Netherlands and Sweden

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 11

simplest variant o the normalised age-perormance curve or onshore UK wind installations

together with the equivalent curves or onshore and offshore Danish installations he curves

illustrated are calculated rom the multiplicative error components model described in the

Appendix with log(load actor) as the dependent variable and a quadratic in plant age to repre-

sent the variation in plant perormance with age9 he rate o decline in perormance is greatest or offshore installations in Denmark with a all

rom load actors o over 40 at ages 0 and 1 to less than 15 by at ages 9 and 10 Onshore

installations in the UK show a more rapid rate o decline ndash 09 percentage points per year over

the first 10 years o operation ndash than is the case or Denmark though the normalised Danish

perormance curve lies below the UK curve or the first our years For the UK the normalised

load actor alls to just over 15 at age 10 and to 11 at age 15 With such low load actors it

seems likely that many wind arms will be re-powered ndash ie the turbines will be replaced ndash once

they reach the age o 10 or at most 15

Figure 1 Performance degradation due to age using equal weights

0

5

10

15

20

25

30

35

40

45

50

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

L o

a d f a c t o r ( )

Age (years)

UK onshore DK onshore DK offshore

Note Normalised load actors in Source Authorrsquos estimates

10 here are two plausible explanations or the observed decline in average load actors as wind

arms age he first is that the turbines become less efficient over time as a result o mechanical

wear and tear erosion o the turbine blades and related actors he second is that the turbines

experience more requent breakdowns and their operators take more time to bring them back

into service because they are less concerned about the perormance o older plants Both

reasons may be relevant in different circumstances and it is not possible to identiy a primary

explanation rom the data he requency o extended shutdowns does seem to increase with

age but this could be a reflection o the timing o planned maintenance operations rather than

breakdowns Whatever the cause the reduction in perormance with age is much greater than

would be expected or thermal generating plants

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1452

12 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

11 he age-perormance curves in Figure 1 are derived rom equations estimated by giving equal

weight to each wind arm irrespective o its generating capacity Figure 2 illustrates similar

curves but in this case the estimates are constructed by weighting each wind arm by its generat-

ing capacity reerred to as capacity-weighted his gives a better representation o perormance

degradation per MW o generating capacity he striking result is that the rate o decline inthe perormance o UK onshore wind installations is significantly aster when capacity weights

are used which implies that large wind arms in the UK experience a more rapid decline than

smaller ones he differences are smaller in Denmark but the capacity-weighted decline in

perormance is more rapid than the equal weights decline or Danish onshore installations

Figure 2 Performance degradation due to age using capacity weights

0

5

10

15

20

25

30

35

40

45

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

L o a d f a c t o r ( )

Age (years)

UK onshore DK onshore DK offshore

Note Normalised load actors in Source Authorrsquos estimates

Differences between wind farms

12 he specification o the statistical model means that the unit fixed effects identiy the specific

perormance characteristics o individual wind arms afer adjusting or wind availability and

age hese characteristics may reflect the site andor design o the installation as well as the

way in which it is operated A positive value or the unit fixed effect is equivalent to shifing

the perormance curve up by some percentage applied uniormly at each age while a negative

value shifs it down It ollows that the distribution o unit fixed effects across installations can

be used to compare the relative effectiveness o new installations by location date o commis-

sioning etc Care is required when making comparisons between or example UK and Danish

onshore wind installations because these have been normalised separately so the unit fixed

effects reer to different underlying averages6

6 Some care is required in interpreting the unit fixed effects By deault the average value o the unit fixed effects or

each group is zero but these averages are taken over all observations or each installation in the basic sample For the

purpose o the comparisons in Figure 3 the unit fixed effects have been adjusted so that the average values or each

group are zero when using one value or each installation

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 13

13 o give a sense o the numbers unit fixed effects o (a) 02 or (b) -02 mean that the base load

actor given the age o the plant is multiplied by (a) exp(02) = 122 or (b) exp(-02) = 082 Since

the base load actor or a plant o age 0 is 231 these fixed effects translate to load actors in

the first year o operation o (a) 282 and (b) 189 For a wind arm with 50 MW o gener-

ating capacity the difference amounts to an additional 40700 MWh o electricity output peryear Using the standard figures cited by RenewableUK this would be sufficient to supply 12300

homes or a year his range (040) is slightly greater than the interquartile range o unit fixed

effects or UK onshore wind arms (035) Again or comparison operators o thermal power

plants would regard any difference o even 10 up or down in efficiency relative to what is

expected as a striking and perhaps worrying indication o either good or bad perormance

Figure 3 Range of unit fixed effects by category

- 1

- 5

0

5

1

A d j f i x e d e f f e c t ndash q u a d r a t i c

UK onshore DK onshore DK offshore

Source Authorrsquos estimates

14 A notable eature o the distributions o the unit fixed effects that are summarised in the box

plots shown in Figure 3 is the range o variation between the best and worst perormers espe-

cially or UK onshore and Danish offshore wind arms Box plots are constructed so that the

top and bottom o the box correspond to the upper and lower quartiles while the bars at topand bottom correspond to the maximum and minimum values excluding outliers7 A plant at

the upper quartile o the distribution o UK onshore wind arms will generate an annual output

that is about 40 higher than a plant at the lower quartile he equivalent figure or Danish

onshore wind plants is an increase o 30 while or Danish offshore wind arms the difference

between the upper and lower quartiles is equivalent to 25 times the output o a lower quartile

plant By any standard such differences are important

7 he upper quartile is the point in a distribution such that the 25 o values exceed it while the lower quartile is the

point in a distribution such that 75 o values exceed it he line in the middle o the box marks the median the

point exceeded by 50 o values Outliers are defined as values that lie outside the range o ukey adjacent values

which are defined as the upperlower quartiles +- 15 x interquartile range

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1652

14 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

15 Outside the range between the upper and lower quartiles the top and bottom segments o

the distributions extend urther up and down or UK onshore wind arms than or those in

Denmark he rather different pattern or Danish offshore wind installations is partly a conse-

quence o the relatively small size o the sample but the evidence points to considerably greater

variability in the perormance o offshore wind installations than or onshore installations16 It is somewhat surprising to observe the magnitude o the differences in the perormance in

onshore wind arms in the UK Given the nature o the subsidy regimes and the high capital

cost o developing new installations it might be expected that operators have a strong incen-

tive to identiy the best locations and then choose equipment that will deliver the maximum

amount o electricity output at a high level o reliability For UK onshore operators it seems that

good or bad perormance is somewhat o a lottery However i the subsidies provided by ROCs

are sufficient to underwrite investment in inefficient plants ndash as appears to be the case ndash then

those subsidies are extremely generous or plants that operate close to the efficient rontier

As location is likely to be the main actor that determines the perormance o a specific plant

relative to all other plants the inerence must be that many wind plants have been developed on

sites with poor wind characteristics

Figure 4 Initial load factors by year of commissioning for onshore wind

10

15

20

25

30

35

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

I n i t i a l l o a d f a c t o r ( )

Year commissioned

UK onshore DK onshore Linear (UK onshore) Linear (DK onshore)

Source Authorrsquos estimates

17 here is another disturbing characteristic o UK wind arms which is revealed by examination

o the unit fixed effects and illustrated in Figure 4 It might be expected that the average peror-

mance o recently installed units would be higher than that o older units his should reflect

improvements in turbine design and reliability as well as in the identification o good sites or

wind generation However that is not the pattern or onshore wind arms in the UK he figure

shows the average values o the initial load actors calculated using equally weighted data by

year o commissioning or onshore wind arms in the UK and Denmark together with fitted

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 15

trend lines8 here has been a clear decline in the initial load actors in the UK which has not

occurred in Denmark his pattern is confirmed by more detailed statistical results shown in

the Appendix which take account o location and size o installation

18 he decline in initial perormance is particularly obvious afer 2005 when the rate o commis-

sioning new capacity increased sharply rom less than 500 MW in the 5 years 2000ndash04 to morethan 500 MW per year rom 2005 onwards In contrast the peak years or commissioning new

onshore wind arms in Denmark were 2000ndash02 when 920 MW was installed he rate o new

building (or replacement) picked up again in 2008 but it is still much lower than at the turn o

century Not only is the perormance o the onshore wind plants in the UK commissioned in

recent years significantly worse than that or wind arms commissioned beore 2005 but it is

possible that this is a direct consequence o an overly rapid expansion o capacity In essence

the evidence suggests that the industry does not have the capability to identiy develop and

operate new onshore wind arms at the rate envisaged by UK government targets while main-

taining a satisactory average level o perormance

Implications for future policy and performance

19 he pattern o perormance degradation with age and over time identified in this analysis has

implications or the assessment o new investment in wind arms and or the design o the

proposed Electricity Market Reorm (EMR) here are two related issues that are affected by

the results

a) What are the implications or the design o policies intended to promote the adoption

and expansion o wind generation in the UK his includes the structure o current

subsidies as well as the specification o the contracts that are proposed under the EMR A

related question concerns the expected lie o new wind arms

b) How might the average perormance o wind generation develop over the next decade

his is absolutely critical to any assessment o the amount o wind capacity that will be

required to meet the UK governmentrsquos targets or the share o electricity ndash and more

generally energy ndash rom renewable sources he government has assumed that the

average load actors or both onshore and offshore wind arms will either remain stable

or increase in uture he prospects will be very different i these assumptions are not

borne out

20 he first question can be addressed by considering the discounted sum o cumulative net

output rom a new wind arm on the assumption that its perormance matches the averages

identified in the analysis his takes account o both perormance degradation over time plus

the annual cost o maintaining the plant which may be expressed as a deduction rom gross

production so as to calculate the contribution o expected output at age s to recovering the

original capital cost o building the plant hese figures have to be discounted back to the date

o commissioning in order to assess the net present value o the initial investment in the plant

he curves or constant and observed perormance using a discount rate o 9 (a typical cost o

capital or wind generation) are shown in Figure 5 he black line is based on the ull set o age

8 he initial load actors are calculated as the normalised load actor or UKDanish wind arms at age 0 multiplied by

where is the unit fixed effect or installation i

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1852

16 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

effects while the red dashed line is based on the smoothed quadratic perormance curve and

the blue dashed line shows what the pattern would be i operating perormance was not affected

by age he calculation is extended out to age 25 since DECCrsquos standard comparisons assume

that wind turbines have a lie o 25 years

Figure 5 Impact of performance degradation on discounted cumulative output

0

10

20

30

40

50

60

70

80

90

100

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

o

f d i s c o u n t e d c u m u l a t i v e o u t p u t

Age of plant (years)

Constant Quadratic Full age effects

Source Authorrsquos estimates

21 he key points in the figure are that 80 o the discounted cumulative output o a new wind

plant is likely to be produced in the first 10 years o its lie and 90 in the first 14 years his

is consistent with the structure o Danish subsidies or onshore wind arms which extend over

a typical period o 125 years Since sites or wind arms are scarce and involve the payment o

significant rents that may be linked to output it is very unlikely that any new installation o

wind turbines will have an expected lie o more than 15 years Instead as has happened in the

past wind operators will have a strong incentive to decommission plants afer no more than 15

years and replace the turbines with newer equipment

22 As a consequence any economic assessment o wind generation should not be based on an

expected lie which is longer than 15 years In recent work reported in evidence to the House

o Commons Select Committee on Energy and Climate Change I assumed that wind plants

would have a residual value equal to 20 o their initial cost in real terms at the end o 15 years

he analysis in this paper suggests that this is too avourable an assumption Given the costs o

decommissioning old turbines the residual value is likely to be well below 10 o their initial

cost and the decision point may be at 10 rather than 15 years

23 he corollary o this observation is that it makes little sense to offer long term contracts or 20

years or more that guarantee prices or eed-in tariffs (Fis) to wind operators A contract length

o 10 to 12 years would be sufficient to remove most o the market risk associated with invest-

ment in wind generation In this respect the subsidy arrangements implemented in Denmark

are better designed

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 17

24 At the same time the offer o subsidies andor guaranteed prices may have serious adverse

consequences or the efficiency o wind generation Returning to the unit fixed effects measur-

ing the perormance o wind plants commissioned afer 2005 only 28 out o 159 units have an

operating perormance that exceeds the average or the period hose 28 units account or 360

MW o capacity whereas the remaining 131 units account or 2810 MW o capacity Not onlyare recent plants less efficient than the average or the whole period but the plants which are

below-average in efficiency are typically larger than the more efficient ones and account or a

disproportionate share o the recent additions to generating capacity

25 In mid-2012 there was approximately 4500 MW o onshore wind capacity operating in the

UK By 2020 it is expected that there will be at least 10000 MW (based on proposed wind

arms with planning consent) and perhaps as much 15000 MW o wind capacity in operation

he government is relying upon a substantial contribution rom onshore wind arms towards

meeting its targets or renewable energy in 2020 o provide context the governmentrsquos targets

or renewable energy in 2020 speciy a total o 234 Wh o energy rom all renewable sources

o which 90ndash140 Wh is expected to come rom wind and biomass electricity generation9

he projections or biomass electricity are extremely vague and have to be reconciled with

an assumed increase o 3ndash4 times in the amount o biomass used separately or heat So in

concrete terms it is reasonable to assume that wind generation will account or at least 90 Wh

with 24ndash32 Wh rom onshore generation and the remainder rom offshore

26 he average load actor or onshore wind over the 10 year period rom April 2002 to March

2012 was 256 which is influenced by the large amounts o new capacity added rom 2005

onwards I this load actor were achieved in 2020 the total amount o onshore wind capacityrequired to meet the onshore generation target would be 107 to 143 GW he average load

actor or offshore wind installations (on an unchanged configuration basis) or the five year

period 2007ndash11 was 32010 I this load actor were achieved in 2020 the total amount o

offshore wind capacity required to meet the remainder o the 90 Wh target would be 206

to 235 GW though these estimates would all to 180 to 205 GW i the average load actor

matched the historic average value o 367 or Danish offshore wind installations

27 he question that has to be addressed is whether these load actors are consistent with (a) the

investment programmes that would be required to achieve the total amounts o generation

capacity implied by the governmentrsquos targets and (b) the age profiles o the perormance o

wind arms discussed in this paper he short answer is that this is simply not possible i the

initial load actor or new wind arms continues to decline at the rate observed the past decade

So even as a starting point it is necessary to assume that this underlying deterioration in wind

arm perormance ceases his would be a major change in itsel and it is not obvious what

actors might lead to such a result

28 he results illustrated below are based on a vintage model o perormance that takes account

o the actual expansion o capacity up to the end o 2011 plus whatever investment in new or

replacement turbines is required to achieve the target levels o wind capacity in 2020 o high-light the ull range o potential outcomes these targets are either 10 GW or 15 GW o onshore

9 Department o Energy and Climate Change ndash UK Renewable Energy Road Map July 2011 Figure 2

10 he equivalent average load actor or onshore wind arms was 264

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2052

18 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

wind and either 18 or 24 GW o offshore wind he levels o investment required to meet these

targets depend upon the expected lie o wind turbines For onshore turbines an expected lie

o 10 years is consistent with re-powering decisions over the last decade while an expected lie

o 15 years seems to be the maximum consistent with the age-perormance profile For offshore

turbines the analysis uses expected lives o 15 and 20 years

Figure 6A Projected load factors for UK onshore wind (total capacity of 10 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

Figure 6B Projected load factors for UK onshore wind (total capacity of 15 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

29 Figure 6A shows projected values o the average normalised load actors or UK onshore wind

arms in 2015 and 2020 using a target capacity in 2020 equal to 10 GW plus the age-peror-

mance profiles or the UK and Denmark (DK) Since vintage effects are important in this

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 19

calculation the profile o new or replacement investment affects the results In this case the

level o gross investment is assumed to be constant rom 2013 onwards at either 930 or 1220

MW per year ndash the range is defined by the assumed lie o turbines Achieving a target o 15

GW in 2020 (Figure 6B) would require a level o investment rom 2013 onwards o either 1540

or 1930 MW per year Such figures are 3ndash4 times the average level o investment over theperiod 2005ndash11 so the lower target may provide a better indication o what will happen

30 All o the estimates are calibrated to match the observed average load actor in 2007ndash11 so that

they should be interpreted as relative rather than absolute values Using the age-perormance

profile or UK onshore wind installations the average normalised load actor is likely to all by

about 19 rom 2007ndash11 to 2015 and by 23ndash28 to 2020 I the age-perormance curves or

Danish onshore installations were to apply the reduction in the average normalised load actor

would be considerably less but there would still be a all o up to 5 by 2020

31 In the medium term the all is greater i turbines operate or 15 years because the average age

o all turbines will be higher and thus the degradation in perormance due to age will have a

larger effect By 2020 this effect reduces the average normalised load actor rom 203 or a 10

year lie to 195 or a 15 year lie here is a complex economic and commercial trade-off here

o meet a target o 10 GW in 2020 it is necessary to increase the level o investment in onshore

capacity by 30 rom 2013 Almost all o this will take the orm o re-powering existing wind

arms11 he costs are airly high because this will usually involve new development consents

decommissioning the original turbines and replacing them with a smaller number o larger and

more powerul turbines

32 Estimates o re-powering costs are generally treated as being commercially sensitive but broadproject costs cited in specialist magazines suggest that the unit costs are likely to be at least

pound1000 per kW or more than pound300 million per year in aggregate On the other hand the increase

in the amount o energy generated would be about 12 Wh or a 10 year turbine lie rather

than a 15 year turbine lie though this does not allow or the loss o production during re-pow-

ering Under the best circumstances the payback period may be quite short but the best option

rom a broader economic and social perspective is not obvious

33 Even under the best scenario 10 GW o onshore wind capacity will only generate 178 Wh o

electricity in 2020 well below the lower end o the target range or onshore wind generation

specified in DECCrsquos road map In act allowing or interactions between the age-perormance

profile and investment the amount o onshore wind capacity required in 2020 to generate 24

Wh would be a minimum o 134 GW At the top end o DECCrsquos target range the capacity

required in 2020 to generate 32 Wh would be a minimum o 178 GW Since these estimates

are based on a 10 year turbine lie the level o investment required rom 2013 onwards would

be 1650 MW per year or the bottom o the range and 2350 MW or the top end o the range

In the last 8 years total investment in onshore wind has only ever exceeded 600 MW in one

11 Re-powering is usually defined as the ull replacement o existing turbines whereas retrofitting covers the replace-

ment o various mechanical or electrical components In the data analysis re-powering involves the creation o a

new unit whereas retrofitting affects the perormance o a continuing unit Hence the potential benefits o retrofit-

ting are taken into account in constructing the age-perormance curves

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2252

20 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

year (883 MW in 2008) so it seems rather unlikely that even the lower end o DECCrsquos range or

onshore wind generation will actually be realised

34 O course this assessment would be different i UK onshore wind operators were able to match

the Danish age-perormance curve here are many actors which differentiate the UK and

Danish wind industries he average scale o wind arms in the UK is larger they are moreremote and there appears to be significantly greater variation between operating and potential

new sites in achieved perormance It would thereore require a very optimistic observer to

conclude that the difference in age-perormance curves between the two countries will disap-

pear in the short or medium term

Figure 7 Projected load factors for UK offshore wind in 2020

20

22

24

26

28

30

32

34

18 GW ndash life 15 yrs 18 GW ndash life 20 yrs 24 GW ndash life 15 yrs 24 GW ndash life 20 yrs

L o a d f a c t o r ( )

2007ndash11 2015 2020

Note Based on Danish age-perormance curves Source Authorrsquos estimates

35 Figure 7 shows the results rom carrying out a similar exercise or offshore wind arms in the

UK he only evidence on the age-perormance curves or offshore installations is based upon

the Danish experience so these have been used to construct the projections Further the profile

o investment assumes a progressive increase in additions to capacity rom 2013 to 2016 and

then a stable level o investment rom 2017 to 2020 he DECC road map involves a very large

commitment to offshore wind investment ndash in the range 25-35 GW o new capacity annually

rom 2016 onwards As a consequence the average age o offshore capacity in 2020 will be quite

low and the ull effects o the age-perormance curves will only be elt during the decade ollow-

ing 2020 Even so the average normalised load actor in 2020 will be about 15 lower than or

2007ndash11 while by 2025 it will be about 30 below the base level

36 On these projections 232 GW o offshore capacity is the minimum necessary to generate 58

Wh o electricity rom offshore wind in 2020 his requirement will increase to 298 GW by

2025 simply to offset the effect o the ageing o offshore generation capacity on total output

37 Combining the estimates or onshore and offshore wind generation the total amount o wind

capacity required in 2020 to produce 90 Wh o electricity rom wind will be in the range 39

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 21

to 41 GW once allowance is made or the observed age-perormance curves or both onshore

and offshore installations hese estimates are roughly a third higher than official predictions o

28ndash31 GW o wind capacity in 2020

38 urning the figures round gives a slightly different picture o the reality o the governmentrsquos

strategy on renewable electricity generation A total o 30 GW o wind capacity in 2020 (12GW onshore 18 GW offshore) will generate about 64 Wh o electricity in a normal year

Since electricity generation rom wind and biomass are expected to account or 50 o renew-

able energy in 2020 the consequence is that biomass is expected to account or 53 Wh In

2011 landfill gas and sewage sludge digestion accounted or 44 o electricity rom bioenergy

(DUKES ables 64) here are limited opportunities to expand production rom these sources

so a continuation o historic rates o growth would mean that renewable generation rom other

sources o bioenergy would have to increase rom 72 Wh in 2011 to 46 Wh in 2020 here

are severe constraints on increasing the amounts o electricity generated rom animal biomass

biodegradable municipal solid waste and anaerobic digestion so it would be optimistic to

assume that these will contribute more than 6-8 Wh in 2020 (up rom 26 Wh in 2011)

39 In practice i the projections or renewable energy are to be believed they rest upon an assump-

tion that there will be a large increase in the amount o electricity that is generated rom plant

biomass ndash either on its own or through co-firing with coal his increase will involve increas-

ing the amount o electricity generated in these ways rom 46 Wh in 2011 to 38-40 Wh in

2020 Since the amounts o straw and UK-grown timber available or this purpose are strictly

limited this will come down to the amount o wood chips that can be imported and the cost o

doing this40 he historic average load actor or biomass plants is just over 60 so about 7500 MW o

biomass capacity will be required to supply 40 Wh o electricity he weight o wood chips per

MWh o electricity depends upon moisture content storage boiler design and other actors but

an indicative range is 085 to 115 tonnes per MWh o wood chips with a moisture content o

30 According to DECC and Forestry Commission estimates the UK imported 265 million

tonnes o steam coal and 26 million tonnes o other wood products (including wood chips)

in 2011 It seems that the net effect o the governmentrsquos strategy will be to replace 15 million

tonnes o coal imported in 2011 by 32 to 44 million tonnes o imported wood chips his is a

somewhat strange way o increasing energy security

Conclusion

41 Wind power is a highly capital-intensive technology or generating electricity Its merits rely

entirely upon a substitution o capital or uel inputs he same is true or hydro or tidal or wave

power In comparison with hydro power wind is a low quality resource because o its variability

and because it cannot be stored So the economic case or wind power must rest on obtaining

the most out o the wind that is available

42 While the decline in the achieved perormance o onshore wind turbines in Denmark is

much less than that or the UK or offshore nonetheless the decline in expected output under

standardised wind conditions over 10 years is 10 unweighted and 13 capacity weighted

hese declines accelerate afer age 10 so that the reductions in perormance are 17 and 20

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2452

22 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

respectively afer 15 years For UK onshore wind arms the reduction in perormance due to

age is much worse at 27 unweighted and 69 capacity weighted by age 10

43 Evidence on the perormance o Danish offshore installations is both restricted and so poor

that there may be concern that the results are affected by a small number o outliers Still the

sample contains a reasonable number o sites with at least 5 years o operating experience andthe decline in perormance by age 5 is 38 unweighted and 26 capacity weighted

44 In addition to these results there is strong evidence that the average normalised load actor or

new onshore wind installations in the UK has allen significantly over the period rom 2000 to

2011 his is consistent with a pattern in which the most avourable sites are developed first

Equally it could mean that wind developers have been unable to keep up with the rate o new

investment while maintaining the quality o development and operations For example the site

design or selection o turbine characteristics may make less effective use o the available wind

resources or the sites available than was the case in the past

45 Whatever the reasons the deterioration in initial perormance means that the expected returns

rom the expansion in wind capacity both or investors and in terms o the reduction in CO983090

emissions have been alling without a concomitant decrease in the private and social costs

that are borne by customers and the general public Clearly this is unsatisactory at best and it

suggests that the benefits claimed or current policies cannot be taken at ace value

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 23

Appendix Data and Methods

A Data for the United Kingdom

he raw data used or this study was extracted by the Renewable Energy Foundation rom the

Renewables and CHP Register database compiled by Ogem12 his inormation is used in the

administration o the market in Renewable Obligation Certificates (ROCs) he Renewables

Obligation is the primary mechanism by which large scale generators o renewable electric-

ity receive subsidies so the operators o wind arms have a very strong incentive to submit

complete and up-to-date inormation to the database he data extracted covered all onshore

wind generators with at least 05 MW o generating capacity which are eligible or ROCs or the

10 year period April 2002 to March 2012 he key variables are the monthly output o electric-

ity (used or the purpose o allocating ROCs) and nominal generating capacity here are 282

reporting units in the final dataset A reporting unit may correspond to an entire wind arm orto separate phases o development Complications arise when an existing wind arm is re-pow-

ered ie when old wind turbines are replaced by newer and usually more powerul turbines

he database reports this as a change in the generating capacity o the recording unit but in

this analysis repowering is treated as the termination o the old recording unit and the creation

o a new recording unit

A considerable amount o data checking and cleaning was required beore the data could

be analysed he first step was to add the age o each reporting unit his was calculated by

reerence to the month in which the turbines were commissioned For wind arms in operation

in or beore June 2002 the commissioning date was obtained rom various online sources ndash

notably the thewindpowernet database o UK wind arms13 which was cross-checked against

the inormation provided on the websites o wind operators and other sources For recording

units whose first data relates to dates afer June 2002 the commissioning date was initially

assumed to be the month or which data is first reported Cross-checks were carried out or

all recording units which commenced reporting in the period July 2002 to December 2003 In

more than 80 o cases the externally-reported commissioning date is within one month o the

first month or which data is reported he exceptions ndash Bu Farm and Mablethorpe ndash appear to

be due to delays by small operators in reporting data to Ogem

In a number o cases it is clear rom the data that plants were not in ull operation at the

time when data was first reported to the database here are two indicators o partial or delayed

commissioning

bull For an initial series o months the calculated load actor is very low ndash usually below 5 ndash

and then increases abruptly In such cases the data or the early months is dropped and the

commissioning date is treated as the month immediately preceding the first month with a

lsquonormalrsquo load actor14

12 httpswwwrenewablesandchpogemgovuk

13 See httpwwwthewindpowernetwindarms_list_enphp

14 he month in which a plant is reported as having been commissioned is not included in the analysis

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2652

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 25

the decline in the plantrsquos load actor as it ages As an example consider a plant whose reported

capacity was increased rom 99 MW to 10 MW at age 3 I this was simply a change in report-

ing practice this would reduce the estimated load actor rom age 3 onwards By increasing

the initial capacity rom 99 MW to 10 MW ndash or by reducing the capacity rom 10 MW to 99

MW rom age 3 onwards ndash this bias is removed It is not important which adjustment is madebecause the difference between the alternative assumptions is captured by the unit fixed effect

he numbers o separate sites and o observations per site mean that it is not possible to

provide a simple visual representation o the data Figure 8 provides a summary by illustrating

the distributions o load actors by the age o the wind arm pooled across all years he ranges

o the load actors or each age are wide while the medians show no clear trend by age hus a

naiumlve analysis might conclude that there is nothing to investigate However as will be explained

below the distributions mask crucial differences between the perormance o wind arms as

they age because they do not control or the differences in location and wind availability over

time

B Data for Denmark

he data or Danish wind arms used in this study comes rom a database compiled by the

Danish Energy Agency covering the characteristics and perormance o all wind turbines rom

2002 up to the end o August 201216 he basic recording unit or the register is the individual

wind turbine but where a number o turbines are connected to a single meter an identical

average output per turbine is recorded or all turbines in the group he presence o what are in

effect repeated observations increases the noise in the data In addition the model that is used

or the analysis assumes that site operator and turbine characteristics are common or all o theunits that make up a single wind arm o address this point separate wind arms have been

identified by grouping all turbines which have identical values or turbine manuacturer and

type local authority date o commissioning and date o decommissioning (where applicable)

he register contains inormation on 7569 turbines commissioned rom 1977 onwards his

includes a large number o small turbines which would not all within the scope o the ROC

scheme in the UK Hence or comparability with the UK data all turbines commissioned

prior to 1992 were dropped as well as all wind arms with a generating capacity o less than 1

MW he resulting sample covered 823 onshore wind arms and 30 offshore wind arms with

monthly load actors or various periods rom January 2002 to August 2012 Given the cut-off

or wind arms commissioned prior to 1992 the maximum age or an onshore wind arm was 20

years and the maximum age or an offshore wind arm was 17 years

he total capacity o the onshore wind arms recorded in the Danish database was 2757 MW

he average size o an onshore wind arm in Denmark is small ndash only 33 MW as compared

with 161 MW or the UK ndash because 84 o the sample consists o installations with between

1 and 4 turbines he largest onshore installation has 39 turbines with a capacity o 234 MW

while the onshore installation with the greatest capacity has 19 turbines with a capacity o 584

MW Overall the sample o Danish onshore wind arms is older and smaller than the sample

o UK onshore wind arms reflecting the much longer history o wind generation in Denmark

16 httpwwwensdkEN-USINFOFACSANDFIGURESENERGY_SAISICS_AND_INDICAORS

OVERVIEWOFHEENERGYSECORREGISEROFWINDURBINESSiderForsideaspx

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2852

26 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

he larger scale o UK wind arms may account or some o the differences in perormance

curves or the two countries

he sample o offshore wind arms in Denmark is quite small with 30 installations in

total Still it is useul to analyse the perormance o these installations since there is very little

comparable data or the UK he sample covers 394 turbines with an average o just over 13turbines per installation and an average capacity o 288 MW he largest installation which

was commissioned in 2003 has 72 turbines with a total capacity o 1656 MW Again offshore

installations are smaller and older in Denmark than is the current pattern in the UK he first

two wind arms in the sample were commissioned in the years 1995 and 2000 but the sample

increases significantly rom 2002 onwards o avoid reliance upon a sample o 1 or 2 wind

arms to determine specific coefficients or ages gt 10 years all observations with an age greater

than 10 were omitted rom the analysis17

A somewhat different problem with the offshore wind arm data arose because the coeffi-

cients estimated using capacity weights (see below) were heavily influenced by the largest windarm his has had an extremely erratic profile o perormance over time with a very low average

load actor he treatment o this wind arm is discussed in more detail in Section F below

C Specication and estimation methods

As Figure 8 illustrates it is necessary to go beyond simple summary statistics in order to assess

whether there is any systematic relationship between age and the average perormance o wind

arms he perormance o wind arms varies over both time and space some months and years

have more or less wind than the long term average while specific characteristics o wind arms

ndash including location type o turbine and operating regime ndash will influence the perormance oeach plant under identical wind conditions

he combination o (a) site-specific characteristics which are constant over time and (b)

period-specific characteristics which are constant across wind arms can be represented by

what statisticians call an error components model with fixed effects or each wind arm and

each time period he dependent variable in the model consists o a sequence o load actors ndash

denoted by or unit i in month t ndash or each wind arm (the panel unit) over time he load

actors are calculated by dividing the total output in MWh by (24 x number o days in month x

reported generating capacity in MW) and are multiplied by 100 to convert to percentages

he additive version o the error components model may be written as

(1)

in which denotes the age o plant i in period t and is some unction that is either a

representation o or an approximation to the normalised age-perormance relationship he

multiplicative version is similar except that is replaced by where ln() denotes the

natural logarithm he terms are wind arm or unit fixed effects which capture actors such

as location type o turbine operating regimes etc he terms are period fixed effects which

capture the influence o wind conditions over the UK or Denmark as a whole as well as seasonal

maintenance and demand Without loss o generality it may be assumed that the fixed effects are

17 For the avoidance o doubt this means that no attempt was made to estimate the effect o age on the perormance o

offshore wind arms beyond age 10 while the data or the oldest wind arms was included in the analysis up to and

including age 10

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 27

normalised so that ie all o the error components have a zero mean while the

mean o the random error is zero by definition he terms are random errors which capture

random variations in wind conditions that are not specific to the site turbine breakdowns and

other actors that are uncorrelated with either the site or the time period It is assumed that the

mean o the random error is zero while additional assumptions depend upon the method o

estimation that is adopted

he model specified in equation (1) can be estimated in a variety o ways18 A crucial issue is

whether the are or may be significantly correlated with the age terms or any other independent

regressors I it is assumed that there is no such correlation the model is usually reerred to as

a random effects model In that case the equation can be estimated by a variant o least squares

(LS) generalised least squares (GLS) generalised methods o moments (GMM) or maximum

likelihood (ML) ndash all o which should be yield parameter estimates that will converge in prob-

ability to the true parameter values as the sample size increases An alternative but less restric-

tive specification is known as the fixed effects model in which no assumption is made about

the correlation between the unit fixed effects and other regressors he limitation o the fixed

effects model is that it cannot be used to estimate parameters or regressors that vary across

panel units but are constant over time or a panel unit (time-invariant regressors) ndash eg the

year in which a wind arm was commissioned its location or its rated generating capacity he

influence o such variables is captured by the unit fixed effects

In order to avoid the risk o obtaining biased estimates o the age effects the results reported

in this note are based upon estimates using the fixed effects model he influence o time-invar-

iant regressors is examined in a second stage by regressing the estimates o the unit fixed effectson these variables he simplest method o estimating the fixed effects model is known as the

lsquowithinrsquo estimator in which the mean value or each unit is subtracted rom all observations or

that unit For example i is a general polynomial o order M in A equation (1) becomes

(2)

so that the model can be estimated by least square with the inclusion o dummy variables or

each period t he variance-covariance matrix o the coefficients have been estimated by using

(a) a robust sandwich estimator adjusted or clustering by panel unit and (b) a bootstrap esti-

mator with 400 repetitions he estimates o the standard errors will be consistent even i the

random error component is serially correlated within panel units andor has different variances

across panel units19 Since the number o panel units and the average number o time periods

per panel units are both large (other than or Danish offshore wind arms) the least square esti-

mates o the parameters will be unbiased under these assumptions

he bootstrap method is expensive to carry out but it provides a useul cross-check on

the standard errors generated by other methods Unortunately the bootstrap method o

18 See Chapter 21 o AC Cameron amp PK rivedi ndash Microeconometrics Methods and Applications (Cambridge

Cambridge University Press 2005) or a standard textbook treatment o estimation methods or panel data models

19 he estimation was carried out using the xtreg procedure in Stata Version 12 he methods o constructing the

sandwich and bootstrap estimates o the variance-covariance matrix are described in Chapters 8 amp 13 o AC

Cameron amp PK rivedi ndash Microeconometrics using Stata (Revised Edition College Station exas Stata Press 2010)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 10: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 9

The Performance of Wind Farms in the United Kingdom

and Denmark

Introduction

1 Any assessment o the costs o wind power must rely heavily upon assumptions about theaverage load actor that will be achieved by new wind installations over their lietime It is

standard practice to calculate average load actors by year and country or onshore and offshore

installations as shown in able 1 (page 40)1 However such estimates do not provide a

reliable statistical basis or assessing the uture perormance o wind arms in aggregate Part

o the reason is that the amount o wind in any month or year is influenced by long term mete-

orological cycles that have periods o many years notably the North Atlantic Oscillation In

addition average load actors do not allow or changes in the composition o wind installations

by location age size and other actors

2 As the number o wind arms operating in developed countries has grown engineers have an

opportunity to analyse operating experience over extended periods o time With any (relatively)

novel technology the incidence o temporary or permanent breakdowns may be expected to all

over time so it is difficult to disentangle the effects o age on operating perormance rom the

technological immaturity o turbines installed 10 or 20 years ago Nonetheless the technology

or onshore wind turbines has been reasonably mature since the early 2000s his is documented

by data produced by the US Department o Energy which shows that the decline in capital costs

that is characteristic o immature technologies slowed afer 2000 and was reversed afer 20042

3 With at least 10 years o operating data since 2000 it should be possible to examine how thetypical operating perormance o wind installations in the United Kingdom and Denmark

varies with the age o the turbines he estimated age-perormance curves should then be

incorporated in estimates o levelised costs which are ofen used to compare the costs o wind

and other orms o generation

4 he Appendix provides a detailed description o the data and methods employed or this study

In both the UK and Denmark wind operators have a large incentive to produce electricity

whenever there is sufficient wind available since marginal operating costs are small while the

operator receives a much higher price In the UK this price is the sum o the market price plus

the value o (a) the Renewable Obligation Certificates (ROCs) awarded or each MWh o elec-

tricity produced and (b) the associated exemption rom the Climate Change Levy In practice

the effective market value o ROCs and associated incentives means that an onshore wind

operator earns roughly double the average market price o electricity per MWh In Denmark

most onshore wind capacity receives a price premium equivalent to about pound28 per MWh on top

1 he load actor is calculated as the ratio o the amount o electricity actually produced by a turbine or wind arm

over a period o a month or a year divided by the amount o output that would have been produced had it operated

at ull nameplate capacity or the entire period his is expressed as a percentage so that reported load actors lie

between 0 and 100

2 R Wiser amp M Bolinger (2012) 2011 Wind Technologies Market Report Lawrence Berkeley National Laboratory

US Department o Energy It should be noted that the installed cost o wind plants has allen since 2010 but these

changes reflect the cyclical orces o demand and supply that are well established across the electricity sector and

which apply to ossil-uel as well as renewable generators

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1252

10 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

o the market price or about 125 years o operation at a typical load actor3 Offshore capacity

is eligible or a similar price premium but most o it is contracted via tenders that offer a guar-

anteed eed-in tariff he tender tariffs have varied rom pound56 per MWh to twice that figure

extending over 16-18 years o operation at a typical load actor

Age-performance curves

5 A standard indicator o the operating perormance o a wind turbine is its normalised load

actor defined as the total output over some period divided by the maximum potential output

adjusted or wind availability Normalisation or wind availability is not straightorward I

detailed data on wind speeds at specific locations is available then the potential output rom

individual turbines or wind arms can be calculated using standard power curves Since wind

speeds can vary substantially over small distances such calculations are both difficult to imple-

ment and prone to errors when applied to large numbers o wind generators using public

sources o data4

6 Many analyses work in the opposite direction For example the Danish Wind Index ndash the

longest running series o data on wind availability in Europe ndash is constructed rom the common

component o monthly variations in wind output rom a large sample o wind turbines5 his

study adopts a similar approach Wind availability is treated as an unobserved actor whose

contribution to monthly variations in wind output is estimated by a series o fixed period

(monthyear) effects in a statistical model o monthly output or all wind arms he reasons

or adopting this approach are discussed in Section D o the Appendix his explains why the

method must perorm better than the easible alternative o using a single index o average

wind speed or the UK or Denmark

7 he analysis relies upon standard techniques that are widely used in the biological and medical

sciences as well economics and engineering Full details are given in the Appendix he idea is

to treat the actual load actor (ie without adjustment or wind conditions) or a specific wind

generator in any time period as being determined by a set o components which reflect specific

site conditions and the age-related perormance o wind plants in general together with the

period fixed effects and an uncorrelated random error he onshore wind datasets or the UK

and Denmark used or the analysis are large with monthly observations on 282 installations in

the UK and 823 installations in Denmark with an age range rom 0 to 19 years he offshore

wind dataset or Denmark is rather smaller covering only 30 installations but it can be used to

obtain reasonable estimates o perormance up to age 10 he results discussed here are based

upon least squares estimation (see page 27) combined with standard errors that are robust to

various departures rom classical assumptions about the nature o the data generation process

8 he results o the statistical analysis demonstrate an unambiguous and statistically significant

decline in the operating perormance o wind arms as they grow older Figure 1 shows the

3 Conversions to GBP are based on an exchange rate o pound1 = DKK 918 (as at 19 October 2012)

4 All wind generators collect their own inormation on wind speeds in order to manage their plants but they do not

publish such inormation and will not have access to inormation collected by other operators

5 Details o the construction o the Danish Wind Index are given at wwwvindstatdk A paper by Boccard ndash N Boccard

(2009) lsquoCapacity actor o wind power realized values vs estimatesrsquo Energy Policy Vol 37 pp 2679-88 ndash includes a

discussion o the construction o similar indices o wind availability in Germany the UK Netherlands and Sweden

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 11

simplest variant o the normalised age-perormance curve or onshore UK wind installations

together with the equivalent curves or onshore and offshore Danish installations he curves

illustrated are calculated rom the multiplicative error components model described in the

Appendix with log(load actor) as the dependent variable and a quadratic in plant age to repre-

sent the variation in plant perormance with age9 he rate o decline in perormance is greatest or offshore installations in Denmark with a all

rom load actors o over 40 at ages 0 and 1 to less than 15 by at ages 9 and 10 Onshore

installations in the UK show a more rapid rate o decline ndash 09 percentage points per year over

the first 10 years o operation ndash than is the case or Denmark though the normalised Danish

perormance curve lies below the UK curve or the first our years For the UK the normalised

load actor alls to just over 15 at age 10 and to 11 at age 15 With such low load actors it

seems likely that many wind arms will be re-powered ndash ie the turbines will be replaced ndash once

they reach the age o 10 or at most 15

Figure 1 Performance degradation due to age using equal weights

0

5

10

15

20

25

30

35

40

45

50

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

L o

a d f a c t o r ( )

Age (years)

UK onshore DK onshore DK offshore

Note Normalised load actors in Source Authorrsquos estimates

10 here are two plausible explanations or the observed decline in average load actors as wind

arms age he first is that the turbines become less efficient over time as a result o mechanical

wear and tear erosion o the turbine blades and related actors he second is that the turbines

experience more requent breakdowns and their operators take more time to bring them back

into service because they are less concerned about the perormance o older plants Both

reasons may be relevant in different circumstances and it is not possible to identiy a primary

explanation rom the data he requency o extended shutdowns does seem to increase with

age but this could be a reflection o the timing o planned maintenance operations rather than

breakdowns Whatever the cause the reduction in perormance with age is much greater than

would be expected or thermal generating plants

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1452

12 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

11 he age-perormance curves in Figure 1 are derived rom equations estimated by giving equal

weight to each wind arm irrespective o its generating capacity Figure 2 illustrates similar

curves but in this case the estimates are constructed by weighting each wind arm by its generat-

ing capacity reerred to as capacity-weighted his gives a better representation o perormance

degradation per MW o generating capacity he striking result is that the rate o decline inthe perormance o UK onshore wind installations is significantly aster when capacity weights

are used which implies that large wind arms in the UK experience a more rapid decline than

smaller ones he differences are smaller in Denmark but the capacity-weighted decline in

perormance is more rapid than the equal weights decline or Danish onshore installations

Figure 2 Performance degradation due to age using capacity weights

0

5

10

15

20

25

30

35

40

45

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

L o a d f a c t o r ( )

Age (years)

UK onshore DK onshore DK offshore

Note Normalised load actors in Source Authorrsquos estimates

Differences between wind farms

12 he specification o the statistical model means that the unit fixed effects identiy the specific

perormance characteristics o individual wind arms afer adjusting or wind availability and

age hese characteristics may reflect the site andor design o the installation as well as the

way in which it is operated A positive value or the unit fixed effect is equivalent to shifing

the perormance curve up by some percentage applied uniormly at each age while a negative

value shifs it down It ollows that the distribution o unit fixed effects across installations can

be used to compare the relative effectiveness o new installations by location date o commis-

sioning etc Care is required when making comparisons between or example UK and Danish

onshore wind installations because these have been normalised separately so the unit fixed

effects reer to different underlying averages6

6 Some care is required in interpreting the unit fixed effects By deault the average value o the unit fixed effects or

each group is zero but these averages are taken over all observations or each installation in the basic sample For the

purpose o the comparisons in Figure 3 the unit fixed effects have been adjusted so that the average values or each

group are zero when using one value or each installation

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 13

13 o give a sense o the numbers unit fixed effects o (a) 02 or (b) -02 mean that the base load

actor given the age o the plant is multiplied by (a) exp(02) = 122 or (b) exp(-02) = 082 Since

the base load actor or a plant o age 0 is 231 these fixed effects translate to load actors in

the first year o operation o (a) 282 and (b) 189 For a wind arm with 50 MW o gener-

ating capacity the difference amounts to an additional 40700 MWh o electricity output peryear Using the standard figures cited by RenewableUK this would be sufficient to supply 12300

homes or a year his range (040) is slightly greater than the interquartile range o unit fixed

effects or UK onshore wind arms (035) Again or comparison operators o thermal power

plants would regard any difference o even 10 up or down in efficiency relative to what is

expected as a striking and perhaps worrying indication o either good or bad perormance

Figure 3 Range of unit fixed effects by category

- 1

- 5

0

5

1

A d j f i x e d e f f e c t ndash q u a d r a t i c

UK onshore DK onshore DK offshore

Source Authorrsquos estimates

14 A notable eature o the distributions o the unit fixed effects that are summarised in the box

plots shown in Figure 3 is the range o variation between the best and worst perormers espe-

cially or UK onshore and Danish offshore wind arms Box plots are constructed so that the

top and bottom o the box correspond to the upper and lower quartiles while the bars at topand bottom correspond to the maximum and minimum values excluding outliers7 A plant at

the upper quartile o the distribution o UK onshore wind arms will generate an annual output

that is about 40 higher than a plant at the lower quartile he equivalent figure or Danish

onshore wind plants is an increase o 30 while or Danish offshore wind arms the difference

between the upper and lower quartiles is equivalent to 25 times the output o a lower quartile

plant By any standard such differences are important

7 he upper quartile is the point in a distribution such that the 25 o values exceed it while the lower quartile is the

point in a distribution such that 75 o values exceed it he line in the middle o the box marks the median the

point exceeded by 50 o values Outliers are defined as values that lie outside the range o ukey adjacent values

which are defined as the upperlower quartiles +- 15 x interquartile range

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1652

14 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

15 Outside the range between the upper and lower quartiles the top and bottom segments o

the distributions extend urther up and down or UK onshore wind arms than or those in

Denmark he rather different pattern or Danish offshore wind installations is partly a conse-

quence o the relatively small size o the sample but the evidence points to considerably greater

variability in the perormance o offshore wind installations than or onshore installations16 It is somewhat surprising to observe the magnitude o the differences in the perormance in

onshore wind arms in the UK Given the nature o the subsidy regimes and the high capital

cost o developing new installations it might be expected that operators have a strong incen-

tive to identiy the best locations and then choose equipment that will deliver the maximum

amount o electricity output at a high level o reliability For UK onshore operators it seems that

good or bad perormance is somewhat o a lottery However i the subsidies provided by ROCs

are sufficient to underwrite investment in inefficient plants ndash as appears to be the case ndash then

those subsidies are extremely generous or plants that operate close to the efficient rontier

As location is likely to be the main actor that determines the perormance o a specific plant

relative to all other plants the inerence must be that many wind plants have been developed on

sites with poor wind characteristics

Figure 4 Initial load factors by year of commissioning for onshore wind

10

15

20

25

30

35

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

I n i t i a l l o a d f a c t o r ( )

Year commissioned

UK onshore DK onshore Linear (UK onshore) Linear (DK onshore)

Source Authorrsquos estimates

17 here is another disturbing characteristic o UK wind arms which is revealed by examination

o the unit fixed effects and illustrated in Figure 4 It might be expected that the average peror-

mance o recently installed units would be higher than that o older units his should reflect

improvements in turbine design and reliability as well as in the identification o good sites or

wind generation However that is not the pattern or onshore wind arms in the UK he figure

shows the average values o the initial load actors calculated using equally weighted data by

year o commissioning or onshore wind arms in the UK and Denmark together with fitted

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 15

trend lines8 here has been a clear decline in the initial load actors in the UK which has not

occurred in Denmark his pattern is confirmed by more detailed statistical results shown in

the Appendix which take account o location and size o installation

18 he decline in initial perormance is particularly obvious afer 2005 when the rate o commis-

sioning new capacity increased sharply rom less than 500 MW in the 5 years 2000ndash04 to morethan 500 MW per year rom 2005 onwards In contrast the peak years or commissioning new

onshore wind arms in Denmark were 2000ndash02 when 920 MW was installed he rate o new

building (or replacement) picked up again in 2008 but it is still much lower than at the turn o

century Not only is the perormance o the onshore wind plants in the UK commissioned in

recent years significantly worse than that or wind arms commissioned beore 2005 but it is

possible that this is a direct consequence o an overly rapid expansion o capacity In essence

the evidence suggests that the industry does not have the capability to identiy develop and

operate new onshore wind arms at the rate envisaged by UK government targets while main-

taining a satisactory average level o perormance

Implications for future policy and performance

19 he pattern o perormance degradation with age and over time identified in this analysis has

implications or the assessment o new investment in wind arms and or the design o the

proposed Electricity Market Reorm (EMR) here are two related issues that are affected by

the results

a) What are the implications or the design o policies intended to promote the adoption

and expansion o wind generation in the UK his includes the structure o current

subsidies as well as the specification o the contracts that are proposed under the EMR A

related question concerns the expected lie o new wind arms

b) How might the average perormance o wind generation develop over the next decade

his is absolutely critical to any assessment o the amount o wind capacity that will be

required to meet the UK governmentrsquos targets or the share o electricity ndash and more

generally energy ndash rom renewable sources he government has assumed that the

average load actors or both onshore and offshore wind arms will either remain stable

or increase in uture he prospects will be very different i these assumptions are not

borne out

20 he first question can be addressed by considering the discounted sum o cumulative net

output rom a new wind arm on the assumption that its perormance matches the averages

identified in the analysis his takes account o both perormance degradation over time plus

the annual cost o maintaining the plant which may be expressed as a deduction rom gross

production so as to calculate the contribution o expected output at age s to recovering the

original capital cost o building the plant hese figures have to be discounted back to the date

o commissioning in order to assess the net present value o the initial investment in the plant

he curves or constant and observed perormance using a discount rate o 9 (a typical cost o

capital or wind generation) are shown in Figure 5 he black line is based on the ull set o age

8 he initial load actors are calculated as the normalised load actor or UKDanish wind arms at age 0 multiplied by

where is the unit fixed effect or installation i

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1852

16 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

effects while the red dashed line is based on the smoothed quadratic perormance curve and

the blue dashed line shows what the pattern would be i operating perormance was not affected

by age he calculation is extended out to age 25 since DECCrsquos standard comparisons assume

that wind turbines have a lie o 25 years

Figure 5 Impact of performance degradation on discounted cumulative output

0

10

20

30

40

50

60

70

80

90

100

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

o

f d i s c o u n t e d c u m u l a t i v e o u t p u t

Age of plant (years)

Constant Quadratic Full age effects

Source Authorrsquos estimates

21 he key points in the figure are that 80 o the discounted cumulative output o a new wind

plant is likely to be produced in the first 10 years o its lie and 90 in the first 14 years his

is consistent with the structure o Danish subsidies or onshore wind arms which extend over

a typical period o 125 years Since sites or wind arms are scarce and involve the payment o

significant rents that may be linked to output it is very unlikely that any new installation o

wind turbines will have an expected lie o more than 15 years Instead as has happened in the

past wind operators will have a strong incentive to decommission plants afer no more than 15

years and replace the turbines with newer equipment

22 As a consequence any economic assessment o wind generation should not be based on an

expected lie which is longer than 15 years In recent work reported in evidence to the House

o Commons Select Committee on Energy and Climate Change I assumed that wind plants

would have a residual value equal to 20 o their initial cost in real terms at the end o 15 years

he analysis in this paper suggests that this is too avourable an assumption Given the costs o

decommissioning old turbines the residual value is likely to be well below 10 o their initial

cost and the decision point may be at 10 rather than 15 years

23 he corollary o this observation is that it makes little sense to offer long term contracts or 20

years or more that guarantee prices or eed-in tariffs (Fis) to wind operators A contract length

o 10 to 12 years would be sufficient to remove most o the market risk associated with invest-

ment in wind generation In this respect the subsidy arrangements implemented in Denmark

are better designed

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 17

24 At the same time the offer o subsidies andor guaranteed prices may have serious adverse

consequences or the efficiency o wind generation Returning to the unit fixed effects measur-

ing the perormance o wind plants commissioned afer 2005 only 28 out o 159 units have an

operating perormance that exceeds the average or the period hose 28 units account or 360

MW o capacity whereas the remaining 131 units account or 2810 MW o capacity Not onlyare recent plants less efficient than the average or the whole period but the plants which are

below-average in efficiency are typically larger than the more efficient ones and account or a

disproportionate share o the recent additions to generating capacity

25 In mid-2012 there was approximately 4500 MW o onshore wind capacity operating in the

UK By 2020 it is expected that there will be at least 10000 MW (based on proposed wind

arms with planning consent) and perhaps as much 15000 MW o wind capacity in operation

he government is relying upon a substantial contribution rom onshore wind arms towards

meeting its targets or renewable energy in 2020 o provide context the governmentrsquos targets

or renewable energy in 2020 speciy a total o 234 Wh o energy rom all renewable sources

o which 90ndash140 Wh is expected to come rom wind and biomass electricity generation9

he projections or biomass electricity are extremely vague and have to be reconciled with

an assumed increase o 3ndash4 times in the amount o biomass used separately or heat So in

concrete terms it is reasonable to assume that wind generation will account or at least 90 Wh

with 24ndash32 Wh rom onshore generation and the remainder rom offshore

26 he average load actor or onshore wind over the 10 year period rom April 2002 to March

2012 was 256 which is influenced by the large amounts o new capacity added rom 2005

onwards I this load actor were achieved in 2020 the total amount o onshore wind capacityrequired to meet the onshore generation target would be 107 to 143 GW he average load

actor or offshore wind installations (on an unchanged configuration basis) or the five year

period 2007ndash11 was 32010 I this load actor were achieved in 2020 the total amount o

offshore wind capacity required to meet the remainder o the 90 Wh target would be 206

to 235 GW though these estimates would all to 180 to 205 GW i the average load actor

matched the historic average value o 367 or Danish offshore wind installations

27 he question that has to be addressed is whether these load actors are consistent with (a) the

investment programmes that would be required to achieve the total amounts o generation

capacity implied by the governmentrsquos targets and (b) the age profiles o the perormance o

wind arms discussed in this paper he short answer is that this is simply not possible i the

initial load actor or new wind arms continues to decline at the rate observed the past decade

So even as a starting point it is necessary to assume that this underlying deterioration in wind

arm perormance ceases his would be a major change in itsel and it is not obvious what

actors might lead to such a result

28 he results illustrated below are based on a vintage model o perormance that takes account

o the actual expansion o capacity up to the end o 2011 plus whatever investment in new or

replacement turbines is required to achieve the target levels o wind capacity in 2020 o high-light the ull range o potential outcomes these targets are either 10 GW or 15 GW o onshore

9 Department o Energy and Climate Change ndash UK Renewable Energy Road Map July 2011 Figure 2

10 he equivalent average load actor or onshore wind arms was 264

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2052

18 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

wind and either 18 or 24 GW o offshore wind he levels o investment required to meet these

targets depend upon the expected lie o wind turbines For onshore turbines an expected lie

o 10 years is consistent with re-powering decisions over the last decade while an expected lie

o 15 years seems to be the maximum consistent with the age-perormance profile For offshore

turbines the analysis uses expected lives o 15 and 20 years

Figure 6A Projected load factors for UK onshore wind (total capacity of 10 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

Figure 6B Projected load factors for UK onshore wind (total capacity of 15 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

29 Figure 6A shows projected values o the average normalised load actors or UK onshore wind

arms in 2015 and 2020 using a target capacity in 2020 equal to 10 GW plus the age-peror-

mance profiles or the UK and Denmark (DK) Since vintage effects are important in this

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 19

calculation the profile o new or replacement investment affects the results In this case the

level o gross investment is assumed to be constant rom 2013 onwards at either 930 or 1220

MW per year ndash the range is defined by the assumed lie o turbines Achieving a target o 15

GW in 2020 (Figure 6B) would require a level o investment rom 2013 onwards o either 1540

or 1930 MW per year Such figures are 3ndash4 times the average level o investment over theperiod 2005ndash11 so the lower target may provide a better indication o what will happen

30 All o the estimates are calibrated to match the observed average load actor in 2007ndash11 so that

they should be interpreted as relative rather than absolute values Using the age-perormance

profile or UK onshore wind installations the average normalised load actor is likely to all by

about 19 rom 2007ndash11 to 2015 and by 23ndash28 to 2020 I the age-perormance curves or

Danish onshore installations were to apply the reduction in the average normalised load actor

would be considerably less but there would still be a all o up to 5 by 2020

31 In the medium term the all is greater i turbines operate or 15 years because the average age

o all turbines will be higher and thus the degradation in perormance due to age will have a

larger effect By 2020 this effect reduces the average normalised load actor rom 203 or a 10

year lie to 195 or a 15 year lie here is a complex economic and commercial trade-off here

o meet a target o 10 GW in 2020 it is necessary to increase the level o investment in onshore

capacity by 30 rom 2013 Almost all o this will take the orm o re-powering existing wind

arms11 he costs are airly high because this will usually involve new development consents

decommissioning the original turbines and replacing them with a smaller number o larger and

more powerul turbines

32 Estimates o re-powering costs are generally treated as being commercially sensitive but broadproject costs cited in specialist magazines suggest that the unit costs are likely to be at least

pound1000 per kW or more than pound300 million per year in aggregate On the other hand the increase

in the amount o energy generated would be about 12 Wh or a 10 year turbine lie rather

than a 15 year turbine lie though this does not allow or the loss o production during re-pow-

ering Under the best circumstances the payback period may be quite short but the best option

rom a broader economic and social perspective is not obvious

33 Even under the best scenario 10 GW o onshore wind capacity will only generate 178 Wh o

electricity in 2020 well below the lower end o the target range or onshore wind generation

specified in DECCrsquos road map In act allowing or interactions between the age-perormance

profile and investment the amount o onshore wind capacity required in 2020 to generate 24

Wh would be a minimum o 134 GW At the top end o DECCrsquos target range the capacity

required in 2020 to generate 32 Wh would be a minimum o 178 GW Since these estimates

are based on a 10 year turbine lie the level o investment required rom 2013 onwards would

be 1650 MW per year or the bottom o the range and 2350 MW or the top end o the range

In the last 8 years total investment in onshore wind has only ever exceeded 600 MW in one

11 Re-powering is usually defined as the ull replacement o existing turbines whereas retrofitting covers the replace-

ment o various mechanical or electrical components In the data analysis re-powering involves the creation o a

new unit whereas retrofitting affects the perormance o a continuing unit Hence the potential benefits o retrofit-

ting are taken into account in constructing the age-perormance curves

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2252

20 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

year (883 MW in 2008) so it seems rather unlikely that even the lower end o DECCrsquos range or

onshore wind generation will actually be realised

34 O course this assessment would be different i UK onshore wind operators were able to match

the Danish age-perormance curve here are many actors which differentiate the UK and

Danish wind industries he average scale o wind arms in the UK is larger they are moreremote and there appears to be significantly greater variation between operating and potential

new sites in achieved perormance It would thereore require a very optimistic observer to

conclude that the difference in age-perormance curves between the two countries will disap-

pear in the short or medium term

Figure 7 Projected load factors for UK offshore wind in 2020

20

22

24

26

28

30

32

34

18 GW ndash life 15 yrs 18 GW ndash life 20 yrs 24 GW ndash life 15 yrs 24 GW ndash life 20 yrs

L o a d f a c t o r ( )

2007ndash11 2015 2020

Note Based on Danish age-perormance curves Source Authorrsquos estimates

35 Figure 7 shows the results rom carrying out a similar exercise or offshore wind arms in the

UK he only evidence on the age-perormance curves or offshore installations is based upon

the Danish experience so these have been used to construct the projections Further the profile

o investment assumes a progressive increase in additions to capacity rom 2013 to 2016 and

then a stable level o investment rom 2017 to 2020 he DECC road map involves a very large

commitment to offshore wind investment ndash in the range 25-35 GW o new capacity annually

rom 2016 onwards As a consequence the average age o offshore capacity in 2020 will be quite

low and the ull effects o the age-perormance curves will only be elt during the decade ollow-

ing 2020 Even so the average normalised load actor in 2020 will be about 15 lower than or

2007ndash11 while by 2025 it will be about 30 below the base level

36 On these projections 232 GW o offshore capacity is the minimum necessary to generate 58

Wh o electricity rom offshore wind in 2020 his requirement will increase to 298 GW by

2025 simply to offset the effect o the ageing o offshore generation capacity on total output

37 Combining the estimates or onshore and offshore wind generation the total amount o wind

capacity required in 2020 to produce 90 Wh o electricity rom wind will be in the range 39

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 21

to 41 GW once allowance is made or the observed age-perormance curves or both onshore

and offshore installations hese estimates are roughly a third higher than official predictions o

28ndash31 GW o wind capacity in 2020

38 urning the figures round gives a slightly different picture o the reality o the governmentrsquos

strategy on renewable electricity generation A total o 30 GW o wind capacity in 2020 (12GW onshore 18 GW offshore) will generate about 64 Wh o electricity in a normal year

Since electricity generation rom wind and biomass are expected to account or 50 o renew-

able energy in 2020 the consequence is that biomass is expected to account or 53 Wh In

2011 landfill gas and sewage sludge digestion accounted or 44 o electricity rom bioenergy

(DUKES ables 64) here are limited opportunities to expand production rom these sources

so a continuation o historic rates o growth would mean that renewable generation rom other

sources o bioenergy would have to increase rom 72 Wh in 2011 to 46 Wh in 2020 here

are severe constraints on increasing the amounts o electricity generated rom animal biomass

biodegradable municipal solid waste and anaerobic digestion so it would be optimistic to

assume that these will contribute more than 6-8 Wh in 2020 (up rom 26 Wh in 2011)

39 In practice i the projections or renewable energy are to be believed they rest upon an assump-

tion that there will be a large increase in the amount o electricity that is generated rom plant

biomass ndash either on its own or through co-firing with coal his increase will involve increas-

ing the amount o electricity generated in these ways rom 46 Wh in 2011 to 38-40 Wh in

2020 Since the amounts o straw and UK-grown timber available or this purpose are strictly

limited this will come down to the amount o wood chips that can be imported and the cost o

doing this40 he historic average load actor or biomass plants is just over 60 so about 7500 MW o

biomass capacity will be required to supply 40 Wh o electricity he weight o wood chips per

MWh o electricity depends upon moisture content storage boiler design and other actors but

an indicative range is 085 to 115 tonnes per MWh o wood chips with a moisture content o

30 According to DECC and Forestry Commission estimates the UK imported 265 million

tonnes o steam coal and 26 million tonnes o other wood products (including wood chips)

in 2011 It seems that the net effect o the governmentrsquos strategy will be to replace 15 million

tonnes o coal imported in 2011 by 32 to 44 million tonnes o imported wood chips his is a

somewhat strange way o increasing energy security

Conclusion

41 Wind power is a highly capital-intensive technology or generating electricity Its merits rely

entirely upon a substitution o capital or uel inputs he same is true or hydro or tidal or wave

power In comparison with hydro power wind is a low quality resource because o its variability

and because it cannot be stored So the economic case or wind power must rest on obtaining

the most out o the wind that is available

42 While the decline in the achieved perormance o onshore wind turbines in Denmark is

much less than that or the UK or offshore nonetheless the decline in expected output under

standardised wind conditions over 10 years is 10 unweighted and 13 capacity weighted

hese declines accelerate afer age 10 so that the reductions in perormance are 17 and 20

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2452

22 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

respectively afer 15 years For UK onshore wind arms the reduction in perormance due to

age is much worse at 27 unweighted and 69 capacity weighted by age 10

43 Evidence on the perormance o Danish offshore installations is both restricted and so poor

that there may be concern that the results are affected by a small number o outliers Still the

sample contains a reasonable number o sites with at least 5 years o operating experience andthe decline in perormance by age 5 is 38 unweighted and 26 capacity weighted

44 In addition to these results there is strong evidence that the average normalised load actor or

new onshore wind installations in the UK has allen significantly over the period rom 2000 to

2011 his is consistent with a pattern in which the most avourable sites are developed first

Equally it could mean that wind developers have been unable to keep up with the rate o new

investment while maintaining the quality o development and operations For example the site

design or selection o turbine characteristics may make less effective use o the available wind

resources or the sites available than was the case in the past

45 Whatever the reasons the deterioration in initial perormance means that the expected returns

rom the expansion in wind capacity both or investors and in terms o the reduction in CO983090

emissions have been alling without a concomitant decrease in the private and social costs

that are borne by customers and the general public Clearly this is unsatisactory at best and it

suggests that the benefits claimed or current policies cannot be taken at ace value

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 23

Appendix Data and Methods

A Data for the United Kingdom

he raw data used or this study was extracted by the Renewable Energy Foundation rom the

Renewables and CHP Register database compiled by Ogem12 his inormation is used in the

administration o the market in Renewable Obligation Certificates (ROCs) he Renewables

Obligation is the primary mechanism by which large scale generators o renewable electric-

ity receive subsidies so the operators o wind arms have a very strong incentive to submit

complete and up-to-date inormation to the database he data extracted covered all onshore

wind generators with at least 05 MW o generating capacity which are eligible or ROCs or the

10 year period April 2002 to March 2012 he key variables are the monthly output o electric-

ity (used or the purpose o allocating ROCs) and nominal generating capacity here are 282

reporting units in the final dataset A reporting unit may correspond to an entire wind arm orto separate phases o development Complications arise when an existing wind arm is re-pow-

ered ie when old wind turbines are replaced by newer and usually more powerul turbines

he database reports this as a change in the generating capacity o the recording unit but in

this analysis repowering is treated as the termination o the old recording unit and the creation

o a new recording unit

A considerable amount o data checking and cleaning was required beore the data could

be analysed he first step was to add the age o each reporting unit his was calculated by

reerence to the month in which the turbines were commissioned For wind arms in operation

in or beore June 2002 the commissioning date was obtained rom various online sources ndash

notably the thewindpowernet database o UK wind arms13 which was cross-checked against

the inormation provided on the websites o wind operators and other sources For recording

units whose first data relates to dates afer June 2002 the commissioning date was initially

assumed to be the month or which data is first reported Cross-checks were carried out or

all recording units which commenced reporting in the period July 2002 to December 2003 In

more than 80 o cases the externally-reported commissioning date is within one month o the

first month or which data is reported he exceptions ndash Bu Farm and Mablethorpe ndash appear to

be due to delays by small operators in reporting data to Ogem

In a number o cases it is clear rom the data that plants were not in ull operation at the

time when data was first reported to the database here are two indicators o partial or delayed

commissioning

bull For an initial series o months the calculated load actor is very low ndash usually below 5 ndash

and then increases abruptly In such cases the data or the early months is dropped and the

commissioning date is treated as the month immediately preceding the first month with a

lsquonormalrsquo load actor14

12 httpswwwrenewablesandchpogemgovuk

13 See httpwwwthewindpowernetwindarms_list_enphp

14 he month in which a plant is reported as having been commissioned is not included in the analysis

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2652

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 25

the decline in the plantrsquos load actor as it ages As an example consider a plant whose reported

capacity was increased rom 99 MW to 10 MW at age 3 I this was simply a change in report-

ing practice this would reduce the estimated load actor rom age 3 onwards By increasing

the initial capacity rom 99 MW to 10 MW ndash or by reducing the capacity rom 10 MW to 99

MW rom age 3 onwards ndash this bias is removed It is not important which adjustment is madebecause the difference between the alternative assumptions is captured by the unit fixed effect

he numbers o separate sites and o observations per site mean that it is not possible to

provide a simple visual representation o the data Figure 8 provides a summary by illustrating

the distributions o load actors by the age o the wind arm pooled across all years he ranges

o the load actors or each age are wide while the medians show no clear trend by age hus a

naiumlve analysis might conclude that there is nothing to investigate However as will be explained

below the distributions mask crucial differences between the perormance o wind arms as

they age because they do not control or the differences in location and wind availability over

time

B Data for Denmark

he data or Danish wind arms used in this study comes rom a database compiled by the

Danish Energy Agency covering the characteristics and perormance o all wind turbines rom

2002 up to the end o August 201216 he basic recording unit or the register is the individual

wind turbine but where a number o turbines are connected to a single meter an identical

average output per turbine is recorded or all turbines in the group he presence o what are in

effect repeated observations increases the noise in the data In addition the model that is used

or the analysis assumes that site operator and turbine characteristics are common or all o theunits that make up a single wind arm o address this point separate wind arms have been

identified by grouping all turbines which have identical values or turbine manuacturer and

type local authority date o commissioning and date o decommissioning (where applicable)

he register contains inormation on 7569 turbines commissioned rom 1977 onwards his

includes a large number o small turbines which would not all within the scope o the ROC

scheme in the UK Hence or comparability with the UK data all turbines commissioned

prior to 1992 were dropped as well as all wind arms with a generating capacity o less than 1

MW he resulting sample covered 823 onshore wind arms and 30 offshore wind arms with

monthly load actors or various periods rom January 2002 to August 2012 Given the cut-off

or wind arms commissioned prior to 1992 the maximum age or an onshore wind arm was 20

years and the maximum age or an offshore wind arm was 17 years

he total capacity o the onshore wind arms recorded in the Danish database was 2757 MW

he average size o an onshore wind arm in Denmark is small ndash only 33 MW as compared

with 161 MW or the UK ndash because 84 o the sample consists o installations with between

1 and 4 turbines he largest onshore installation has 39 turbines with a capacity o 234 MW

while the onshore installation with the greatest capacity has 19 turbines with a capacity o 584

MW Overall the sample o Danish onshore wind arms is older and smaller than the sample

o UK onshore wind arms reflecting the much longer history o wind generation in Denmark

16 httpwwwensdkEN-USINFOFACSANDFIGURESENERGY_SAISICS_AND_INDICAORS

OVERVIEWOFHEENERGYSECORREGISEROFWINDURBINESSiderForsideaspx

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2852

26 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

he larger scale o UK wind arms may account or some o the differences in perormance

curves or the two countries

he sample o offshore wind arms in Denmark is quite small with 30 installations in

total Still it is useul to analyse the perormance o these installations since there is very little

comparable data or the UK he sample covers 394 turbines with an average o just over 13turbines per installation and an average capacity o 288 MW he largest installation which

was commissioned in 2003 has 72 turbines with a total capacity o 1656 MW Again offshore

installations are smaller and older in Denmark than is the current pattern in the UK he first

two wind arms in the sample were commissioned in the years 1995 and 2000 but the sample

increases significantly rom 2002 onwards o avoid reliance upon a sample o 1 or 2 wind

arms to determine specific coefficients or ages gt 10 years all observations with an age greater

than 10 were omitted rom the analysis17

A somewhat different problem with the offshore wind arm data arose because the coeffi-

cients estimated using capacity weights (see below) were heavily influenced by the largest windarm his has had an extremely erratic profile o perormance over time with a very low average

load actor he treatment o this wind arm is discussed in more detail in Section F below

C Specication and estimation methods

As Figure 8 illustrates it is necessary to go beyond simple summary statistics in order to assess

whether there is any systematic relationship between age and the average perormance o wind

arms he perormance o wind arms varies over both time and space some months and years

have more or less wind than the long term average while specific characteristics o wind arms

ndash including location type o turbine and operating regime ndash will influence the perormance oeach plant under identical wind conditions

he combination o (a) site-specific characteristics which are constant over time and (b)

period-specific characteristics which are constant across wind arms can be represented by

what statisticians call an error components model with fixed effects or each wind arm and

each time period he dependent variable in the model consists o a sequence o load actors ndash

denoted by or unit i in month t ndash or each wind arm (the panel unit) over time he load

actors are calculated by dividing the total output in MWh by (24 x number o days in month x

reported generating capacity in MW) and are multiplied by 100 to convert to percentages

he additive version o the error components model may be written as

(1)

in which denotes the age o plant i in period t and is some unction that is either a

representation o or an approximation to the normalised age-perormance relationship he

multiplicative version is similar except that is replaced by where ln() denotes the

natural logarithm he terms are wind arm or unit fixed effects which capture actors such

as location type o turbine operating regimes etc he terms are period fixed effects which

capture the influence o wind conditions over the UK or Denmark as a whole as well as seasonal

maintenance and demand Without loss o generality it may be assumed that the fixed effects are

17 For the avoidance o doubt this means that no attempt was made to estimate the effect o age on the perormance o

offshore wind arms beyond age 10 while the data or the oldest wind arms was included in the analysis up to and

including age 10

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 27

normalised so that ie all o the error components have a zero mean while the

mean o the random error is zero by definition he terms are random errors which capture

random variations in wind conditions that are not specific to the site turbine breakdowns and

other actors that are uncorrelated with either the site or the time period It is assumed that the

mean o the random error is zero while additional assumptions depend upon the method o

estimation that is adopted

he model specified in equation (1) can be estimated in a variety o ways18 A crucial issue is

whether the are or may be significantly correlated with the age terms or any other independent

regressors I it is assumed that there is no such correlation the model is usually reerred to as

a random effects model In that case the equation can be estimated by a variant o least squares

(LS) generalised least squares (GLS) generalised methods o moments (GMM) or maximum

likelihood (ML) ndash all o which should be yield parameter estimates that will converge in prob-

ability to the true parameter values as the sample size increases An alternative but less restric-

tive specification is known as the fixed effects model in which no assumption is made about

the correlation between the unit fixed effects and other regressors he limitation o the fixed

effects model is that it cannot be used to estimate parameters or regressors that vary across

panel units but are constant over time or a panel unit (time-invariant regressors) ndash eg the

year in which a wind arm was commissioned its location or its rated generating capacity he

influence o such variables is captured by the unit fixed effects

In order to avoid the risk o obtaining biased estimates o the age effects the results reported

in this note are based upon estimates using the fixed effects model he influence o time-invar-

iant regressors is examined in a second stage by regressing the estimates o the unit fixed effectson these variables he simplest method o estimating the fixed effects model is known as the

lsquowithinrsquo estimator in which the mean value or each unit is subtracted rom all observations or

that unit For example i is a general polynomial o order M in A equation (1) becomes

(2)

so that the model can be estimated by least square with the inclusion o dummy variables or

each period t he variance-covariance matrix o the coefficients have been estimated by using

(a) a robust sandwich estimator adjusted or clustering by panel unit and (b) a bootstrap esti-

mator with 400 repetitions he estimates o the standard errors will be consistent even i the

random error component is serially correlated within panel units andor has different variances

across panel units19 Since the number o panel units and the average number o time periods

per panel units are both large (other than or Danish offshore wind arms) the least square esti-

mates o the parameters will be unbiased under these assumptions

he bootstrap method is expensive to carry out but it provides a useul cross-check on

the standard errors generated by other methods Unortunately the bootstrap method o

18 See Chapter 21 o AC Cameron amp PK rivedi ndash Microeconometrics Methods and Applications (Cambridge

Cambridge University Press 2005) or a standard textbook treatment o estimation methods or panel data models

19 he estimation was carried out using the xtreg procedure in Stata Version 12 he methods o constructing the

sandwich and bootstrap estimates o the variance-covariance matrix are described in Chapters 8 amp 13 o AC

Cameron amp PK rivedi ndash Microeconometrics using Stata (Revised Edition College Station exas Stata Press 2010)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 11: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 9

The Performance of Wind Farms in the United Kingdom

and Denmark

Introduction

1 Any assessment o the costs o wind power must rely heavily upon assumptions about theaverage load actor that will be achieved by new wind installations over their lietime It is

standard practice to calculate average load actors by year and country or onshore and offshore

installations as shown in able 1 (page 40)1 However such estimates do not provide a

reliable statistical basis or assessing the uture perormance o wind arms in aggregate Part

o the reason is that the amount o wind in any month or year is influenced by long term mete-

orological cycles that have periods o many years notably the North Atlantic Oscillation In

addition average load actors do not allow or changes in the composition o wind installations

by location age size and other actors

2 As the number o wind arms operating in developed countries has grown engineers have an

opportunity to analyse operating experience over extended periods o time With any (relatively)

novel technology the incidence o temporary or permanent breakdowns may be expected to all

over time so it is difficult to disentangle the effects o age on operating perormance rom the

technological immaturity o turbines installed 10 or 20 years ago Nonetheless the technology

or onshore wind turbines has been reasonably mature since the early 2000s his is documented

by data produced by the US Department o Energy which shows that the decline in capital costs

that is characteristic o immature technologies slowed afer 2000 and was reversed afer 20042

3 With at least 10 years o operating data since 2000 it should be possible to examine how thetypical operating perormance o wind installations in the United Kingdom and Denmark

varies with the age o the turbines he estimated age-perormance curves should then be

incorporated in estimates o levelised costs which are ofen used to compare the costs o wind

and other orms o generation

4 he Appendix provides a detailed description o the data and methods employed or this study

In both the UK and Denmark wind operators have a large incentive to produce electricity

whenever there is sufficient wind available since marginal operating costs are small while the

operator receives a much higher price In the UK this price is the sum o the market price plus

the value o (a) the Renewable Obligation Certificates (ROCs) awarded or each MWh o elec-

tricity produced and (b) the associated exemption rom the Climate Change Levy In practice

the effective market value o ROCs and associated incentives means that an onshore wind

operator earns roughly double the average market price o electricity per MWh In Denmark

most onshore wind capacity receives a price premium equivalent to about pound28 per MWh on top

1 he load actor is calculated as the ratio o the amount o electricity actually produced by a turbine or wind arm

over a period o a month or a year divided by the amount o output that would have been produced had it operated

at ull nameplate capacity or the entire period his is expressed as a percentage so that reported load actors lie

between 0 and 100

2 R Wiser amp M Bolinger (2012) 2011 Wind Technologies Market Report Lawrence Berkeley National Laboratory

US Department o Energy It should be noted that the installed cost o wind plants has allen since 2010 but these

changes reflect the cyclical orces o demand and supply that are well established across the electricity sector and

which apply to ossil-uel as well as renewable generators

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1252

10 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

o the market price or about 125 years o operation at a typical load actor3 Offshore capacity

is eligible or a similar price premium but most o it is contracted via tenders that offer a guar-

anteed eed-in tariff he tender tariffs have varied rom pound56 per MWh to twice that figure

extending over 16-18 years o operation at a typical load actor

Age-performance curves

5 A standard indicator o the operating perormance o a wind turbine is its normalised load

actor defined as the total output over some period divided by the maximum potential output

adjusted or wind availability Normalisation or wind availability is not straightorward I

detailed data on wind speeds at specific locations is available then the potential output rom

individual turbines or wind arms can be calculated using standard power curves Since wind

speeds can vary substantially over small distances such calculations are both difficult to imple-

ment and prone to errors when applied to large numbers o wind generators using public

sources o data4

6 Many analyses work in the opposite direction For example the Danish Wind Index ndash the

longest running series o data on wind availability in Europe ndash is constructed rom the common

component o monthly variations in wind output rom a large sample o wind turbines5 his

study adopts a similar approach Wind availability is treated as an unobserved actor whose

contribution to monthly variations in wind output is estimated by a series o fixed period

(monthyear) effects in a statistical model o monthly output or all wind arms he reasons

or adopting this approach are discussed in Section D o the Appendix his explains why the

method must perorm better than the easible alternative o using a single index o average

wind speed or the UK or Denmark

7 he analysis relies upon standard techniques that are widely used in the biological and medical

sciences as well economics and engineering Full details are given in the Appendix he idea is

to treat the actual load actor (ie without adjustment or wind conditions) or a specific wind

generator in any time period as being determined by a set o components which reflect specific

site conditions and the age-related perormance o wind plants in general together with the

period fixed effects and an uncorrelated random error he onshore wind datasets or the UK

and Denmark used or the analysis are large with monthly observations on 282 installations in

the UK and 823 installations in Denmark with an age range rom 0 to 19 years he offshore

wind dataset or Denmark is rather smaller covering only 30 installations but it can be used to

obtain reasonable estimates o perormance up to age 10 he results discussed here are based

upon least squares estimation (see page 27) combined with standard errors that are robust to

various departures rom classical assumptions about the nature o the data generation process

8 he results o the statistical analysis demonstrate an unambiguous and statistically significant

decline in the operating perormance o wind arms as they grow older Figure 1 shows the

3 Conversions to GBP are based on an exchange rate o pound1 = DKK 918 (as at 19 October 2012)

4 All wind generators collect their own inormation on wind speeds in order to manage their plants but they do not

publish such inormation and will not have access to inormation collected by other operators

5 Details o the construction o the Danish Wind Index are given at wwwvindstatdk A paper by Boccard ndash N Boccard

(2009) lsquoCapacity actor o wind power realized values vs estimatesrsquo Energy Policy Vol 37 pp 2679-88 ndash includes a

discussion o the construction o similar indices o wind availability in Germany the UK Netherlands and Sweden

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 11

simplest variant o the normalised age-perormance curve or onshore UK wind installations

together with the equivalent curves or onshore and offshore Danish installations he curves

illustrated are calculated rom the multiplicative error components model described in the

Appendix with log(load actor) as the dependent variable and a quadratic in plant age to repre-

sent the variation in plant perormance with age9 he rate o decline in perormance is greatest or offshore installations in Denmark with a all

rom load actors o over 40 at ages 0 and 1 to less than 15 by at ages 9 and 10 Onshore

installations in the UK show a more rapid rate o decline ndash 09 percentage points per year over

the first 10 years o operation ndash than is the case or Denmark though the normalised Danish

perormance curve lies below the UK curve or the first our years For the UK the normalised

load actor alls to just over 15 at age 10 and to 11 at age 15 With such low load actors it

seems likely that many wind arms will be re-powered ndash ie the turbines will be replaced ndash once

they reach the age o 10 or at most 15

Figure 1 Performance degradation due to age using equal weights

0

5

10

15

20

25

30

35

40

45

50

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

L o

a d f a c t o r ( )

Age (years)

UK onshore DK onshore DK offshore

Note Normalised load actors in Source Authorrsquos estimates

10 here are two plausible explanations or the observed decline in average load actors as wind

arms age he first is that the turbines become less efficient over time as a result o mechanical

wear and tear erosion o the turbine blades and related actors he second is that the turbines

experience more requent breakdowns and their operators take more time to bring them back

into service because they are less concerned about the perormance o older plants Both

reasons may be relevant in different circumstances and it is not possible to identiy a primary

explanation rom the data he requency o extended shutdowns does seem to increase with

age but this could be a reflection o the timing o planned maintenance operations rather than

breakdowns Whatever the cause the reduction in perormance with age is much greater than

would be expected or thermal generating plants

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1452

12 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

11 he age-perormance curves in Figure 1 are derived rom equations estimated by giving equal

weight to each wind arm irrespective o its generating capacity Figure 2 illustrates similar

curves but in this case the estimates are constructed by weighting each wind arm by its generat-

ing capacity reerred to as capacity-weighted his gives a better representation o perormance

degradation per MW o generating capacity he striking result is that the rate o decline inthe perormance o UK onshore wind installations is significantly aster when capacity weights

are used which implies that large wind arms in the UK experience a more rapid decline than

smaller ones he differences are smaller in Denmark but the capacity-weighted decline in

perormance is more rapid than the equal weights decline or Danish onshore installations

Figure 2 Performance degradation due to age using capacity weights

0

5

10

15

20

25

30

35

40

45

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

L o a d f a c t o r ( )

Age (years)

UK onshore DK onshore DK offshore

Note Normalised load actors in Source Authorrsquos estimates

Differences between wind farms

12 he specification o the statistical model means that the unit fixed effects identiy the specific

perormance characteristics o individual wind arms afer adjusting or wind availability and

age hese characteristics may reflect the site andor design o the installation as well as the

way in which it is operated A positive value or the unit fixed effect is equivalent to shifing

the perormance curve up by some percentage applied uniormly at each age while a negative

value shifs it down It ollows that the distribution o unit fixed effects across installations can

be used to compare the relative effectiveness o new installations by location date o commis-

sioning etc Care is required when making comparisons between or example UK and Danish

onshore wind installations because these have been normalised separately so the unit fixed

effects reer to different underlying averages6

6 Some care is required in interpreting the unit fixed effects By deault the average value o the unit fixed effects or

each group is zero but these averages are taken over all observations or each installation in the basic sample For the

purpose o the comparisons in Figure 3 the unit fixed effects have been adjusted so that the average values or each

group are zero when using one value or each installation

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 13

13 o give a sense o the numbers unit fixed effects o (a) 02 or (b) -02 mean that the base load

actor given the age o the plant is multiplied by (a) exp(02) = 122 or (b) exp(-02) = 082 Since

the base load actor or a plant o age 0 is 231 these fixed effects translate to load actors in

the first year o operation o (a) 282 and (b) 189 For a wind arm with 50 MW o gener-

ating capacity the difference amounts to an additional 40700 MWh o electricity output peryear Using the standard figures cited by RenewableUK this would be sufficient to supply 12300

homes or a year his range (040) is slightly greater than the interquartile range o unit fixed

effects or UK onshore wind arms (035) Again or comparison operators o thermal power

plants would regard any difference o even 10 up or down in efficiency relative to what is

expected as a striking and perhaps worrying indication o either good or bad perormance

Figure 3 Range of unit fixed effects by category

- 1

- 5

0

5

1

A d j f i x e d e f f e c t ndash q u a d r a t i c

UK onshore DK onshore DK offshore

Source Authorrsquos estimates

14 A notable eature o the distributions o the unit fixed effects that are summarised in the box

plots shown in Figure 3 is the range o variation between the best and worst perormers espe-

cially or UK onshore and Danish offshore wind arms Box plots are constructed so that the

top and bottom o the box correspond to the upper and lower quartiles while the bars at topand bottom correspond to the maximum and minimum values excluding outliers7 A plant at

the upper quartile o the distribution o UK onshore wind arms will generate an annual output

that is about 40 higher than a plant at the lower quartile he equivalent figure or Danish

onshore wind plants is an increase o 30 while or Danish offshore wind arms the difference

between the upper and lower quartiles is equivalent to 25 times the output o a lower quartile

plant By any standard such differences are important

7 he upper quartile is the point in a distribution such that the 25 o values exceed it while the lower quartile is the

point in a distribution such that 75 o values exceed it he line in the middle o the box marks the median the

point exceeded by 50 o values Outliers are defined as values that lie outside the range o ukey adjacent values

which are defined as the upperlower quartiles +- 15 x interquartile range

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1652

14 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

15 Outside the range between the upper and lower quartiles the top and bottom segments o

the distributions extend urther up and down or UK onshore wind arms than or those in

Denmark he rather different pattern or Danish offshore wind installations is partly a conse-

quence o the relatively small size o the sample but the evidence points to considerably greater

variability in the perormance o offshore wind installations than or onshore installations16 It is somewhat surprising to observe the magnitude o the differences in the perormance in

onshore wind arms in the UK Given the nature o the subsidy regimes and the high capital

cost o developing new installations it might be expected that operators have a strong incen-

tive to identiy the best locations and then choose equipment that will deliver the maximum

amount o electricity output at a high level o reliability For UK onshore operators it seems that

good or bad perormance is somewhat o a lottery However i the subsidies provided by ROCs

are sufficient to underwrite investment in inefficient plants ndash as appears to be the case ndash then

those subsidies are extremely generous or plants that operate close to the efficient rontier

As location is likely to be the main actor that determines the perormance o a specific plant

relative to all other plants the inerence must be that many wind plants have been developed on

sites with poor wind characteristics

Figure 4 Initial load factors by year of commissioning for onshore wind

10

15

20

25

30

35

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

I n i t i a l l o a d f a c t o r ( )

Year commissioned

UK onshore DK onshore Linear (UK onshore) Linear (DK onshore)

Source Authorrsquos estimates

17 here is another disturbing characteristic o UK wind arms which is revealed by examination

o the unit fixed effects and illustrated in Figure 4 It might be expected that the average peror-

mance o recently installed units would be higher than that o older units his should reflect

improvements in turbine design and reliability as well as in the identification o good sites or

wind generation However that is not the pattern or onshore wind arms in the UK he figure

shows the average values o the initial load actors calculated using equally weighted data by

year o commissioning or onshore wind arms in the UK and Denmark together with fitted

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 15

trend lines8 here has been a clear decline in the initial load actors in the UK which has not

occurred in Denmark his pattern is confirmed by more detailed statistical results shown in

the Appendix which take account o location and size o installation

18 he decline in initial perormance is particularly obvious afer 2005 when the rate o commis-

sioning new capacity increased sharply rom less than 500 MW in the 5 years 2000ndash04 to morethan 500 MW per year rom 2005 onwards In contrast the peak years or commissioning new

onshore wind arms in Denmark were 2000ndash02 when 920 MW was installed he rate o new

building (or replacement) picked up again in 2008 but it is still much lower than at the turn o

century Not only is the perormance o the onshore wind plants in the UK commissioned in

recent years significantly worse than that or wind arms commissioned beore 2005 but it is

possible that this is a direct consequence o an overly rapid expansion o capacity In essence

the evidence suggests that the industry does not have the capability to identiy develop and

operate new onshore wind arms at the rate envisaged by UK government targets while main-

taining a satisactory average level o perormance

Implications for future policy and performance

19 he pattern o perormance degradation with age and over time identified in this analysis has

implications or the assessment o new investment in wind arms and or the design o the

proposed Electricity Market Reorm (EMR) here are two related issues that are affected by

the results

a) What are the implications or the design o policies intended to promote the adoption

and expansion o wind generation in the UK his includes the structure o current

subsidies as well as the specification o the contracts that are proposed under the EMR A

related question concerns the expected lie o new wind arms

b) How might the average perormance o wind generation develop over the next decade

his is absolutely critical to any assessment o the amount o wind capacity that will be

required to meet the UK governmentrsquos targets or the share o electricity ndash and more

generally energy ndash rom renewable sources he government has assumed that the

average load actors or both onshore and offshore wind arms will either remain stable

or increase in uture he prospects will be very different i these assumptions are not

borne out

20 he first question can be addressed by considering the discounted sum o cumulative net

output rom a new wind arm on the assumption that its perormance matches the averages

identified in the analysis his takes account o both perormance degradation over time plus

the annual cost o maintaining the plant which may be expressed as a deduction rom gross

production so as to calculate the contribution o expected output at age s to recovering the

original capital cost o building the plant hese figures have to be discounted back to the date

o commissioning in order to assess the net present value o the initial investment in the plant

he curves or constant and observed perormance using a discount rate o 9 (a typical cost o

capital or wind generation) are shown in Figure 5 he black line is based on the ull set o age

8 he initial load actors are calculated as the normalised load actor or UKDanish wind arms at age 0 multiplied by

where is the unit fixed effect or installation i

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1852

16 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

effects while the red dashed line is based on the smoothed quadratic perormance curve and

the blue dashed line shows what the pattern would be i operating perormance was not affected

by age he calculation is extended out to age 25 since DECCrsquos standard comparisons assume

that wind turbines have a lie o 25 years

Figure 5 Impact of performance degradation on discounted cumulative output

0

10

20

30

40

50

60

70

80

90

100

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

o

f d i s c o u n t e d c u m u l a t i v e o u t p u t

Age of plant (years)

Constant Quadratic Full age effects

Source Authorrsquos estimates

21 he key points in the figure are that 80 o the discounted cumulative output o a new wind

plant is likely to be produced in the first 10 years o its lie and 90 in the first 14 years his

is consistent with the structure o Danish subsidies or onshore wind arms which extend over

a typical period o 125 years Since sites or wind arms are scarce and involve the payment o

significant rents that may be linked to output it is very unlikely that any new installation o

wind turbines will have an expected lie o more than 15 years Instead as has happened in the

past wind operators will have a strong incentive to decommission plants afer no more than 15

years and replace the turbines with newer equipment

22 As a consequence any economic assessment o wind generation should not be based on an

expected lie which is longer than 15 years In recent work reported in evidence to the House

o Commons Select Committee on Energy and Climate Change I assumed that wind plants

would have a residual value equal to 20 o their initial cost in real terms at the end o 15 years

he analysis in this paper suggests that this is too avourable an assumption Given the costs o

decommissioning old turbines the residual value is likely to be well below 10 o their initial

cost and the decision point may be at 10 rather than 15 years

23 he corollary o this observation is that it makes little sense to offer long term contracts or 20

years or more that guarantee prices or eed-in tariffs (Fis) to wind operators A contract length

o 10 to 12 years would be sufficient to remove most o the market risk associated with invest-

ment in wind generation In this respect the subsidy arrangements implemented in Denmark

are better designed

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 17

24 At the same time the offer o subsidies andor guaranteed prices may have serious adverse

consequences or the efficiency o wind generation Returning to the unit fixed effects measur-

ing the perormance o wind plants commissioned afer 2005 only 28 out o 159 units have an

operating perormance that exceeds the average or the period hose 28 units account or 360

MW o capacity whereas the remaining 131 units account or 2810 MW o capacity Not onlyare recent plants less efficient than the average or the whole period but the plants which are

below-average in efficiency are typically larger than the more efficient ones and account or a

disproportionate share o the recent additions to generating capacity

25 In mid-2012 there was approximately 4500 MW o onshore wind capacity operating in the

UK By 2020 it is expected that there will be at least 10000 MW (based on proposed wind

arms with planning consent) and perhaps as much 15000 MW o wind capacity in operation

he government is relying upon a substantial contribution rom onshore wind arms towards

meeting its targets or renewable energy in 2020 o provide context the governmentrsquos targets

or renewable energy in 2020 speciy a total o 234 Wh o energy rom all renewable sources

o which 90ndash140 Wh is expected to come rom wind and biomass electricity generation9

he projections or biomass electricity are extremely vague and have to be reconciled with

an assumed increase o 3ndash4 times in the amount o biomass used separately or heat So in

concrete terms it is reasonable to assume that wind generation will account or at least 90 Wh

with 24ndash32 Wh rom onshore generation and the remainder rom offshore

26 he average load actor or onshore wind over the 10 year period rom April 2002 to March

2012 was 256 which is influenced by the large amounts o new capacity added rom 2005

onwards I this load actor were achieved in 2020 the total amount o onshore wind capacityrequired to meet the onshore generation target would be 107 to 143 GW he average load

actor or offshore wind installations (on an unchanged configuration basis) or the five year

period 2007ndash11 was 32010 I this load actor were achieved in 2020 the total amount o

offshore wind capacity required to meet the remainder o the 90 Wh target would be 206

to 235 GW though these estimates would all to 180 to 205 GW i the average load actor

matched the historic average value o 367 or Danish offshore wind installations

27 he question that has to be addressed is whether these load actors are consistent with (a) the

investment programmes that would be required to achieve the total amounts o generation

capacity implied by the governmentrsquos targets and (b) the age profiles o the perormance o

wind arms discussed in this paper he short answer is that this is simply not possible i the

initial load actor or new wind arms continues to decline at the rate observed the past decade

So even as a starting point it is necessary to assume that this underlying deterioration in wind

arm perormance ceases his would be a major change in itsel and it is not obvious what

actors might lead to such a result

28 he results illustrated below are based on a vintage model o perormance that takes account

o the actual expansion o capacity up to the end o 2011 plus whatever investment in new or

replacement turbines is required to achieve the target levels o wind capacity in 2020 o high-light the ull range o potential outcomes these targets are either 10 GW or 15 GW o onshore

9 Department o Energy and Climate Change ndash UK Renewable Energy Road Map July 2011 Figure 2

10 he equivalent average load actor or onshore wind arms was 264

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2052

18 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

wind and either 18 or 24 GW o offshore wind he levels o investment required to meet these

targets depend upon the expected lie o wind turbines For onshore turbines an expected lie

o 10 years is consistent with re-powering decisions over the last decade while an expected lie

o 15 years seems to be the maximum consistent with the age-perormance profile For offshore

turbines the analysis uses expected lives o 15 and 20 years

Figure 6A Projected load factors for UK onshore wind (total capacity of 10 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

Figure 6B Projected load factors for UK onshore wind (total capacity of 15 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

29 Figure 6A shows projected values o the average normalised load actors or UK onshore wind

arms in 2015 and 2020 using a target capacity in 2020 equal to 10 GW plus the age-peror-

mance profiles or the UK and Denmark (DK) Since vintage effects are important in this

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 19

calculation the profile o new or replacement investment affects the results In this case the

level o gross investment is assumed to be constant rom 2013 onwards at either 930 or 1220

MW per year ndash the range is defined by the assumed lie o turbines Achieving a target o 15

GW in 2020 (Figure 6B) would require a level o investment rom 2013 onwards o either 1540

or 1930 MW per year Such figures are 3ndash4 times the average level o investment over theperiod 2005ndash11 so the lower target may provide a better indication o what will happen

30 All o the estimates are calibrated to match the observed average load actor in 2007ndash11 so that

they should be interpreted as relative rather than absolute values Using the age-perormance

profile or UK onshore wind installations the average normalised load actor is likely to all by

about 19 rom 2007ndash11 to 2015 and by 23ndash28 to 2020 I the age-perormance curves or

Danish onshore installations were to apply the reduction in the average normalised load actor

would be considerably less but there would still be a all o up to 5 by 2020

31 In the medium term the all is greater i turbines operate or 15 years because the average age

o all turbines will be higher and thus the degradation in perormance due to age will have a

larger effect By 2020 this effect reduces the average normalised load actor rom 203 or a 10

year lie to 195 or a 15 year lie here is a complex economic and commercial trade-off here

o meet a target o 10 GW in 2020 it is necessary to increase the level o investment in onshore

capacity by 30 rom 2013 Almost all o this will take the orm o re-powering existing wind

arms11 he costs are airly high because this will usually involve new development consents

decommissioning the original turbines and replacing them with a smaller number o larger and

more powerul turbines

32 Estimates o re-powering costs are generally treated as being commercially sensitive but broadproject costs cited in specialist magazines suggest that the unit costs are likely to be at least

pound1000 per kW or more than pound300 million per year in aggregate On the other hand the increase

in the amount o energy generated would be about 12 Wh or a 10 year turbine lie rather

than a 15 year turbine lie though this does not allow or the loss o production during re-pow-

ering Under the best circumstances the payback period may be quite short but the best option

rom a broader economic and social perspective is not obvious

33 Even under the best scenario 10 GW o onshore wind capacity will only generate 178 Wh o

electricity in 2020 well below the lower end o the target range or onshore wind generation

specified in DECCrsquos road map In act allowing or interactions between the age-perormance

profile and investment the amount o onshore wind capacity required in 2020 to generate 24

Wh would be a minimum o 134 GW At the top end o DECCrsquos target range the capacity

required in 2020 to generate 32 Wh would be a minimum o 178 GW Since these estimates

are based on a 10 year turbine lie the level o investment required rom 2013 onwards would

be 1650 MW per year or the bottom o the range and 2350 MW or the top end o the range

In the last 8 years total investment in onshore wind has only ever exceeded 600 MW in one

11 Re-powering is usually defined as the ull replacement o existing turbines whereas retrofitting covers the replace-

ment o various mechanical or electrical components In the data analysis re-powering involves the creation o a

new unit whereas retrofitting affects the perormance o a continuing unit Hence the potential benefits o retrofit-

ting are taken into account in constructing the age-perormance curves

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2252

20 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

year (883 MW in 2008) so it seems rather unlikely that even the lower end o DECCrsquos range or

onshore wind generation will actually be realised

34 O course this assessment would be different i UK onshore wind operators were able to match

the Danish age-perormance curve here are many actors which differentiate the UK and

Danish wind industries he average scale o wind arms in the UK is larger they are moreremote and there appears to be significantly greater variation between operating and potential

new sites in achieved perormance It would thereore require a very optimistic observer to

conclude that the difference in age-perormance curves between the two countries will disap-

pear in the short or medium term

Figure 7 Projected load factors for UK offshore wind in 2020

20

22

24

26

28

30

32

34

18 GW ndash life 15 yrs 18 GW ndash life 20 yrs 24 GW ndash life 15 yrs 24 GW ndash life 20 yrs

L o a d f a c t o r ( )

2007ndash11 2015 2020

Note Based on Danish age-perormance curves Source Authorrsquos estimates

35 Figure 7 shows the results rom carrying out a similar exercise or offshore wind arms in the

UK he only evidence on the age-perormance curves or offshore installations is based upon

the Danish experience so these have been used to construct the projections Further the profile

o investment assumes a progressive increase in additions to capacity rom 2013 to 2016 and

then a stable level o investment rom 2017 to 2020 he DECC road map involves a very large

commitment to offshore wind investment ndash in the range 25-35 GW o new capacity annually

rom 2016 onwards As a consequence the average age o offshore capacity in 2020 will be quite

low and the ull effects o the age-perormance curves will only be elt during the decade ollow-

ing 2020 Even so the average normalised load actor in 2020 will be about 15 lower than or

2007ndash11 while by 2025 it will be about 30 below the base level

36 On these projections 232 GW o offshore capacity is the minimum necessary to generate 58

Wh o electricity rom offshore wind in 2020 his requirement will increase to 298 GW by

2025 simply to offset the effect o the ageing o offshore generation capacity on total output

37 Combining the estimates or onshore and offshore wind generation the total amount o wind

capacity required in 2020 to produce 90 Wh o electricity rom wind will be in the range 39

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 21

to 41 GW once allowance is made or the observed age-perormance curves or both onshore

and offshore installations hese estimates are roughly a third higher than official predictions o

28ndash31 GW o wind capacity in 2020

38 urning the figures round gives a slightly different picture o the reality o the governmentrsquos

strategy on renewable electricity generation A total o 30 GW o wind capacity in 2020 (12GW onshore 18 GW offshore) will generate about 64 Wh o electricity in a normal year

Since electricity generation rom wind and biomass are expected to account or 50 o renew-

able energy in 2020 the consequence is that biomass is expected to account or 53 Wh In

2011 landfill gas and sewage sludge digestion accounted or 44 o electricity rom bioenergy

(DUKES ables 64) here are limited opportunities to expand production rom these sources

so a continuation o historic rates o growth would mean that renewable generation rom other

sources o bioenergy would have to increase rom 72 Wh in 2011 to 46 Wh in 2020 here

are severe constraints on increasing the amounts o electricity generated rom animal biomass

biodegradable municipal solid waste and anaerobic digestion so it would be optimistic to

assume that these will contribute more than 6-8 Wh in 2020 (up rom 26 Wh in 2011)

39 In practice i the projections or renewable energy are to be believed they rest upon an assump-

tion that there will be a large increase in the amount o electricity that is generated rom plant

biomass ndash either on its own or through co-firing with coal his increase will involve increas-

ing the amount o electricity generated in these ways rom 46 Wh in 2011 to 38-40 Wh in

2020 Since the amounts o straw and UK-grown timber available or this purpose are strictly

limited this will come down to the amount o wood chips that can be imported and the cost o

doing this40 he historic average load actor or biomass plants is just over 60 so about 7500 MW o

biomass capacity will be required to supply 40 Wh o electricity he weight o wood chips per

MWh o electricity depends upon moisture content storage boiler design and other actors but

an indicative range is 085 to 115 tonnes per MWh o wood chips with a moisture content o

30 According to DECC and Forestry Commission estimates the UK imported 265 million

tonnes o steam coal and 26 million tonnes o other wood products (including wood chips)

in 2011 It seems that the net effect o the governmentrsquos strategy will be to replace 15 million

tonnes o coal imported in 2011 by 32 to 44 million tonnes o imported wood chips his is a

somewhat strange way o increasing energy security

Conclusion

41 Wind power is a highly capital-intensive technology or generating electricity Its merits rely

entirely upon a substitution o capital or uel inputs he same is true or hydro or tidal or wave

power In comparison with hydro power wind is a low quality resource because o its variability

and because it cannot be stored So the economic case or wind power must rest on obtaining

the most out o the wind that is available

42 While the decline in the achieved perormance o onshore wind turbines in Denmark is

much less than that or the UK or offshore nonetheless the decline in expected output under

standardised wind conditions over 10 years is 10 unweighted and 13 capacity weighted

hese declines accelerate afer age 10 so that the reductions in perormance are 17 and 20

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2452

22 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

respectively afer 15 years For UK onshore wind arms the reduction in perormance due to

age is much worse at 27 unweighted and 69 capacity weighted by age 10

43 Evidence on the perormance o Danish offshore installations is both restricted and so poor

that there may be concern that the results are affected by a small number o outliers Still the

sample contains a reasonable number o sites with at least 5 years o operating experience andthe decline in perormance by age 5 is 38 unweighted and 26 capacity weighted

44 In addition to these results there is strong evidence that the average normalised load actor or

new onshore wind installations in the UK has allen significantly over the period rom 2000 to

2011 his is consistent with a pattern in which the most avourable sites are developed first

Equally it could mean that wind developers have been unable to keep up with the rate o new

investment while maintaining the quality o development and operations For example the site

design or selection o turbine characteristics may make less effective use o the available wind

resources or the sites available than was the case in the past

45 Whatever the reasons the deterioration in initial perormance means that the expected returns

rom the expansion in wind capacity both or investors and in terms o the reduction in CO983090

emissions have been alling without a concomitant decrease in the private and social costs

that are borne by customers and the general public Clearly this is unsatisactory at best and it

suggests that the benefits claimed or current policies cannot be taken at ace value

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 23

Appendix Data and Methods

A Data for the United Kingdom

he raw data used or this study was extracted by the Renewable Energy Foundation rom the

Renewables and CHP Register database compiled by Ogem12 his inormation is used in the

administration o the market in Renewable Obligation Certificates (ROCs) he Renewables

Obligation is the primary mechanism by which large scale generators o renewable electric-

ity receive subsidies so the operators o wind arms have a very strong incentive to submit

complete and up-to-date inormation to the database he data extracted covered all onshore

wind generators with at least 05 MW o generating capacity which are eligible or ROCs or the

10 year period April 2002 to March 2012 he key variables are the monthly output o electric-

ity (used or the purpose o allocating ROCs) and nominal generating capacity here are 282

reporting units in the final dataset A reporting unit may correspond to an entire wind arm orto separate phases o development Complications arise when an existing wind arm is re-pow-

ered ie when old wind turbines are replaced by newer and usually more powerul turbines

he database reports this as a change in the generating capacity o the recording unit but in

this analysis repowering is treated as the termination o the old recording unit and the creation

o a new recording unit

A considerable amount o data checking and cleaning was required beore the data could

be analysed he first step was to add the age o each reporting unit his was calculated by

reerence to the month in which the turbines were commissioned For wind arms in operation

in or beore June 2002 the commissioning date was obtained rom various online sources ndash

notably the thewindpowernet database o UK wind arms13 which was cross-checked against

the inormation provided on the websites o wind operators and other sources For recording

units whose first data relates to dates afer June 2002 the commissioning date was initially

assumed to be the month or which data is first reported Cross-checks were carried out or

all recording units which commenced reporting in the period July 2002 to December 2003 In

more than 80 o cases the externally-reported commissioning date is within one month o the

first month or which data is reported he exceptions ndash Bu Farm and Mablethorpe ndash appear to

be due to delays by small operators in reporting data to Ogem

In a number o cases it is clear rom the data that plants were not in ull operation at the

time when data was first reported to the database here are two indicators o partial or delayed

commissioning

bull For an initial series o months the calculated load actor is very low ndash usually below 5 ndash

and then increases abruptly In such cases the data or the early months is dropped and the

commissioning date is treated as the month immediately preceding the first month with a

lsquonormalrsquo load actor14

12 httpswwwrenewablesandchpogemgovuk

13 See httpwwwthewindpowernetwindarms_list_enphp

14 he month in which a plant is reported as having been commissioned is not included in the analysis

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2652

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 25

the decline in the plantrsquos load actor as it ages As an example consider a plant whose reported

capacity was increased rom 99 MW to 10 MW at age 3 I this was simply a change in report-

ing practice this would reduce the estimated load actor rom age 3 onwards By increasing

the initial capacity rom 99 MW to 10 MW ndash or by reducing the capacity rom 10 MW to 99

MW rom age 3 onwards ndash this bias is removed It is not important which adjustment is madebecause the difference between the alternative assumptions is captured by the unit fixed effect

he numbers o separate sites and o observations per site mean that it is not possible to

provide a simple visual representation o the data Figure 8 provides a summary by illustrating

the distributions o load actors by the age o the wind arm pooled across all years he ranges

o the load actors or each age are wide while the medians show no clear trend by age hus a

naiumlve analysis might conclude that there is nothing to investigate However as will be explained

below the distributions mask crucial differences between the perormance o wind arms as

they age because they do not control or the differences in location and wind availability over

time

B Data for Denmark

he data or Danish wind arms used in this study comes rom a database compiled by the

Danish Energy Agency covering the characteristics and perormance o all wind turbines rom

2002 up to the end o August 201216 he basic recording unit or the register is the individual

wind turbine but where a number o turbines are connected to a single meter an identical

average output per turbine is recorded or all turbines in the group he presence o what are in

effect repeated observations increases the noise in the data In addition the model that is used

or the analysis assumes that site operator and turbine characteristics are common or all o theunits that make up a single wind arm o address this point separate wind arms have been

identified by grouping all turbines which have identical values or turbine manuacturer and

type local authority date o commissioning and date o decommissioning (where applicable)

he register contains inormation on 7569 turbines commissioned rom 1977 onwards his

includes a large number o small turbines which would not all within the scope o the ROC

scheme in the UK Hence or comparability with the UK data all turbines commissioned

prior to 1992 were dropped as well as all wind arms with a generating capacity o less than 1

MW he resulting sample covered 823 onshore wind arms and 30 offshore wind arms with

monthly load actors or various periods rom January 2002 to August 2012 Given the cut-off

or wind arms commissioned prior to 1992 the maximum age or an onshore wind arm was 20

years and the maximum age or an offshore wind arm was 17 years

he total capacity o the onshore wind arms recorded in the Danish database was 2757 MW

he average size o an onshore wind arm in Denmark is small ndash only 33 MW as compared

with 161 MW or the UK ndash because 84 o the sample consists o installations with between

1 and 4 turbines he largest onshore installation has 39 turbines with a capacity o 234 MW

while the onshore installation with the greatest capacity has 19 turbines with a capacity o 584

MW Overall the sample o Danish onshore wind arms is older and smaller than the sample

o UK onshore wind arms reflecting the much longer history o wind generation in Denmark

16 httpwwwensdkEN-USINFOFACSANDFIGURESENERGY_SAISICS_AND_INDICAORS

OVERVIEWOFHEENERGYSECORREGISEROFWINDURBINESSiderForsideaspx

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2852

26 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

he larger scale o UK wind arms may account or some o the differences in perormance

curves or the two countries

he sample o offshore wind arms in Denmark is quite small with 30 installations in

total Still it is useul to analyse the perormance o these installations since there is very little

comparable data or the UK he sample covers 394 turbines with an average o just over 13turbines per installation and an average capacity o 288 MW he largest installation which

was commissioned in 2003 has 72 turbines with a total capacity o 1656 MW Again offshore

installations are smaller and older in Denmark than is the current pattern in the UK he first

two wind arms in the sample were commissioned in the years 1995 and 2000 but the sample

increases significantly rom 2002 onwards o avoid reliance upon a sample o 1 or 2 wind

arms to determine specific coefficients or ages gt 10 years all observations with an age greater

than 10 were omitted rom the analysis17

A somewhat different problem with the offshore wind arm data arose because the coeffi-

cients estimated using capacity weights (see below) were heavily influenced by the largest windarm his has had an extremely erratic profile o perormance over time with a very low average

load actor he treatment o this wind arm is discussed in more detail in Section F below

C Specication and estimation methods

As Figure 8 illustrates it is necessary to go beyond simple summary statistics in order to assess

whether there is any systematic relationship between age and the average perormance o wind

arms he perormance o wind arms varies over both time and space some months and years

have more or less wind than the long term average while specific characteristics o wind arms

ndash including location type o turbine and operating regime ndash will influence the perormance oeach plant under identical wind conditions

he combination o (a) site-specific characteristics which are constant over time and (b)

period-specific characteristics which are constant across wind arms can be represented by

what statisticians call an error components model with fixed effects or each wind arm and

each time period he dependent variable in the model consists o a sequence o load actors ndash

denoted by or unit i in month t ndash or each wind arm (the panel unit) over time he load

actors are calculated by dividing the total output in MWh by (24 x number o days in month x

reported generating capacity in MW) and are multiplied by 100 to convert to percentages

he additive version o the error components model may be written as

(1)

in which denotes the age o plant i in period t and is some unction that is either a

representation o or an approximation to the normalised age-perormance relationship he

multiplicative version is similar except that is replaced by where ln() denotes the

natural logarithm he terms are wind arm or unit fixed effects which capture actors such

as location type o turbine operating regimes etc he terms are period fixed effects which

capture the influence o wind conditions over the UK or Denmark as a whole as well as seasonal

maintenance and demand Without loss o generality it may be assumed that the fixed effects are

17 For the avoidance o doubt this means that no attempt was made to estimate the effect o age on the perormance o

offshore wind arms beyond age 10 while the data or the oldest wind arms was included in the analysis up to and

including age 10

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 27

normalised so that ie all o the error components have a zero mean while the

mean o the random error is zero by definition he terms are random errors which capture

random variations in wind conditions that are not specific to the site turbine breakdowns and

other actors that are uncorrelated with either the site or the time period It is assumed that the

mean o the random error is zero while additional assumptions depend upon the method o

estimation that is adopted

he model specified in equation (1) can be estimated in a variety o ways18 A crucial issue is

whether the are or may be significantly correlated with the age terms or any other independent

regressors I it is assumed that there is no such correlation the model is usually reerred to as

a random effects model In that case the equation can be estimated by a variant o least squares

(LS) generalised least squares (GLS) generalised methods o moments (GMM) or maximum

likelihood (ML) ndash all o which should be yield parameter estimates that will converge in prob-

ability to the true parameter values as the sample size increases An alternative but less restric-

tive specification is known as the fixed effects model in which no assumption is made about

the correlation between the unit fixed effects and other regressors he limitation o the fixed

effects model is that it cannot be used to estimate parameters or regressors that vary across

panel units but are constant over time or a panel unit (time-invariant regressors) ndash eg the

year in which a wind arm was commissioned its location or its rated generating capacity he

influence o such variables is captured by the unit fixed effects

In order to avoid the risk o obtaining biased estimates o the age effects the results reported

in this note are based upon estimates using the fixed effects model he influence o time-invar-

iant regressors is examined in a second stage by regressing the estimates o the unit fixed effectson these variables he simplest method o estimating the fixed effects model is known as the

lsquowithinrsquo estimator in which the mean value or each unit is subtracted rom all observations or

that unit For example i is a general polynomial o order M in A equation (1) becomes

(2)

so that the model can be estimated by least square with the inclusion o dummy variables or

each period t he variance-covariance matrix o the coefficients have been estimated by using

(a) a robust sandwich estimator adjusted or clustering by panel unit and (b) a bootstrap esti-

mator with 400 repetitions he estimates o the standard errors will be consistent even i the

random error component is serially correlated within panel units andor has different variances

across panel units19 Since the number o panel units and the average number o time periods

per panel units are both large (other than or Danish offshore wind arms) the least square esti-

mates o the parameters will be unbiased under these assumptions

he bootstrap method is expensive to carry out but it provides a useul cross-check on

the standard errors generated by other methods Unortunately the bootstrap method o

18 See Chapter 21 o AC Cameron amp PK rivedi ndash Microeconometrics Methods and Applications (Cambridge

Cambridge University Press 2005) or a standard textbook treatment o estimation methods or panel data models

19 he estimation was carried out using the xtreg procedure in Stata Version 12 he methods o constructing the

sandwich and bootstrap estimates o the variance-covariance matrix are described in Chapters 8 amp 13 o AC

Cameron amp PK rivedi ndash Microeconometrics using Stata (Revised Edition College Station exas Stata Press 2010)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 12: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1252

10 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

o the market price or about 125 years o operation at a typical load actor3 Offshore capacity

is eligible or a similar price premium but most o it is contracted via tenders that offer a guar-

anteed eed-in tariff he tender tariffs have varied rom pound56 per MWh to twice that figure

extending over 16-18 years o operation at a typical load actor

Age-performance curves

5 A standard indicator o the operating perormance o a wind turbine is its normalised load

actor defined as the total output over some period divided by the maximum potential output

adjusted or wind availability Normalisation or wind availability is not straightorward I

detailed data on wind speeds at specific locations is available then the potential output rom

individual turbines or wind arms can be calculated using standard power curves Since wind

speeds can vary substantially over small distances such calculations are both difficult to imple-

ment and prone to errors when applied to large numbers o wind generators using public

sources o data4

6 Many analyses work in the opposite direction For example the Danish Wind Index ndash the

longest running series o data on wind availability in Europe ndash is constructed rom the common

component o monthly variations in wind output rom a large sample o wind turbines5 his

study adopts a similar approach Wind availability is treated as an unobserved actor whose

contribution to monthly variations in wind output is estimated by a series o fixed period

(monthyear) effects in a statistical model o monthly output or all wind arms he reasons

or adopting this approach are discussed in Section D o the Appendix his explains why the

method must perorm better than the easible alternative o using a single index o average

wind speed or the UK or Denmark

7 he analysis relies upon standard techniques that are widely used in the biological and medical

sciences as well economics and engineering Full details are given in the Appendix he idea is

to treat the actual load actor (ie without adjustment or wind conditions) or a specific wind

generator in any time period as being determined by a set o components which reflect specific

site conditions and the age-related perormance o wind plants in general together with the

period fixed effects and an uncorrelated random error he onshore wind datasets or the UK

and Denmark used or the analysis are large with monthly observations on 282 installations in

the UK and 823 installations in Denmark with an age range rom 0 to 19 years he offshore

wind dataset or Denmark is rather smaller covering only 30 installations but it can be used to

obtain reasonable estimates o perormance up to age 10 he results discussed here are based

upon least squares estimation (see page 27) combined with standard errors that are robust to

various departures rom classical assumptions about the nature o the data generation process

8 he results o the statistical analysis demonstrate an unambiguous and statistically significant

decline in the operating perormance o wind arms as they grow older Figure 1 shows the

3 Conversions to GBP are based on an exchange rate o pound1 = DKK 918 (as at 19 October 2012)

4 All wind generators collect their own inormation on wind speeds in order to manage their plants but they do not

publish such inormation and will not have access to inormation collected by other operators

5 Details o the construction o the Danish Wind Index are given at wwwvindstatdk A paper by Boccard ndash N Boccard

(2009) lsquoCapacity actor o wind power realized values vs estimatesrsquo Energy Policy Vol 37 pp 2679-88 ndash includes a

discussion o the construction o similar indices o wind availability in Germany the UK Netherlands and Sweden

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 11

simplest variant o the normalised age-perormance curve or onshore UK wind installations

together with the equivalent curves or onshore and offshore Danish installations he curves

illustrated are calculated rom the multiplicative error components model described in the

Appendix with log(load actor) as the dependent variable and a quadratic in plant age to repre-

sent the variation in plant perormance with age9 he rate o decline in perormance is greatest or offshore installations in Denmark with a all

rom load actors o over 40 at ages 0 and 1 to less than 15 by at ages 9 and 10 Onshore

installations in the UK show a more rapid rate o decline ndash 09 percentage points per year over

the first 10 years o operation ndash than is the case or Denmark though the normalised Danish

perormance curve lies below the UK curve or the first our years For the UK the normalised

load actor alls to just over 15 at age 10 and to 11 at age 15 With such low load actors it

seems likely that many wind arms will be re-powered ndash ie the turbines will be replaced ndash once

they reach the age o 10 or at most 15

Figure 1 Performance degradation due to age using equal weights

0

5

10

15

20

25

30

35

40

45

50

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

L o

a d f a c t o r ( )

Age (years)

UK onshore DK onshore DK offshore

Note Normalised load actors in Source Authorrsquos estimates

10 here are two plausible explanations or the observed decline in average load actors as wind

arms age he first is that the turbines become less efficient over time as a result o mechanical

wear and tear erosion o the turbine blades and related actors he second is that the turbines

experience more requent breakdowns and their operators take more time to bring them back

into service because they are less concerned about the perormance o older plants Both

reasons may be relevant in different circumstances and it is not possible to identiy a primary

explanation rom the data he requency o extended shutdowns does seem to increase with

age but this could be a reflection o the timing o planned maintenance operations rather than

breakdowns Whatever the cause the reduction in perormance with age is much greater than

would be expected or thermal generating plants

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1452

12 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

11 he age-perormance curves in Figure 1 are derived rom equations estimated by giving equal

weight to each wind arm irrespective o its generating capacity Figure 2 illustrates similar

curves but in this case the estimates are constructed by weighting each wind arm by its generat-

ing capacity reerred to as capacity-weighted his gives a better representation o perormance

degradation per MW o generating capacity he striking result is that the rate o decline inthe perormance o UK onshore wind installations is significantly aster when capacity weights

are used which implies that large wind arms in the UK experience a more rapid decline than

smaller ones he differences are smaller in Denmark but the capacity-weighted decline in

perormance is more rapid than the equal weights decline or Danish onshore installations

Figure 2 Performance degradation due to age using capacity weights

0

5

10

15

20

25

30

35

40

45

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

L o a d f a c t o r ( )

Age (years)

UK onshore DK onshore DK offshore

Note Normalised load actors in Source Authorrsquos estimates

Differences between wind farms

12 he specification o the statistical model means that the unit fixed effects identiy the specific

perormance characteristics o individual wind arms afer adjusting or wind availability and

age hese characteristics may reflect the site andor design o the installation as well as the

way in which it is operated A positive value or the unit fixed effect is equivalent to shifing

the perormance curve up by some percentage applied uniormly at each age while a negative

value shifs it down It ollows that the distribution o unit fixed effects across installations can

be used to compare the relative effectiveness o new installations by location date o commis-

sioning etc Care is required when making comparisons between or example UK and Danish

onshore wind installations because these have been normalised separately so the unit fixed

effects reer to different underlying averages6

6 Some care is required in interpreting the unit fixed effects By deault the average value o the unit fixed effects or

each group is zero but these averages are taken over all observations or each installation in the basic sample For the

purpose o the comparisons in Figure 3 the unit fixed effects have been adjusted so that the average values or each

group are zero when using one value or each installation

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 13

13 o give a sense o the numbers unit fixed effects o (a) 02 or (b) -02 mean that the base load

actor given the age o the plant is multiplied by (a) exp(02) = 122 or (b) exp(-02) = 082 Since

the base load actor or a plant o age 0 is 231 these fixed effects translate to load actors in

the first year o operation o (a) 282 and (b) 189 For a wind arm with 50 MW o gener-

ating capacity the difference amounts to an additional 40700 MWh o electricity output peryear Using the standard figures cited by RenewableUK this would be sufficient to supply 12300

homes or a year his range (040) is slightly greater than the interquartile range o unit fixed

effects or UK onshore wind arms (035) Again or comparison operators o thermal power

plants would regard any difference o even 10 up or down in efficiency relative to what is

expected as a striking and perhaps worrying indication o either good or bad perormance

Figure 3 Range of unit fixed effects by category

- 1

- 5

0

5

1

A d j f i x e d e f f e c t ndash q u a d r a t i c

UK onshore DK onshore DK offshore

Source Authorrsquos estimates

14 A notable eature o the distributions o the unit fixed effects that are summarised in the box

plots shown in Figure 3 is the range o variation between the best and worst perormers espe-

cially or UK onshore and Danish offshore wind arms Box plots are constructed so that the

top and bottom o the box correspond to the upper and lower quartiles while the bars at topand bottom correspond to the maximum and minimum values excluding outliers7 A plant at

the upper quartile o the distribution o UK onshore wind arms will generate an annual output

that is about 40 higher than a plant at the lower quartile he equivalent figure or Danish

onshore wind plants is an increase o 30 while or Danish offshore wind arms the difference

between the upper and lower quartiles is equivalent to 25 times the output o a lower quartile

plant By any standard such differences are important

7 he upper quartile is the point in a distribution such that the 25 o values exceed it while the lower quartile is the

point in a distribution such that 75 o values exceed it he line in the middle o the box marks the median the

point exceeded by 50 o values Outliers are defined as values that lie outside the range o ukey adjacent values

which are defined as the upperlower quartiles +- 15 x interquartile range

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1652

14 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

15 Outside the range between the upper and lower quartiles the top and bottom segments o

the distributions extend urther up and down or UK onshore wind arms than or those in

Denmark he rather different pattern or Danish offshore wind installations is partly a conse-

quence o the relatively small size o the sample but the evidence points to considerably greater

variability in the perormance o offshore wind installations than or onshore installations16 It is somewhat surprising to observe the magnitude o the differences in the perormance in

onshore wind arms in the UK Given the nature o the subsidy regimes and the high capital

cost o developing new installations it might be expected that operators have a strong incen-

tive to identiy the best locations and then choose equipment that will deliver the maximum

amount o electricity output at a high level o reliability For UK onshore operators it seems that

good or bad perormance is somewhat o a lottery However i the subsidies provided by ROCs

are sufficient to underwrite investment in inefficient plants ndash as appears to be the case ndash then

those subsidies are extremely generous or plants that operate close to the efficient rontier

As location is likely to be the main actor that determines the perormance o a specific plant

relative to all other plants the inerence must be that many wind plants have been developed on

sites with poor wind characteristics

Figure 4 Initial load factors by year of commissioning for onshore wind

10

15

20

25

30

35

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

I n i t i a l l o a d f a c t o r ( )

Year commissioned

UK onshore DK onshore Linear (UK onshore) Linear (DK onshore)

Source Authorrsquos estimates

17 here is another disturbing characteristic o UK wind arms which is revealed by examination

o the unit fixed effects and illustrated in Figure 4 It might be expected that the average peror-

mance o recently installed units would be higher than that o older units his should reflect

improvements in turbine design and reliability as well as in the identification o good sites or

wind generation However that is not the pattern or onshore wind arms in the UK he figure

shows the average values o the initial load actors calculated using equally weighted data by

year o commissioning or onshore wind arms in the UK and Denmark together with fitted

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 15

trend lines8 here has been a clear decline in the initial load actors in the UK which has not

occurred in Denmark his pattern is confirmed by more detailed statistical results shown in

the Appendix which take account o location and size o installation

18 he decline in initial perormance is particularly obvious afer 2005 when the rate o commis-

sioning new capacity increased sharply rom less than 500 MW in the 5 years 2000ndash04 to morethan 500 MW per year rom 2005 onwards In contrast the peak years or commissioning new

onshore wind arms in Denmark were 2000ndash02 when 920 MW was installed he rate o new

building (or replacement) picked up again in 2008 but it is still much lower than at the turn o

century Not only is the perormance o the onshore wind plants in the UK commissioned in

recent years significantly worse than that or wind arms commissioned beore 2005 but it is

possible that this is a direct consequence o an overly rapid expansion o capacity In essence

the evidence suggests that the industry does not have the capability to identiy develop and

operate new onshore wind arms at the rate envisaged by UK government targets while main-

taining a satisactory average level o perormance

Implications for future policy and performance

19 he pattern o perormance degradation with age and over time identified in this analysis has

implications or the assessment o new investment in wind arms and or the design o the

proposed Electricity Market Reorm (EMR) here are two related issues that are affected by

the results

a) What are the implications or the design o policies intended to promote the adoption

and expansion o wind generation in the UK his includes the structure o current

subsidies as well as the specification o the contracts that are proposed under the EMR A

related question concerns the expected lie o new wind arms

b) How might the average perormance o wind generation develop over the next decade

his is absolutely critical to any assessment o the amount o wind capacity that will be

required to meet the UK governmentrsquos targets or the share o electricity ndash and more

generally energy ndash rom renewable sources he government has assumed that the

average load actors or both onshore and offshore wind arms will either remain stable

or increase in uture he prospects will be very different i these assumptions are not

borne out

20 he first question can be addressed by considering the discounted sum o cumulative net

output rom a new wind arm on the assumption that its perormance matches the averages

identified in the analysis his takes account o both perormance degradation over time plus

the annual cost o maintaining the plant which may be expressed as a deduction rom gross

production so as to calculate the contribution o expected output at age s to recovering the

original capital cost o building the plant hese figures have to be discounted back to the date

o commissioning in order to assess the net present value o the initial investment in the plant

he curves or constant and observed perormance using a discount rate o 9 (a typical cost o

capital or wind generation) are shown in Figure 5 he black line is based on the ull set o age

8 he initial load actors are calculated as the normalised load actor or UKDanish wind arms at age 0 multiplied by

where is the unit fixed effect or installation i

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1852

16 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

effects while the red dashed line is based on the smoothed quadratic perormance curve and

the blue dashed line shows what the pattern would be i operating perormance was not affected

by age he calculation is extended out to age 25 since DECCrsquos standard comparisons assume

that wind turbines have a lie o 25 years

Figure 5 Impact of performance degradation on discounted cumulative output

0

10

20

30

40

50

60

70

80

90

100

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

o

f d i s c o u n t e d c u m u l a t i v e o u t p u t

Age of plant (years)

Constant Quadratic Full age effects

Source Authorrsquos estimates

21 he key points in the figure are that 80 o the discounted cumulative output o a new wind

plant is likely to be produced in the first 10 years o its lie and 90 in the first 14 years his

is consistent with the structure o Danish subsidies or onshore wind arms which extend over

a typical period o 125 years Since sites or wind arms are scarce and involve the payment o

significant rents that may be linked to output it is very unlikely that any new installation o

wind turbines will have an expected lie o more than 15 years Instead as has happened in the

past wind operators will have a strong incentive to decommission plants afer no more than 15

years and replace the turbines with newer equipment

22 As a consequence any economic assessment o wind generation should not be based on an

expected lie which is longer than 15 years In recent work reported in evidence to the House

o Commons Select Committee on Energy and Climate Change I assumed that wind plants

would have a residual value equal to 20 o their initial cost in real terms at the end o 15 years

he analysis in this paper suggests that this is too avourable an assumption Given the costs o

decommissioning old turbines the residual value is likely to be well below 10 o their initial

cost and the decision point may be at 10 rather than 15 years

23 he corollary o this observation is that it makes little sense to offer long term contracts or 20

years or more that guarantee prices or eed-in tariffs (Fis) to wind operators A contract length

o 10 to 12 years would be sufficient to remove most o the market risk associated with invest-

ment in wind generation In this respect the subsidy arrangements implemented in Denmark

are better designed

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 17

24 At the same time the offer o subsidies andor guaranteed prices may have serious adverse

consequences or the efficiency o wind generation Returning to the unit fixed effects measur-

ing the perormance o wind plants commissioned afer 2005 only 28 out o 159 units have an

operating perormance that exceeds the average or the period hose 28 units account or 360

MW o capacity whereas the remaining 131 units account or 2810 MW o capacity Not onlyare recent plants less efficient than the average or the whole period but the plants which are

below-average in efficiency are typically larger than the more efficient ones and account or a

disproportionate share o the recent additions to generating capacity

25 In mid-2012 there was approximately 4500 MW o onshore wind capacity operating in the

UK By 2020 it is expected that there will be at least 10000 MW (based on proposed wind

arms with planning consent) and perhaps as much 15000 MW o wind capacity in operation

he government is relying upon a substantial contribution rom onshore wind arms towards

meeting its targets or renewable energy in 2020 o provide context the governmentrsquos targets

or renewable energy in 2020 speciy a total o 234 Wh o energy rom all renewable sources

o which 90ndash140 Wh is expected to come rom wind and biomass electricity generation9

he projections or biomass electricity are extremely vague and have to be reconciled with

an assumed increase o 3ndash4 times in the amount o biomass used separately or heat So in

concrete terms it is reasonable to assume that wind generation will account or at least 90 Wh

with 24ndash32 Wh rom onshore generation and the remainder rom offshore

26 he average load actor or onshore wind over the 10 year period rom April 2002 to March

2012 was 256 which is influenced by the large amounts o new capacity added rom 2005

onwards I this load actor were achieved in 2020 the total amount o onshore wind capacityrequired to meet the onshore generation target would be 107 to 143 GW he average load

actor or offshore wind installations (on an unchanged configuration basis) or the five year

period 2007ndash11 was 32010 I this load actor were achieved in 2020 the total amount o

offshore wind capacity required to meet the remainder o the 90 Wh target would be 206

to 235 GW though these estimates would all to 180 to 205 GW i the average load actor

matched the historic average value o 367 or Danish offshore wind installations

27 he question that has to be addressed is whether these load actors are consistent with (a) the

investment programmes that would be required to achieve the total amounts o generation

capacity implied by the governmentrsquos targets and (b) the age profiles o the perormance o

wind arms discussed in this paper he short answer is that this is simply not possible i the

initial load actor or new wind arms continues to decline at the rate observed the past decade

So even as a starting point it is necessary to assume that this underlying deterioration in wind

arm perormance ceases his would be a major change in itsel and it is not obvious what

actors might lead to such a result

28 he results illustrated below are based on a vintage model o perormance that takes account

o the actual expansion o capacity up to the end o 2011 plus whatever investment in new or

replacement turbines is required to achieve the target levels o wind capacity in 2020 o high-light the ull range o potential outcomes these targets are either 10 GW or 15 GW o onshore

9 Department o Energy and Climate Change ndash UK Renewable Energy Road Map July 2011 Figure 2

10 he equivalent average load actor or onshore wind arms was 264

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2052

18 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

wind and either 18 or 24 GW o offshore wind he levels o investment required to meet these

targets depend upon the expected lie o wind turbines For onshore turbines an expected lie

o 10 years is consistent with re-powering decisions over the last decade while an expected lie

o 15 years seems to be the maximum consistent with the age-perormance profile For offshore

turbines the analysis uses expected lives o 15 and 20 years

Figure 6A Projected load factors for UK onshore wind (total capacity of 10 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

Figure 6B Projected load factors for UK onshore wind (total capacity of 15 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

29 Figure 6A shows projected values o the average normalised load actors or UK onshore wind

arms in 2015 and 2020 using a target capacity in 2020 equal to 10 GW plus the age-peror-

mance profiles or the UK and Denmark (DK) Since vintage effects are important in this

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 19

calculation the profile o new or replacement investment affects the results In this case the

level o gross investment is assumed to be constant rom 2013 onwards at either 930 or 1220

MW per year ndash the range is defined by the assumed lie o turbines Achieving a target o 15

GW in 2020 (Figure 6B) would require a level o investment rom 2013 onwards o either 1540

or 1930 MW per year Such figures are 3ndash4 times the average level o investment over theperiod 2005ndash11 so the lower target may provide a better indication o what will happen

30 All o the estimates are calibrated to match the observed average load actor in 2007ndash11 so that

they should be interpreted as relative rather than absolute values Using the age-perormance

profile or UK onshore wind installations the average normalised load actor is likely to all by

about 19 rom 2007ndash11 to 2015 and by 23ndash28 to 2020 I the age-perormance curves or

Danish onshore installations were to apply the reduction in the average normalised load actor

would be considerably less but there would still be a all o up to 5 by 2020

31 In the medium term the all is greater i turbines operate or 15 years because the average age

o all turbines will be higher and thus the degradation in perormance due to age will have a

larger effect By 2020 this effect reduces the average normalised load actor rom 203 or a 10

year lie to 195 or a 15 year lie here is a complex economic and commercial trade-off here

o meet a target o 10 GW in 2020 it is necessary to increase the level o investment in onshore

capacity by 30 rom 2013 Almost all o this will take the orm o re-powering existing wind

arms11 he costs are airly high because this will usually involve new development consents

decommissioning the original turbines and replacing them with a smaller number o larger and

more powerul turbines

32 Estimates o re-powering costs are generally treated as being commercially sensitive but broadproject costs cited in specialist magazines suggest that the unit costs are likely to be at least

pound1000 per kW or more than pound300 million per year in aggregate On the other hand the increase

in the amount o energy generated would be about 12 Wh or a 10 year turbine lie rather

than a 15 year turbine lie though this does not allow or the loss o production during re-pow-

ering Under the best circumstances the payback period may be quite short but the best option

rom a broader economic and social perspective is not obvious

33 Even under the best scenario 10 GW o onshore wind capacity will only generate 178 Wh o

electricity in 2020 well below the lower end o the target range or onshore wind generation

specified in DECCrsquos road map In act allowing or interactions between the age-perormance

profile and investment the amount o onshore wind capacity required in 2020 to generate 24

Wh would be a minimum o 134 GW At the top end o DECCrsquos target range the capacity

required in 2020 to generate 32 Wh would be a minimum o 178 GW Since these estimates

are based on a 10 year turbine lie the level o investment required rom 2013 onwards would

be 1650 MW per year or the bottom o the range and 2350 MW or the top end o the range

In the last 8 years total investment in onshore wind has only ever exceeded 600 MW in one

11 Re-powering is usually defined as the ull replacement o existing turbines whereas retrofitting covers the replace-

ment o various mechanical or electrical components In the data analysis re-powering involves the creation o a

new unit whereas retrofitting affects the perormance o a continuing unit Hence the potential benefits o retrofit-

ting are taken into account in constructing the age-perormance curves

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2252

20 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

year (883 MW in 2008) so it seems rather unlikely that even the lower end o DECCrsquos range or

onshore wind generation will actually be realised

34 O course this assessment would be different i UK onshore wind operators were able to match

the Danish age-perormance curve here are many actors which differentiate the UK and

Danish wind industries he average scale o wind arms in the UK is larger they are moreremote and there appears to be significantly greater variation between operating and potential

new sites in achieved perormance It would thereore require a very optimistic observer to

conclude that the difference in age-perormance curves between the two countries will disap-

pear in the short or medium term

Figure 7 Projected load factors for UK offshore wind in 2020

20

22

24

26

28

30

32

34

18 GW ndash life 15 yrs 18 GW ndash life 20 yrs 24 GW ndash life 15 yrs 24 GW ndash life 20 yrs

L o a d f a c t o r ( )

2007ndash11 2015 2020

Note Based on Danish age-perormance curves Source Authorrsquos estimates

35 Figure 7 shows the results rom carrying out a similar exercise or offshore wind arms in the

UK he only evidence on the age-perormance curves or offshore installations is based upon

the Danish experience so these have been used to construct the projections Further the profile

o investment assumes a progressive increase in additions to capacity rom 2013 to 2016 and

then a stable level o investment rom 2017 to 2020 he DECC road map involves a very large

commitment to offshore wind investment ndash in the range 25-35 GW o new capacity annually

rom 2016 onwards As a consequence the average age o offshore capacity in 2020 will be quite

low and the ull effects o the age-perormance curves will only be elt during the decade ollow-

ing 2020 Even so the average normalised load actor in 2020 will be about 15 lower than or

2007ndash11 while by 2025 it will be about 30 below the base level

36 On these projections 232 GW o offshore capacity is the minimum necessary to generate 58

Wh o electricity rom offshore wind in 2020 his requirement will increase to 298 GW by

2025 simply to offset the effect o the ageing o offshore generation capacity on total output

37 Combining the estimates or onshore and offshore wind generation the total amount o wind

capacity required in 2020 to produce 90 Wh o electricity rom wind will be in the range 39

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 21

to 41 GW once allowance is made or the observed age-perormance curves or both onshore

and offshore installations hese estimates are roughly a third higher than official predictions o

28ndash31 GW o wind capacity in 2020

38 urning the figures round gives a slightly different picture o the reality o the governmentrsquos

strategy on renewable electricity generation A total o 30 GW o wind capacity in 2020 (12GW onshore 18 GW offshore) will generate about 64 Wh o electricity in a normal year

Since electricity generation rom wind and biomass are expected to account or 50 o renew-

able energy in 2020 the consequence is that biomass is expected to account or 53 Wh In

2011 landfill gas and sewage sludge digestion accounted or 44 o electricity rom bioenergy

(DUKES ables 64) here are limited opportunities to expand production rom these sources

so a continuation o historic rates o growth would mean that renewable generation rom other

sources o bioenergy would have to increase rom 72 Wh in 2011 to 46 Wh in 2020 here

are severe constraints on increasing the amounts o electricity generated rom animal biomass

biodegradable municipal solid waste and anaerobic digestion so it would be optimistic to

assume that these will contribute more than 6-8 Wh in 2020 (up rom 26 Wh in 2011)

39 In practice i the projections or renewable energy are to be believed they rest upon an assump-

tion that there will be a large increase in the amount o electricity that is generated rom plant

biomass ndash either on its own or through co-firing with coal his increase will involve increas-

ing the amount o electricity generated in these ways rom 46 Wh in 2011 to 38-40 Wh in

2020 Since the amounts o straw and UK-grown timber available or this purpose are strictly

limited this will come down to the amount o wood chips that can be imported and the cost o

doing this40 he historic average load actor or biomass plants is just over 60 so about 7500 MW o

biomass capacity will be required to supply 40 Wh o electricity he weight o wood chips per

MWh o electricity depends upon moisture content storage boiler design and other actors but

an indicative range is 085 to 115 tonnes per MWh o wood chips with a moisture content o

30 According to DECC and Forestry Commission estimates the UK imported 265 million

tonnes o steam coal and 26 million tonnes o other wood products (including wood chips)

in 2011 It seems that the net effect o the governmentrsquos strategy will be to replace 15 million

tonnes o coal imported in 2011 by 32 to 44 million tonnes o imported wood chips his is a

somewhat strange way o increasing energy security

Conclusion

41 Wind power is a highly capital-intensive technology or generating electricity Its merits rely

entirely upon a substitution o capital or uel inputs he same is true or hydro or tidal or wave

power In comparison with hydro power wind is a low quality resource because o its variability

and because it cannot be stored So the economic case or wind power must rest on obtaining

the most out o the wind that is available

42 While the decline in the achieved perormance o onshore wind turbines in Denmark is

much less than that or the UK or offshore nonetheless the decline in expected output under

standardised wind conditions over 10 years is 10 unweighted and 13 capacity weighted

hese declines accelerate afer age 10 so that the reductions in perormance are 17 and 20

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2452

22 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

respectively afer 15 years For UK onshore wind arms the reduction in perormance due to

age is much worse at 27 unweighted and 69 capacity weighted by age 10

43 Evidence on the perormance o Danish offshore installations is both restricted and so poor

that there may be concern that the results are affected by a small number o outliers Still the

sample contains a reasonable number o sites with at least 5 years o operating experience andthe decline in perormance by age 5 is 38 unweighted and 26 capacity weighted

44 In addition to these results there is strong evidence that the average normalised load actor or

new onshore wind installations in the UK has allen significantly over the period rom 2000 to

2011 his is consistent with a pattern in which the most avourable sites are developed first

Equally it could mean that wind developers have been unable to keep up with the rate o new

investment while maintaining the quality o development and operations For example the site

design or selection o turbine characteristics may make less effective use o the available wind

resources or the sites available than was the case in the past

45 Whatever the reasons the deterioration in initial perormance means that the expected returns

rom the expansion in wind capacity both or investors and in terms o the reduction in CO983090

emissions have been alling without a concomitant decrease in the private and social costs

that are borne by customers and the general public Clearly this is unsatisactory at best and it

suggests that the benefits claimed or current policies cannot be taken at ace value

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 23

Appendix Data and Methods

A Data for the United Kingdom

he raw data used or this study was extracted by the Renewable Energy Foundation rom the

Renewables and CHP Register database compiled by Ogem12 his inormation is used in the

administration o the market in Renewable Obligation Certificates (ROCs) he Renewables

Obligation is the primary mechanism by which large scale generators o renewable electric-

ity receive subsidies so the operators o wind arms have a very strong incentive to submit

complete and up-to-date inormation to the database he data extracted covered all onshore

wind generators with at least 05 MW o generating capacity which are eligible or ROCs or the

10 year period April 2002 to March 2012 he key variables are the monthly output o electric-

ity (used or the purpose o allocating ROCs) and nominal generating capacity here are 282

reporting units in the final dataset A reporting unit may correspond to an entire wind arm orto separate phases o development Complications arise when an existing wind arm is re-pow-

ered ie when old wind turbines are replaced by newer and usually more powerul turbines

he database reports this as a change in the generating capacity o the recording unit but in

this analysis repowering is treated as the termination o the old recording unit and the creation

o a new recording unit

A considerable amount o data checking and cleaning was required beore the data could

be analysed he first step was to add the age o each reporting unit his was calculated by

reerence to the month in which the turbines were commissioned For wind arms in operation

in or beore June 2002 the commissioning date was obtained rom various online sources ndash

notably the thewindpowernet database o UK wind arms13 which was cross-checked against

the inormation provided on the websites o wind operators and other sources For recording

units whose first data relates to dates afer June 2002 the commissioning date was initially

assumed to be the month or which data is first reported Cross-checks were carried out or

all recording units which commenced reporting in the period July 2002 to December 2003 In

more than 80 o cases the externally-reported commissioning date is within one month o the

first month or which data is reported he exceptions ndash Bu Farm and Mablethorpe ndash appear to

be due to delays by small operators in reporting data to Ogem

In a number o cases it is clear rom the data that plants were not in ull operation at the

time when data was first reported to the database here are two indicators o partial or delayed

commissioning

bull For an initial series o months the calculated load actor is very low ndash usually below 5 ndash

and then increases abruptly In such cases the data or the early months is dropped and the

commissioning date is treated as the month immediately preceding the first month with a

lsquonormalrsquo load actor14

12 httpswwwrenewablesandchpogemgovuk

13 See httpwwwthewindpowernetwindarms_list_enphp

14 he month in which a plant is reported as having been commissioned is not included in the analysis

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2652

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 25

the decline in the plantrsquos load actor as it ages As an example consider a plant whose reported

capacity was increased rom 99 MW to 10 MW at age 3 I this was simply a change in report-

ing practice this would reduce the estimated load actor rom age 3 onwards By increasing

the initial capacity rom 99 MW to 10 MW ndash or by reducing the capacity rom 10 MW to 99

MW rom age 3 onwards ndash this bias is removed It is not important which adjustment is madebecause the difference between the alternative assumptions is captured by the unit fixed effect

he numbers o separate sites and o observations per site mean that it is not possible to

provide a simple visual representation o the data Figure 8 provides a summary by illustrating

the distributions o load actors by the age o the wind arm pooled across all years he ranges

o the load actors or each age are wide while the medians show no clear trend by age hus a

naiumlve analysis might conclude that there is nothing to investigate However as will be explained

below the distributions mask crucial differences between the perormance o wind arms as

they age because they do not control or the differences in location and wind availability over

time

B Data for Denmark

he data or Danish wind arms used in this study comes rom a database compiled by the

Danish Energy Agency covering the characteristics and perormance o all wind turbines rom

2002 up to the end o August 201216 he basic recording unit or the register is the individual

wind turbine but where a number o turbines are connected to a single meter an identical

average output per turbine is recorded or all turbines in the group he presence o what are in

effect repeated observations increases the noise in the data In addition the model that is used

or the analysis assumes that site operator and turbine characteristics are common or all o theunits that make up a single wind arm o address this point separate wind arms have been

identified by grouping all turbines which have identical values or turbine manuacturer and

type local authority date o commissioning and date o decommissioning (where applicable)

he register contains inormation on 7569 turbines commissioned rom 1977 onwards his

includes a large number o small turbines which would not all within the scope o the ROC

scheme in the UK Hence or comparability with the UK data all turbines commissioned

prior to 1992 were dropped as well as all wind arms with a generating capacity o less than 1

MW he resulting sample covered 823 onshore wind arms and 30 offshore wind arms with

monthly load actors or various periods rom January 2002 to August 2012 Given the cut-off

or wind arms commissioned prior to 1992 the maximum age or an onshore wind arm was 20

years and the maximum age or an offshore wind arm was 17 years

he total capacity o the onshore wind arms recorded in the Danish database was 2757 MW

he average size o an onshore wind arm in Denmark is small ndash only 33 MW as compared

with 161 MW or the UK ndash because 84 o the sample consists o installations with between

1 and 4 turbines he largest onshore installation has 39 turbines with a capacity o 234 MW

while the onshore installation with the greatest capacity has 19 turbines with a capacity o 584

MW Overall the sample o Danish onshore wind arms is older and smaller than the sample

o UK onshore wind arms reflecting the much longer history o wind generation in Denmark

16 httpwwwensdkEN-USINFOFACSANDFIGURESENERGY_SAISICS_AND_INDICAORS

OVERVIEWOFHEENERGYSECORREGISEROFWINDURBINESSiderForsideaspx

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2852

26 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

he larger scale o UK wind arms may account or some o the differences in perormance

curves or the two countries

he sample o offshore wind arms in Denmark is quite small with 30 installations in

total Still it is useul to analyse the perormance o these installations since there is very little

comparable data or the UK he sample covers 394 turbines with an average o just over 13turbines per installation and an average capacity o 288 MW he largest installation which

was commissioned in 2003 has 72 turbines with a total capacity o 1656 MW Again offshore

installations are smaller and older in Denmark than is the current pattern in the UK he first

two wind arms in the sample were commissioned in the years 1995 and 2000 but the sample

increases significantly rom 2002 onwards o avoid reliance upon a sample o 1 or 2 wind

arms to determine specific coefficients or ages gt 10 years all observations with an age greater

than 10 were omitted rom the analysis17

A somewhat different problem with the offshore wind arm data arose because the coeffi-

cients estimated using capacity weights (see below) were heavily influenced by the largest windarm his has had an extremely erratic profile o perormance over time with a very low average

load actor he treatment o this wind arm is discussed in more detail in Section F below

C Specication and estimation methods

As Figure 8 illustrates it is necessary to go beyond simple summary statistics in order to assess

whether there is any systematic relationship between age and the average perormance o wind

arms he perormance o wind arms varies over both time and space some months and years

have more or less wind than the long term average while specific characteristics o wind arms

ndash including location type o turbine and operating regime ndash will influence the perormance oeach plant under identical wind conditions

he combination o (a) site-specific characteristics which are constant over time and (b)

period-specific characteristics which are constant across wind arms can be represented by

what statisticians call an error components model with fixed effects or each wind arm and

each time period he dependent variable in the model consists o a sequence o load actors ndash

denoted by or unit i in month t ndash or each wind arm (the panel unit) over time he load

actors are calculated by dividing the total output in MWh by (24 x number o days in month x

reported generating capacity in MW) and are multiplied by 100 to convert to percentages

he additive version o the error components model may be written as

(1)

in which denotes the age o plant i in period t and is some unction that is either a

representation o or an approximation to the normalised age-perormance relationship he

multiplicative version is similar except that is replaced by where ln() denotes the

natural logarithm he terms are wind arm or unit fixed effects which capture actors such

as location type o turbine operating regimes etc he terms are period fixed effects which

capture the influence o wind conditions over the UK or Denmark as a whole as well as seasonal

maintenance and demand Without loss o generality it may be assumed that the fixed effects are

17 For the avoidance o doubt this means that no attempt was made to estimate the effect o age on the perormance o

offshore wind arms beyond age 10 while the data or the oldest wind arms was included in the analysis up to and

including age 10

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 27

normalised so that ie all o the error components have a zero mean while the

mean o the random error is zero by definition he terms are random errors which capture

random variations in wind conditions that are not specific to the site turbine breakdowns and

other actors that are uncorrelated with either the site or the time period It is assumed that the

mean o the random error is zero while additional assumptions depend upon the method o

estimation that is adopted

he model specified in equation (1) can be estimated in a variety o ways18 A crucial issue is

whether the are or may be significantly correlated with the age terms or any other independent

regressors I it is assumed that there is no such correlation the model is usually reerred to as

a random effects model In that case the equation can be estimated by a variant o least squares

(LS) generalised least squares (GLS) generalised methods o moments (GMM) or maximum

likelihood (ML) ndash all o which should be yield parameter estimates that will converge in prob-

ability to the true parameter values as the sample size increases An alternative but less restric-

tive specification is known as the fixed effects model in which no assumption is made about

the correlation between the unit fixed effects and other regressors he limitation o the fixed

effects model is that it cannot be used to estimate parameters or regressors that vary across

panel units but are constant over time or a panel unit (time-invariant regressors) ndash eg the

year in which a wind arm was commissioned its location or its rated generating capacity he

influence o such variables is captured by the unit fixed effects

In order to avoid the risk o obtaining biased estimates o the age effects the results reported

in this note are based upon estimates using the fixed effects model he influence o time-invar-

iant regressors is examined in a second stage by regressing the estimates o the unit fixed effectson these variables he simplest method o estimating the fixed effects model is known as the

lsquowithinrsquo estimator in which the mean value or each unit is subtracted rom all observations or

that unit For example i is a general polynomial o order M in A equation (1) becomes

(2)

so that the model can be estimated by least square with the inclusion o dummy variables or

each period t he variance-covariance matrix o the coefficients have been estimated by using

(a) a robust sandwich estimator adjusted or clustering by panel unit and (b) a bootstrap esti-

mator with 400 repetitions he estimates o the standard errors will be consistent even i the

random error component is serially correlated within panel units andor has different variances

across panel units19 Since the number o panel units and the average number o time periods

per panel units are both large (other than or Danish offshore wind arms) the least square esti-

mates o the parameters will be unbiased under these assumptions

he bootstrap method is expensive to carry out but it provides a useul cross-check on

the standard errors generated by other methods Unortunately the bootstrap method o

18 See Chapter 21 o AC Cameron amp PK rivedi ndash Microeconometrics Methods and Applications (Cambridge

Cambridge University Press 2005) or a standard textbook treatment o estimation methods or panel data models

19 he estimation was carried out using the xtreg procedure in Stata Version 12 he methods o constructing the

sandwich and bootstrap estimates o the variance-covariance matrix are described in Chapters 8 amp 13 o AC

Cameron amp PK rivedi ndash Microeconometrics using Stata (Revised Edition College Station exas Stata Press 2010)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 13: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 11

simplest variant o the normalised age-perormance curve or onshore UK wind installations

together with the equivalent curves or onshore and offshore Danish installations he curves

illustrated are calculated rom the multiplicative error components model described in the

Appendix with log(load actor) as the dependent variable and a quadratic in plant age to repre-

sent the variation in plant perormance with age9 he rate o decline in perormance is greatest or offshore installations in Denmark with a all

rom load actors o over 40 at ages 0 and 1 to less than 15 by at ages 9 and 10 Onshore

installations in the UK show a more rapid rate o decline ndash 09 percentage points per year over

the first 10 years o operation ndash than is the case or Denmark though the normalised Danish

perormance curve lies below the UK curve or the first our years For the UK the normalised

load actor alls to just over 15 at age 10 and to 11 at age 15 With such low load actors it

seems likely that many wind arms will be re-powered ndash ie the turbines will be replaced ndash once

they reach the age o 10 or at most 15

Figure 1 Performance degradation due to age using equal weights

0

5

10

15

20

25

30

35

40

45

50

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

L o

a d f a c t o r ( )

Age (years)

UK onshore DK onshore DK offshore

Note Normalised load actors in Source Authorrsquos estimates

10 here are two plausible explanations or the observed decline in average load actors as wind

arms age he first is that the turbines become less efficient over time as a result o mechanical

wear and tear erosion o the turbine blades and related actors he second is that the turbines

experience more requent breakdowns and their operators take more time to bring them back

into service because they are less concerned about the perormance o older plants Both

reasons may be relevant in different circumstances and it is not possible to identiy a primary

explanation rom the data he requency o extended shutdowns does seem to increase with

age but this could be a reflection o the timing o planned maintenance operations rather than

breakdowns Whatever the cause the reduction in perormance with age is much greater than

would be expected or thermal generating plants

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1452

12 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

11 he age-perormance curves in Figure 1 are derived rom equations estimated by giving equal

weight to each wind arm irrespective o its generating capacity Figure 2 illustrates similar

curves but in this case the estimates are constructed by weighting each wind arm by its generat-

ing capacity reerred to as capacity-weighted his gives a better representation o perormance

degradation per MW o generating capacity he striking result is that the rate o decline inthe perormance o UK onshore wind installations is significantly aster when capacity weights

are used which implies that large wind arms in the UK experience a more rapid decline than

smaller ones he differences are smaller in Denmark but the capacity-weighted decline in

perormance is more rapid than the equal weights decline or Danish onshore installations

Figure 2 Performance degradation due to age using capacity weights

0

5

10

15

20

25

30

35

40

45

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

L o a d f a c t o r ( )

Age (years)

UK onshore DK onshore DK offshore

Note Normalised load actors in Source Authorrsquos estimates

Differences between wind farms

12 he specification o the statistical model means that the unit fixed effects identiy the specific

perormance characteristics o individual wind arms afer adjusting or wind availability and

age hese characteristics may reflect the site andor design o the installation as well as the

way in which it is operated A positive value or the unit fixed effect is equivalent to shifing

the perormance curve up by some percentage applied uniormly at each age while a negative

value shifs it down It ollows that the distribution o unit fixed effects across installations can

be used to compare the relative effectiveness o new installations by location date o commis-

sioning etc Care is required when making comparisons between or example UK and Danish

onshore wind installations because these have been normalised separately so the unit fixed

effects reer to different underlying averages6

6 Some care is required in interpreting the unit fixed effects By deault the average value o the unit fixed effects or

each group is zero but these averages are taken over all observations or each installation in the basic sample For the

purpose o the comparisons in Figure 3 the unit fixed effects have been adjusted so that the average values or each

group are zero when using one value or each installation

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 13

13 o give a sense o the numbers unit fixed effects o (a) 02 or (b) -02 mean that the base load

actor given the age o the plant is multiplied by (a) exp(02) = 122 or (b) exp(-02) = 082 Since

the base load actor or a plant o age 0 is 231 these fixed effects translate to load actors in

the first year o operation o (a) 282 and (b) 189 For a wind arm with 50 MW o gener-

ating capacity the difference amounts to an additional 40700 MWh o electricity output peryear Using the standard figures cited by RenewableUK this would be sufficient to supply 12300

homes or a year his range (040) is slightly greater than the interquartile range o unit fixed

effects or UK onshore wind arms (035) Again or comparison operators o thermal power

plants would regard any difference o even 10 up or down in efficiency relative to what is

expected as a striking and perhaps worrying indication o either good or bad perormance

Figure 3 Range of unit fixed effects by category

- 1

- 5

0

5

1

A d j f i x e d e f f e c t ndash q u a d r a t i c

UK onshore DK onshore DK offshore

Source Authorrsquos estimates

14 A notable eature o the distributions o the unit fixed effects that are summarised in the box

plots shown in Figure 3 is the range o variation between the best and worst perormers espe-

cially or UK onshore and Danish offshore wind arms Box plots are constructed so that the

top and bottom o the box correspond to the upper and lower quartiles while the bars at topand bottom correspond to the maximum and minimum values excluding outliers7 A plant at

the upper quartile o the distribution o UK onshore wind arms will generate an annual output

that is about 40 higher than a plant at the lower quartile he equivalent figure or Danish

onshore wind plants is an increase o 30 while or Danish offshore wind arms the difference

between the upper and lower quartiles is equivalent to 25 times the output o a lower quartile

plant By any standard such differences are important

7 he upper quartile is the point in a distribution such that the 25 o values exceed it while the lower quartile is the

point in a distribution such that 75 o values exceed it he line in the middle o the box marks the median the

point exceeded by 50 o values Outliers are defined as values that lie outside the range o ukey adjacent values

which are defined as the upperlower quartiles +- 15 x interquartile range

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1652

14 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

15 Outside the range between the upper and lower quartiles the top and bottom segments o

the distributions extend urther up and down or UK onshore wind arms than or those in

Denmark he rather different pattern or Danish offshore wind installations is partly a conse-

quence o the relatively small size o the sample but the evidence points to considerably greater

variability in the perormance o offshore wind installations than or onshore installations16 It is somewhat surprising to observe the magnitude o the differences in the perormance in

onshore wind arms in the UK Given the nature o the subsidy regimes and the high capital

cost o developing new installations it might be expected that operators have a strong incen-

tive to identiy the best locations and then choose equipment that will deliver the maximum

amount o electricity output at a high level o reliability For UK onshore operators it seems that

good or bad perormance is somewhat o a lottery However i the subsidies provided by ROCs

are sufficient to underwrite investment in inefficient plants ndash as appears to be the case ndash then

those subsidies are extremely generous or plants that operate close to the efficient rontier

As location is likely to be the main actor that determines the perormance o a specific plant

relative to all other plants the inerence must be that many wind plants have been developed on

sites with poor wind characteristics

Figure 4 Initial load factors by year of commissioning for onshore wind

10

15

20

25

30

35

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

I n i t i a l l o a d f a c t o r ( )

Year commissioned

UK onshore DK onshore Linear (UK onshore) Linear (DK onshore)

Source Authorrsquos estimates

17 here is another disturbing characteristic o UK wind arms which is revealed by examination

o the unit fixed effects and illustrated in Figure 4 It might be expected that the average peror-

mance o recently installed units would be higher than that o older units his should reflect

improvements in turbine design and reliability as well as in the identification o good sites or

wind generation However that is not the pattern or onshore wind arms in the UK he figure

shows the average values o the initial load actors calculated using equally weighted data by

year o commissioning or onshore wind arms in the UK and Denmark together with fitted

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 15

trend lines8 here has been a clear decline in the initial load actors in the UK which has not

occurred in Denmark his pattern is confirmed by more detailed statistical results shown in

the Appendix which take account o location and size o installation

18 he decline in initial perormance is particularly obvious afer 2005 when the rate o commis-

sioning new capacity increased sharply rom less than 500 MW in the 5 years 2000ndash04 to morethan 500 MW per year rom 2005 onwards In contrast the peak years or commissioning new

onshore wind arms in Denmark were 2000ndash02 when 920 MW was installed he rate o new

building (or replacement) picked up again in 2008 but it is still much lower than at the turn o

century Not only is the perormance o the onshore wind plants in the UK commissioned in

recent years significantly worse than that or wind arms commissioned beore 2005 but it is

possible that this is a direct consequence o an overly rapid expansion o capacity In essence

the evidence suggests that the industry does not have the capability to identiy develop and

operate new onshore wind arms at the rate envisaged by UK government targets while main-

taining a satisactory average level o perormance

Implications for future policy and performance

19 he pattern o perormance degradation with age and over time identified in this analysis has

implications or the assessment o new investment in wind arms and or the design o the

proposed Electricity Market Reorm (EMR) here are two related issues that are affected by

the results

a) What are the implications or the design o policies intended to promote the adoption

and expansion o wind generation in the UK his includes the structure o current

subsidies as well as the specification o the contracts that are proposed under the EMR A

related question concerns the expected lie o new wind arms

b) How might the average perormance o wind generation develop over the next decade

his is absolutely critical to any assessment o the amount o wind capacity that will be

required to meet the UK governmentrsquos targets or the share o electricity ndash and more

generally energy ndash rom renewable sources he government has assumed that the

average load actors or both onshore and offshore wind arms will either remain stable

or increase in uture he prospects will be very different i these assumptions are not

borne out

20 he first question can be addressed by considering the discounted sum o cumulative net

output rom a new wind arm on the assumption that its perormance matches the averages

identified in the analysis his takes account o both perormance degradation over time plus

the annual cost o maintaining the plant which may be expressed as a deduction rom gross

production so as to calculate the contribution o expected output at age s to recovering the

original capital cost o building the plant hese figures have to be discounted back to the date

o commissioning in order to assess the net present value o the initial investment in the plant

he curves or constant and observed perormance using a discount rate o 9 (a typical cost o

capital or wind generation) are shown in Figure 5 he black line is based on the ull set o age

8 he initial load actors are calculated as the normalised load actor or UKDanish wind arms at age 0 multiplied by

where is the unit fixed effect or installation i

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1852

16 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

effects while the red dashed line is based on the smoothed quadratic perormance curve and

the blue dashed line shows what the pattern would be i operating perormance was not affected

by age he calculation is extended out to age 25 since DECCrsquos standard comparisons assume

that wind turbines have a lie o 25 years

Figure 5 Impact of performance degradation on discounted cumulative output

0

10

20

30

40

50

60

70

80

90

100

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

o

f d i s c o u n t e d c u m u l a t i v e o u t p u t

Age of plant (years)

Constant Quadratic Full age effects

Source Authorrsquos estimates

21 he key points in the figure are that 80 o the discounted cumulative output o a new wind

plant is likely to be produced in the first 10 years o its lie and 90 in the first 14 years his

is consistent with the structure o Danish subsidies or onshore wind arms which extend over

a typical period o 125 years Since sites or wind arms are scarce and involve the payment o

significant rents that may be linked to output it is very unlikely that any new installation o

wind turbines will have an expected lie o more than 15 years Instead as has happened in the

past wind operators will have a strong incentive to decommission plants afer no more than 15

years and replace the turbines with newer equipment

22 As a consequence any economic assessment o wind generation should not be based on an

expected lie which is longer than 15 years In recent work reported in evidence to the House

o Commons Select Committee on Energy and Climate Change I assumed that wind plants

would have a residual value equal to 20 o their initial cost in real terms at the end o 15 years

he analysis in this paper suggests that this is too avourable an assumption Given the costs o

decommissioning old turbines the residual value is likely to be well below 10 o their initial

cost and the decision point may be at 10 rather than 15 years

23 he corollary o this observation is that it makes little sense to offer long term contracts or 20

years or more that guarantee prices or eed-in tariffs (Fis) to wind operators A contract length

o 10 to 12 years would be sufficient to remove most o the market risk associated with invest-

ment in wind generation In this respect the subsidy arrangements implemented in Denmark

are better designed

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 17

24 At the same time the offer o subsidies andor guaranteed prices may have serious adverse

consequences or the efficiency o wind generation Returning to the unit fixed effects measur-

ing the perormance o wind plants commissioned afer 2005 only 28 out o 159 units have an

operating perormance that exceeds the average or the period hose 28 units account or 360

MW o capacity whereas the remaining 131 units account or 2810 MW o capacity Not onlyare recent plants less efficient than the average or the whole period but the plants which are

below-average in efficiency are typically larger than the more efficient ones and account or a

disproportionate share o the recent additions to generating capacity

25 In mid-2012 there was approximately 4500 MW o onshore wind capacity operating in the

UK By 2020 it is expected that there will be at least 10000 MW (based on proposed wind

arms with planning consent) and perhaps as much 15000 MW o wind capacity in operation

he government is relying upon a substantial contribution rom onshore wind arms towards

meeting its targets or renewable energy in 2020 o provide context the governmentrsquos targets

or renewable energy in 2020 speciy a total o 234 Wh o energy rom all renewable sources

o which 90ndash140 Wh is expected to come rom wind and biomass electricity generation9

he projections or biomass electricity are extremely vague and have to be reconciled with

an assumed increase o 3ndash4 times in the amount o biomass used separately or heat So in

concrete terms it is reasonable to assume that wind generation will account or at least 90 Wh

with 24ndash32 Wh rom onshore generation and the remainder rom offshore

26 he average load actor or onshore wind over the 10 year period rom April 2002 to March

2012 was 256 which is influenced by the large amounts o new capacity added rom 2005

onwards I this load actor were achieved in 2020 the total amount o onshore wind capacityrequired to meet the onshore generation target would be 107 to 143 GW he average load

actor or offshore wind installations (on an unchanged configuration basis) or the five year

period 2007ndash11 was 32010 I this load actor were achieved in 2020 the total amount o

offshore wind capacity required to meet the remainder o the 90 Wh target would be 206

to 235 GW though these estimates would all to 180 to 205 GW i the average load actor

matched the historic average value o 367 or Danish offshore wind installations

27 he question that has to be addressed is whether these load actors are consistent with (a) the

investment programmes that would be required to achieve the total amounts o generation

capacity implied by the governmentrsquos targets and (b) the age profiles o the perormance o

wind arms discussed in this paper he short answer is that this is simply not possible i the

initial load actor or new wind arms continues to decline at the rate observed the past decade

So even as a starting point it is necessary to assume that this underlying deterioration in wind

arm perormance ceases his would be a major change in itsel and it is not obvious what

actors might lead to such a result

28 he results illustrated below are based on a vintage model o perormance that takes account

o the actual expansion o capacity up to the end o 2011 plus whatever investment in new or

replacement turbines is required to achieve the target levels o wind capacity in 2020 o high-light the ull range o potential outcomes these targets are either 10 GW or 15 GW o onshore

9 Department o Energy and Climate Change ndash UK Renewable Energy Road Map July 2011 Figure 2

10 he equivalent average load actor or onshore wind arms was 264

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2052

18 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

wind and either 18 or 24 GW o offshore wind he levels o investment required to meet these

targets depend upon the expected lie o wind turbines For onshore turbines an expected lie

o 10 years is consistent with re-powering decisions over the last decade while an expected lie

o 15 years seems to be the maximum consistent with the age-perormance profile For offshore

turbines the analysis uses expected lives o 15 and 20 years

Figure 6A Projected load factors for UK onshore wind (total capacity of 10 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

Figure 6B Projected load factors for UK onshore wind (total capacity of 15 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

29 Figure 6A shows projected values o the average normalised load actors or UK onshore wind

arms in 2015 and 2020 using a target capacity in 2020 equal to 10 GW plus the age-peror-

mance profiles or the UK and Denmark (DK) Since vintage effects are important in this

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 19

calculation the profile o new or replacement investment affects the results In this case the

level o gross investment is assumed to be constant rom 2013 onwards at either 930 or 1220

MW per year ndash the range is defined by the assumed lie o turbines Achieving a target o 15

GW in 2020 (Figure 6B) would require a level o investment rom 2013 onwards o either 1540

or 1930 MW per year Such figures are 3ndash4 times the average level o investment over theperiod 2005ndash11 so the lower target may provide a better indication o what will happen

30 All o the estimates are calibrated to match the observed average load actor in 2007ndash11 so that

they should be interpreted as relative rather than absolute values Using the age-perormance

profile or UK onshore wind installations the average normalised load actor is likely to all by

about 19 rom 2007ndash11 to 2015 and by 23ndash28 to 2020 I the age-perormance curves or

Danish onshore installations were to apply the reduction in the average normalised load actor

would be considerably less but there would still be a all o up to 5 by 2020

31 In the medium term the all is greater i turbines operate or 15 years because the average age

o all turbines will be higher and thus the degradation in perormance due to age will have a

larger effect By 2020 this effect reduces the average normalised load actor rom 203 or a 10

year lie to 195 or a 15 year lie here is a complex economic and commercial trade-off here

o meet a target o 10 GW in 2020 it is necessary to increase the level o investment in onshore

capacity by 30 rom 2013 Almost all o this will take the orm o re-powering existing wind

arms11 he costs are airly high because this will usually involve new development consents

decommissioning the original turbines and replacing them with a smaller number o larger and

more powerul turbines

32 Estimates o re-powering costs are generally treated as being commercially sensitive but broadproject costs cited in specialist magazines suggest that the unit costs are likely to be at least

pound1000 per kW or more than pound300 million per year in aggregate On the other hand the increase

in the amount o energy generated would be about 12 Wh or a 10 year turbine lie rather

than a 15 year turbine lie though this does not allow or the loss o production during re-pow-

ering Under the best circumstances the payback period may be quite short but the best option

rom a broader economic and social perspective is not obvious

33 Even under the best scenario 10 GW o onshore wind capacity will only generate 178 Wh o

electricity in 2020 well below the lower end o the target range or onshore wind generation

specified in DECCrsquos road map In act allowing or interactions between the age-perormance

profile and investment the amount o onshore wind capacity required in 2020 to generate 24

Wh would be a minimum o 134 GW At the top end o DECCrsquos target range the capacity

required in 2020 to generate 32 Wh would be a minimum o 178 GW Since these estimates

are based on a 10 year turbine lie the level o investment required rom 2013 onwards would

be 1650 MW per year or the bottom o the range and 2350 MW or the top end o the range

In the last 8 years total investment in onshore wind has only ever exceeded 600 MW in one

11 Re-powering is usually defined as the ull replacement o existing turbines whereas retrofitting covers the replace-

ment o various mechanical or electrical components In the data analysis re-powering involves the creation o a

new unit whereas retrofitting affects the perormance o a continuing unit Hence the potential benefits o retrofit-

ting are taken into account in constructing the age-perormance curves

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2252

20 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

year (883 MW in 2008) so it seems rather unlikely that even the lower end o DECCrsquos range or

onshore wind generation will actually be realised

34 O course this assessment would be different i UK onshore wind operators were able to match

the Danish age-perormance curve here are many actors which differentiate the UK and

Danish wind industries he average scale o wind arms in the UK is larger they are moreremote and there appears to be significantly greater variation between operating and potential

new sites in achieved perormance It would thereore require a very optimistic observer to

conclude that the difference in age-perormance curves between the two countries will disap-

pear in the short or medium term

Figure 7 Projected load factors for UK offshore wind in 2020

20

22

24

26

28

30

32

34

18 GW ndash life 15 yrs 18 GW ndash life 20 yrs 24 GW ndash life 15 yrs 24 GW ndash life 20 yrs

L o a d f a c t o r ( )

2007ndash11 2015 2020

Note Based on Danish age-perormance curves Source Authorrsquos estimates

35 Figure 7 shows the results rom carrying out a similar exercise or offshore wind arms in the

UK he only evidence on the age-perormance curves or offshore installations is based upon

the Danish experience so these have been used to construct the projections Further the profile

o investment assumes a progressive increase in additions to capacity rom 2013 to 2016 and

then a stable level o investment rom 2017 to 2020 he DECC road map involves a very large

commitment to offshore wind investment ndash in the range 25-35 GW o new capacity annually

rom 2016 onwards As a consequence the average age o offshore capacity in 2020 will be quite

low and the ull effects o the age-perormance curves will only be elt during the decade ollow-

ing 2020 Even so the average normalised load actor in 2020 will be about 15 lower than or

2007ndash11 while by 2025 it will be about 30 below the base level

36 On these projections 232 GW o offshore capacity is the minimum necessary to generate 58

Wh o electricity rom offshore wind in 2020 his requirement will increase to 298 GW by

2025 simply to offset the effect o the ageing o offshore generation capacity on total output

37 Combining the estimates or onshore and offshore wind generation the total amount o wind

capacity required in 2020 to produce 90 Wh o electricity rom wind will be in the range 39

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 21

to 41 GW once allowance is made or the observed age-perormance curves or both onshore

and offshore installations hese estimates are roughly a third higher than official predictions o

28ndash31 GW o wind capacity in 2020

38 urning the figures round gives a slightly different picture o the reality o the governmentrsquos

strategy on renewable electricity generation A total o 30 GW o wind capacity in 2020 (12GW onshore 18 GW offshore) will generate about 64 Wh o electricity in a normal year

Since electricity generation rom wind and biomass are expected to account or 50 o renew-

able energy in 2020 the consequence is that biomass is expected to account or 53 Wh In

2011 landfill gas and sewage sludge digestion accounted or 44 o electricity rom bioenergy

(DUKES ables 64) here are limited opportunities to expand production rom these sources

so a continuation o historic rates o growth would mean that renewable generation rom other

sources o bioenergy would have to increase rom 72 Wh in 2011 to 46 Wh in 2020 here

are severe constraints on increasing the amounts o electricity generated rom animal biomass

biodegradable municipal solid waste and anaerobic digestion so it would be optimistic to

assume that these will contribute more than 6-8 Wh in 2020 (up rom 26 Wh in 2011)

39 In practice i the projections or renewable energy are to be believed they rest upon an assump-

tion that there will be a large increase in the amount o electricity that is generated rom plant

biomass ndash either on its own or through co-firing with coal his increase will involve increas-

ing the amount o electricity generated in these ways rom 46 Wh in 2011 to 38-40 Wh in

2020 Since the amounts o straw and UK-grown timber available or this purpose are strictly

limited this will come down to the amount o wood chips that can be imported and the cost o

doing this40 he historic average load actor or biomass plants is just over 60 so about 7500 MW o

biomass capacity will be required to supply 40 Wh o electricity he weight o wood chips per

MWh o electricity depends upon moisture content storage boiler design and other actors but

an indicative range is 085 to 115 tonnes per MWh o wood chips with a moisture content o

30 According to DECC and Forestry Commission estimates the UK imported 265 million

tonnes o steam coal and 26 million tonnes o other wood products (including wood chips)

in 2011 It seems that the net effect o the governmentrsquos strategy will be to replace 15 million

tonnes o coal imported in 2011 by 32 to 44 million tonnes o imported wood chips his is a

somewhat strange way o increasing energy security

Conclusion

41 Wind power is a highly capital-intensive technology or generating electricity Its merits rely

entirely upon a substitution o capital or uel inputs he same is true or hydro or tidal or wave

power In comparison with hydro power wind is a low quality resource because o its variability

and because it cannot be stored So the economic case or wind power must rest on obtaining

the most out o the wind that is available

42 While the decline in the achieved perormance o onshore wind turbines in Denmark is

much less than that or the UK or offshore nonetheless the decline in expected output under

standardised wind conditions over 10 years is 10 unweighted and 13 capacity weighted

hese declines accelerate afer age 10 so that the reductions in perormance are 17 and 20

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2452

22 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

respectively afer 15 years For UK onshore wind arms the reduction in perormance due to

age is much worse at 27 unweighted and 69 capacity weighted by age 10

43 Evidence on the perormance o Danish offshore installations is both restricted and so poor

that there may be concern that the results are affected by a small number o outliers Still the

sample contains a reasonable number o sites with at least 5 years o operating experience andthe decline in perormance by age 5 is 38 unweighted and 26 capacity weighted

44 In addition to these results there is strong evidence that the average normalised load actor or

new onshore wind installations in the UK has allen significantly over the period rom 2000 to

2011 his is consistent with a pattern in which the most avourable sites are developed first

Equally it could mean that wind developers have been unable to keep up with the rate o new

investment while maintaining the quality o development and operations For example the site

design or selection o turbine characteristics may make less effective use o the available wind

resources or the sites available than was the case in the past

45 Whatever the reasons the deterioration in initial perormance means that the expected returns

rom the expansion in wind capacity both or investors and in terms o the reduction in CO983090

emissions have been alling without a concomitant decrease in the private and social costs

that are borne by customers and the general public Clearly this is unsatisactory at best and it

suggests that the benefits claimed or current policies cannot be taken at ace value

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 23

Appendix Data and Methods

A Data for the United Kingdom

he raw data used or this study was extracted by the Renewable Energy Foundation rom the

Renewables and CHP Register database compiled by Ogem12 his inormation is used in the

administration o the market in Renewable Obligation Certificates (ROCs) he Renewables

Obligation is the primary mechanism by which large scale generators o renewable electric-

ity receive subsidies so the operators o wind arms have a very strong incentive to submit

complete and up-to-date inormation to the database he data extracted covered all onshore

wind generators with at least 05 MW o generating capacity which are eligible or ROCs or the

10 year period April 2002 to March 2012 he key variables are the monthly output o electric-

ity (used or the purpose o allocating ROCs) and nominal generating capacity here are 282

reporting units in the final dataset A reporting unit may correspond to an entire wind arm orto separate phases o development Complications arise when an existing wind arm is re-pow-

ered ie when old wind turbines are replaced by newer and usually more powerul turbines

he database reports this as a change in the generating capacity o the recording unit but in

this analysis repowering is treated as the termination o the old recording unit and the creation

o a new recording unit

A considerable amount o data checking and cleaning was required beore the data could

be analysed he first step was to add the age o each reporting unit his was calculated by

reerence to the month in which the turbines were commissioned For wind arms in operation

in or beore June 2002 the commissioning date was obtained rom various online sources ndash

notably the thewindpowernet database o UK wind arms13 which was cross-checked against

the inormation provided on the websites o wind operators and other sources For recording

units whose first data relates to dates afer June 2002 the commissioning date was initially

assumed to be the month or which data is first reported Cross-checks were carried out or

all recording units which commenced reporting in the period July 2002 to December 2003 In

more than 80 o cases the externally-reported commissioning date is within one month o the

first month or which data is reported he exceptions ndash Bu Farm and Mablethorpe ndash appear to

be due to delays by small operators in reporting data to Ogem

In a number o cases it is clear rom the data that plants were not in ull operation at the

time when data was first reported to the database here are two indicators o partial or delayed

commissioning

bull For an initial series o months the calculated load actor is very low ndash usually below 5 ndash

and then increases abruptly In such cases the data or the early months is dropped and the

commissioning date is treated as the month immediately preceding the first month with a

lsquonormalrsquo load actor14

12 httpswwwrenewablesandchpogemgovuk

13 See httpwwwthewindpowernetwindarms_list_enphp

14 he month in which a plant is reported as having been commissioned is not included in the analysis

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2652

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 25

the decline in the plantrsquos load actor as it ages As an example consider a plant whose reported

capacity was increased rom 99 MW to 10 MW at age 3 I this was simply a change in report-

ing practice this would reduce the estimated load actor rom age 3 onwards By increasing

the initial capacity rom 99 MW to 10 MW ndash or by reducing the capacity rom 10 MW to 99

MW rom age 3 onwards ndash this bias is removed It is not important which adjustment is madebecause the difference between the alternative assumptions is captured by the unit fixed effect

he numbers o separate sites and o observations per site mean that it is not possible to

provide a simple visual representation o the data Figure 8 provides a summary by illustrating

the distributions o load actors by the age o the wind arm pooled across all years he ranges

o the load actors or each age are wide while the medians show no clear trend by age hus a

naiumlve analysis might conclude that there is nothing to investigate However as will be explained

below the distributions mask crucial differences between the perormance o wind arms as

they age because they do not control or the differences in location and wind availability over

time

B Data for Denmark

he data or Danish wind arms used in this study comes rom a database compiled by the

Danish Energy Agency covering the characteristics and perormance o all wind turbines rom

2002 up to the end o August 201216 he basic recording unit or the register is the individual

wind turbine but where a number o turbines are connected to a single meter an identical

average output per turbine is recorded or all turbines in the group he presence o what are in

effect repeated observations increases the noise in the data In addition the model that is used

or the analysis assumes that site operator and turbine characteristics are common or all o theunits that make up a single wind arm o address this point separate wind arms have been

identified by grouping all turbines which have identical values or turbine manuacturer and

type local authority date o commissioning and date o decommissioning (where applicable)

he register contains inormation on 7569 turbines commissioned rom 1977 onwards his

includes a large number o small turbines which would not all within the scope o the ROC

scheme in the UK Hence or comparability with the UK data all turbines commissioned

prior to 1992 were dropped as well as all wind arms with a generating capacity o less than 1

MW he resulting sample covered 823 onshore wind arms and 30 offshore wind arms with

monthly load actors or various periods rom January 2002 to August 2012 Given the cut-off

or wind arms commissioned prior to 1992 the maximum age or an onshore wind arm was 20

years and the maximum age or an offshore wind arm was 17 years

he total capacity o the onshore wind arms recorded in the Danish database was 2757 MW

he average size o an onshore wind arm in Denmark is small ndash only 33 MW as compared

with 161 MW or the UK ndash because 84 o the sample consists o installations with between

1 and 4 turbines he largest onshore installation has 39 turbines with a capacity o 234 MW

while the onshore installation with the greatest capacity has 19 turbines with a capacity o 584

MW Overall the sample o Danish onshore wind arms is older and smaller than the sample

o UK onshore wind arms reflecting the much longer history o wind generation in Denmark

16 httpwwwensdkEN-USINFOFACSANDFIGURESENERGY_SAISICS_AND_INDICAORS

OVERVIEWOFHEENERGYSECORREGISEROFWINDURBINESSiderForsideaspx

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2852

26 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

he larger scale o UK wind arms may account or some o the differences in perormance

curves or the two countries

he sample o offshore wind arms in Denmark is quite small with 30 installations in

total Still it is useul to analyse the perormance o these installations since there is very little

comparable data or the UK he sample covers 394 turbines with an average o just over 13turbines per installation and an average capacity o 288 MW he largest installation which

was commissioned in 2003 has 72 turbines with a total capacity o 1656 MW Again offshore

installations are smaller and older in Denmark than is the current pattern in the UK he first

two wind arms in the sample were commissioned in the years 1995 and 2000 but the sample

increases significantly rom 2002 onwards o avoid reliance upon a sample o 1 or 2 wind

arms to determine specific coefficients or ages gt 10 years all observations with an age greater

than 10 were omitted rom the analysis17

A somewhat different problem with the offshore wind arm data arose because the coeffi-

cients estimated using capacity weights (see below) were heavily influenced by the largest windarm his has had an extremely erratic profile o perormance over time with a very low average

load actor he treatment o this wind arm is discussed in more detail in Section F below

C Specication and estimation methods

As Figure 8 illustrates it is necessary to go beyond simple summary statistics in order to assess

whether there is any systematic relationship between age and the average perormance o wind

arms he perormance o wind arms varies over both time and space some months and years

have more or less wind than the long term average while specific characteristics o wind arms

ndash including location type o turbine and operating regime ndash will influence the perormance oeach plant under identical wind conditions

he combination o (a) site-specific characteristics which are constant over time and (b)

period-specific characteristics which are constant across wind arms can be represented by

what statisticians call an error components model with fixed effects or each wind arm and

each time period he dependent variable in the model consists o a sequence o load actors ndash

denoted by or unit i in month t ndash or each wind arm (the panel unit) over time he load

actors are calculated by dividing the total output in MWh by (24 x number o days in month x

reported generating capacity in MW) and are multiplied by 100 to convert to percentages

he additive version o the error components model may be written as

(1)

in which denotes the age o plant i in period t and is some unction that is either a

representation o or an approximation to the normalised age-perormance relationship he

multiplicative version is similar except that is replaced by where ln() denotes the

natural logarithm he terms are wind arm or unit fixed effects which capture actors such

as location type o turbine operating regimes etc he terms are period fixed effects which

capture the influence o wind conditions over the UK or Denmark as a whole as well as seasonal

maintenance and demand Without loss o generality it may be assumed that the fixed effects are

17 For the avoidance o doubt this means that no attempt was made to estimate the effect o age on the perormance o

offshore wind arms beyond age 10 while the data or the oldest wind arms was included in the analysis up to and

including age 10

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 27

normalised so that ie all o the error components have a zero mean while the

mean o the random error is zero by definition he terms are random errors which capture

random variations in wind conditions that are not specific to the site turbine breakdowns and

other actors that are uncorrelated with either the site or the time period It is assumed that the

mean o the random error is zero while additional assumptions depend upon the method o

estimation that is adopted

he model specified in equation (1) can be estimated in a variety o ways18 A crucial issue is

whether the are or may be significantly correlated with the age terms or any other independent

regressors I it is assumed that there is no such correlation the model is usually reerred to as

a random effects model In that case the equation can be estimated by a variant o least squares

(LS) generalised least squares (GLS) generalised methods o moments (GMM) or maximum

likelihood (ML) ndash all o which should be yield parameter estimates that will converge in prob-

ability to the true parameter values as the sample size increases An alternative but less restric-

tive specification is known as the fixed effects model in which no assumption is made about

the correlation between the unit fixed effects and other regressors he limitation o the fixed

effects model is that it cannot be used to estimate parameters or regressors that vary across

panel units but are constant over time or a panel unit (time-invariant regressors) ndash eg the

year in which a wind arm was commissioned its location or its rated generating capacity he

influence o such variables is captured by the unit fixed effects

In order to avoid the risk o obtaining biased estimates o the age effects the results reported

in this note are based upon estimates using the fixed effects model he influence o time-invar-

iant regressors is examined in a second stage by regressing the estimates o the unit fixed effectson these variables he simplest method o estimating the fixed effects model is known as the

lsquowithinrsquo estimator in which the mean value or each unit is subtracted rom all observations or

that unit For example i is a general polynomial o order M in A equation (1) becomes

(2)

so that the model can be estimated by least square with the inclusion o dummy variables or

each period t he variance-covariance matrix o the coefficients have been estimated by using

(a) a robust sandwich estimator adjusted or clustering by panel unit and (b) a bootstrap esti-

mator with 400 repetitions he estimates o the standard errors will be consistent even i the

random error component is serially correlated within panel units andor has different variances

across panel units19 Since the number o panel units and the average number o time periods

per panel units are both large (other than or Danish offshore wind arms) the least square esti-

mates o the parameters will be unbiased under these assumptions

he bootstrap method is expensive to carry out but it provides a useul cross-check on

the standard errors generated by other methods Unortunately the bootstrap method o

18 See Chapter 21 o AC Cameron amp PK rivedi ndash Microeconometrics Methods and Applications (Cambridge

Cambridge University Press 2005) or a standard textbook treatment o estimation methods or panel data models

19 he estimation was carried out using the xtreg procedure in Stata Version 12 he methods o constructing the

sandwich and bootstrap estimates o the variance-covariance matrix are described in Chapters 8 amp 13 o AC

Cameron amp PK rivedi ndash Microeconometrics using Stata (Revised Edition College Station exas Stata Press 2010)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 14: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1452

12 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

11 he age-perormance curves in Figure 1 are derived rom equations estimated by giving equal

weight to each wind arm irrespective o its generating capacity Figure 2 illustrates similar

curves but in this case the estimates are constructed by weighting each wind arm by its generat-

ing capacity reerred to as capacity-weighted his gives a better representation o perormance

degradation per MW o generating capacity he striking result is that the rate o decline inthe perormance o UK onshore wind installations is significantly aster when capacity weights

are used which implies that large wind arms in the UK experience a more rapid decline than

smaller ones he differences are smaller in Denmark but the capacity-weighted decline in

perormance is more rapid than the equal weights decline or Danish onshore installations

Figure 2 Performance degradation due to age using capacity weights

0

5

10

15

20

25

30

35

40

45

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

L o a d f a c t o r ( )

Age (years)

UK onshore DK onshore DK offshore

Note Normalised load actors in Source Authorrsquos estimates

Differences between wind farms

12 he specification o the statistical model means that the unit fixed effects identiy the specific

perormance characteristics o individual wind arms afer adjusting or wind availability and

age hese characteristics may reflect the site andor design o the installation as well as the

way in which it is operated A positive value or the unit fixed effect is equivalent to shifing

the perormance curve up by some percentage applied uniormly at each age while a negative

value shifs it down It ollows that the distribution o unit fixed effects across installations can

be used to compare the relative effectiveness o new installations by location date o commis-

sioning etc Care is required when making comparisons between or example UK and Danish

onshore wind installations because these have been normalised separately so the unit fixed

effects reer to different underlying averages6

6 Some care is required in interpreting the unit fixed effects By deault the average value o the unit fixed effects or

each group is zero but these averages are taken over all observations or each installation in the basic sample For the

purpose o the comparisons in Figure 3 the unit fixed effects have been adjusted so that the average values or each

group are zero when using one value or each installation

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 13

13 o give a sense o the numbers unit fixed effects o (a) 02 or (b) -02 mean that the base load

actor given the age o the plant is multiplied by (a) exp(02) = 122 or (b) exp(-02) = 082 Since

the base load actor or a plant o age 0 is 231 these fixed effects translate to load actors in

the first year o operation o (a) 282 and (b) 189 For a wind arm with 50 MW o gener-

ating capacity the difference amounts to an additional 40700 MWh o electricity output peryear Using the standard figures cited by RenewableUK this would be sufficient to supply 12300

homes or a year his range (040) is slightly greater than the interquartile range o unit fixed

effects or UK onshore wind arms (035) Again or comparison operators o thermal power

plants would regard any difference o even 10 up or down in efficiency relative to what is

expected as a striking and perhaps worrying indication o either good or bad perormance

Figure 3 Range of unit fixed effects by category

- 1

- 5

0

5

1

A d j f i x e d e f f e c t ndash q u a d r a t i c

UK onshore DK onshore DK offshore

Source Authorrsquos estimates

14 A notable eature o the distributions o the unit fixed effects that are summarised in the box

plots shown in Figure 3 is the range o variation between the best and worst perormers espe-

cially or UK onshore and Danish offshore wind arms Box plots are constructed so that the

top and bottom o the box correspond to the upper and lower quartiles while the bars at topand bottom correspond to the maximum and minimum values excluding outliers7 A plant at

the upper quartile o the distribution o UK onshore wind arms will generate an annual output

that is about 40 higher than a plant at the lower quartile he equivalent figure or Danish

onshore wind plants is an increase o 30 while or Danish offshore wind arms the difference

between the upper and lower quartiles is equivalent to 25 times the output o a lower quartile

plant By any standard such differences are important

7 he upper quartile is the point in a distribution such that the 25 o values exceed it while the lower quartile is the

point in a distribution such that 75 o values exceed it he line in the middle o the box marks the median the

point exceeded by 50 o values Outliers are defined as values that lie outside the range o ukey adjacent values

which are defined as the upperlower quartiles +- 15 x interquartile range

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1652

14 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

15 Outside the range between the upper and lower quartiles the top and bottom segments o

the distributions extend urther up and down or UK onshore wind arms than or those in

Denmark he rather different pattern or Danish offshore wind installations is partly a conse-

quence o the relatively small size o the sample but the evidence points to considerably greater

variability in the perormance o offshore wind installations than or onshore installations16 It is somewhat surprising to observe the magnitude o the differences in the perormance in

onshore wind arms in the UK Given the nature o the subsidy regimes and the high capital

cost o developing new installations it might be expected that operators have a strong incen-

tive to identiy the best locations and then choose equipment that will deliver the maximum

amount o electricity output at a high level o reliability For UK onshore operators it seems that

good or bad perormance is somewhat o a lottery However i the subsidies provided by ROCs

are sufficient to underwrite investment in inefficient plants ndash as appears to be the case ndash then

those subsidies are extremely generous or plants that operate close to the efficient rontier

As location is likely to be the main actor that determines the perormance o a specific plant

relative to all other plants the inerence must be that many wind plants have been developed on

sites with poor wind characteristics

Figure 4 Initial load factors by year of commissioning for onshore wind

10

15

20

25

30

35

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

I n i t i a l l o a d f a c t o r ( )

Year commissioned

UK onshore DK onshore Linear (UK onshore) Linear (DK onshore)

Source Authorrsquos estimates

17 here is another disturbing characteristic o UK wind arms which is revealed by examination

o the unit fixed effects and illustrated in Figure 4 It might be expected that the average peror-

mance o recently installed units would be higher than that o older units his should reflect

improvements in turbine design and reliability as well as in the identification o good sites or

wind generation However that is not the pattern or onshore wind arms in the UK he figure

shows the average values o the initial load actors calculated using equally weighted data by

year o commissioning or onshore wind arms in the UK and Denmark together with fitted

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 15

trend lines8 here has been a clear decline in the initial load actors in the UK which has not

occurred in Denmark his pattern is confirmed by more detailed statistical results shown in

the Appendix which take account o location and size o installation

18 he decline in initial perormance is particularly obvious afer 2005 when the rate o commis-

sioning new capacity increased sharply rom less than 500 MW in the 5 years 2000ndash04 to morethan 500 MW per year rom 2005 onwards In contrast the peak years or commissioning new

onshore wind arms in Denmark were 2000ndash02 when 920 MW was installed he rate o new

building (or replacement) picked up again in 2008 but it is still much lower than at the turn o

century Not only is the perormance o the onshore wind plants in the UK commissioned in

recent years significantly worse than that or wind arms commissioned beore 2005 but it is

possible that this is a direct consequence o an overly rapid expansion o capacity In essence

the evidence suggests that the industry does not have the capability to identiy develop and

operate new onshore wind arms at the rate envisaged by UK government targets while main-

taining a satisactory average level o perormance

Implications for future policy and performance

19 he pattern o perormance degradation with age and over time identified in this analysis has

implications or the assessment o new investment in wind arms and or the design o the

proposed Electricity Market Reorm (EMR) here are two related issues that are affected by

the results

a) What are the implications or the design o policies intended to promote the adoption

and expansion o wind generation in the UK his includes the structure o current

subsidies as well as the specification o the contracts that are proposed under the EMR A

related question concerns the expected lie o new wind arms

b) How might the average perormance o wind generation develop over the next decade

his is absolutely critical to any assessment o the amount o wind capacity that will be

required to meet the UK governmentrsquos targets or the share o electricity ndash and more

generally energy ndash rom renewable sources he government has assumed that the

average load actors or both onshore and offshore wind arms will either remain stable

or increase in uture he prospects will be very different i these assumptions are not

borne out

20 he first question can be addressed by considering the discounted sum o cumulative net

output rom a new wind arm on the assumption that its perormance matches the averages

identified in the analysis his takes account o both perormance degradation over time plus

the annual cost o maintaining the plant which may be expressed as a deduction rom gross

production so as to calculate the contribution o expected output at age s to recovering the

original capital cost o building the plant hese figures have to be discounted back to the date

o commissioning in order to assess the net present value o the initial investment in the plant

he curves or constant and observed perormance using a discount rate o 9 (a typical cost o

capital or wind generation) are shown in Figure 5 he black line is based on the ull set o age

8 he initial load actors are calculated as the normalised load actor or UKDanish wind arms at age 0 multiplied by

where is the unit fixed effect or installation i

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1852

16 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

effects while the red dashed line is based on the smoothed quadratic perormance curve and

the blue dashed line shows what the pattern would be i operating perormance was not affected

by age he calculation is extended out to age 25 since DECCrsquos standard comparisons assume

that wind turbines have a lie o 25 years

Figure 5 Impact of performance degradation on discounted cumulative output

0

10

20

30

40

50

60

70

80

90

100

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

o

f d i s c o u n t e d c u m u l a t i v e o u t p u t

Age of plant (years)

Constant Quadratic Full age effects

Source Authorrsquos estimates

21 he key points in the figure are that 80 o the discounted cumulative output o a new wind

plant is likely to be produced in the first 10 years o its lie and 90 in the first 14 years his

is consistent with the structure o Danish subsidies or onshore wind arms which extend over

a typical period o 125 years Since sites or wind arms are scarce and involve the payment o

significant rents that may be linked to output it is very unlikely that any new installation o

wind turbines will have an expected lie o more than 15 years Instead as has happened in the

past wind operators will have a strong incentive to decommission plants afer no more than 15

years and replace the turbines with newer equipment

22 As a consequence any economic assessment o wind generation should not be based on an

expected lie which is longer than 15 years In recent work reported in evidence to the House

o Commons Select Committee on Energy and Climate Change I assumed that wind plants

would have a residual value equal to 20 o their initial cost in real terms at the end o 15 years

he analysis in this paper suggests that this is too avourable an assumption Given the costs o

decommissioning old turbines the residual value is likely to be well below 10 o their initial

cost and the decision point may be at 10 rather than 15 years

23 he corollary o this observation is that it makes little sense to offer long term contracts or 20

years or more that guarantee prices or eed-in tariffs (Fis) to wind operators A contract length

o 10 to 12 years would be sufficient to remove most o the market risk associated with invest-

ment in wind generation In this respect the subsidy arrangements implemented in Denmark

are better designed

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 17

24 At the same time the offer o subsidies andor guaranteed prices may have serious adverse

consequences or the efficiency o wind generation Returning to the unit fixed effects measur-

ing the perormance o wind plants commissioned afer 2005 only 28 out o 159 units have an

operating perormance that exceeds the average or the period hose 28 units account or 360

MW o capacity whereas the remaining 131 units account or 2810 MW o capacity Not onlyare recent plants less efficient than the average or the whole period but the plants which are

below-average in efficiency are typically larger than the more efficient ones and account or a

disproportionate share o the recent additions to generating capacity

25 In mid-2012 there was approximately 4500 MW o onshore wind capacity operating in the

UK By 2020 it is expected that there will be at least 10000 MW (based on proposed wind

arms with planning consent) and perhaps as much 15000 MW o wind capacity in operation

he government is relying upon a substantial contribution rom onshore wind arms towards

meeting its targets or renewable energy in 2020 o provide context the governmentrsquos targets

or renewable energy in 2020 speciy a total o 234 Wh o energy rom all renewable sources

o which 90ndash140 Wh is expected to come rom wind and biomass electricity generation9

he projections or biomass electricity are extremely vague and have to be reconciled with

an assumed increase o 3ndash4 times in the amount o biomass used separately or heat So in

concrete terms it is reasonable to assume that wind generation will account or at least 90 Wh

with 24ndash32 Wh rom onshore generation and the remainder rom offshore

26 he average load actor or onshore wind over the 10 year period rom April 2002 to March

2012 was 256 which is influenced by the large amounts o new capacity added rom 2005

onwards I this load actor were achieved in 2020 the total amount o onshore wind capacityrequired to meet the onshore generation target would be 107 to 143 GW he average load

actor or offshore wind installations (on an unchanged configuration basis) or the five year

period 2007ndash11 was 32010 I this load actor were achieved in 2020 the total amount o

offshore wind capacity required to meet the remainder o the 90 Wh target would be 206

to 235 GW though these estimates would all to 180 to 205 GW i the average load actor

matched the historic average value o 367 or Danish offshore wind installations

27 he question that has to be addressed is whether these load actors are consistent with (a) the

investment programmes that would be required to achieve the total amounts o generation

capacity implied by the governmentrsquos targets and (b) the age profiles o the perormance o

wind arms discussed in this paper he short answer is that this is simply not possible i the

initial load actor or new wind arms continues to decline at the rate observed the past decade

So even as a starting point it is necessary to assume that this underlying deterioration in wind

arm perormance ceases his would be a major change in itsel and it is not obvious what

actors might lead to such a result

28 he results illustrated below are based on a vintage model o perormance that takes account

o the actual expansion o capacity up to the end o 2011 plus whatever investment in new or

replacement turbines is required to achieve the target levels o wind capacity in 2020 o high-light the ull range o potential outcomes these targets are either 10 GW or 15 GW o onshore

9 Department o Energy and Climate Change ndash UK Renewable Energy Road Map July 2011 Figure 2

10 he equivalent average load actor or onshore wind arms was 264

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2052

18 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

wind and either 18 or 24 GW o offshore wind he levels o investment required to meet these

targets depend upon the expected lie o wind turbines For onshore turbines an expected lie

o 10 years is consistent with re-powering decisions over the last decade while an expected lie

o 15 years seems to be the maximum consistent with the age-perormance profile For offshore

turbines the analysis uses expected lives o 15 and 20 years

Figure 6A Projected load factors for UK onshore wind (total capacity of 10 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

Figure 6B Projected load factors for UK onshore wind (total capacity of 15 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

29 Figure 6A shows projected values o the average normalised load actors or UK onshore wind

arms in 2015 and 2020 using a target capacity in 2020 equal to 10 GW plus the age-peror-

mance profiles or the UK and Denmark (DK) Since vintage effects are important in this

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 19

calculation the profile o new or replacement investment affects the results In this case the

level o gross investment is assumed to be constant rom 2013 onwards at either 930 or 1220

MW per year ndash the range is defined by the assumed lie o turbines Achieving a target o 15

GW in 2020 (Figure 6B) would require a level o investment rom 2013 onwards o either 1540

or 1930 MW per year Such figures are 3ndash4 times the average level o investment over theperiod 2005ndash11 so the lower target may provide a better indication o what will happen

30 All o the estimates are calibrated to match the observed average load actor in 2007ndash11 so that

they should be interpreted as relative rather than absolute values Using the age-perormance

profile or UK onshore wind installations the average normalised load actor is likely to all by

about 19 rom 2007ndash11 to 2015 and by 23ndash28 to 2020 I the age-perormance curves or

Danish onshore installations were to apply the reduction in the average normalised load actor

would be considerably less but there would still be a all o up to 5 by 2020

31 In the medium term the all is greater i turbines operate or 15 years because the average age

o all turbines will be higher and thus the degradation in perormance due to age will have a

larger effect By 2020 this effect reduces the average normalised load actor rom 203 or a 10

year lie to 195 or a 15 year lie here is a complex economic and commercial trade-off here

o meet a target o 10 GW in 2020 it is necessary to increase the level o investment in onshore

capacity by 30 rom 2013 Almost all o this will take the orm o re-powering existing wind

arms11 he costs are airly high because this will usually involve new development consents

decommissioning the original turbines and replacing them with a smaller number o larger and

more powerul turbines

32 Estimates o re-powering costs are generally treated as being commercially sensitive but broadproject costs cited in specialist magazines suggest that the unit costs are likely to be at least

pound1000 per kW or more than pound300 million per year in aggregate On the other hand the increase

in the amount o energy generated would be about 12 Wh or a 10 year turbine lie rather

than a 15 year turbine lie though this does not allow or the loss o production during re-pow-

ering Under the best circumstances the payback period may be quite short but the best option

rom a broader economic and social perspective is not obvious

33 Even under the best scenario 10 GW o onshore wind capacity will only generate 178 Wh o

electricity in 2020 well below the lower end o the target range or onshore wind generation

specified in DECCrsquos road map In act allowing or interactions between the age-perormance

profile and investment the amount o onshore wind capacity required in 2020 to generate 24

Wh would be a minimum o 134 GW At the top end o DECCrsquos target range the capacity

required in 2020 to generate 32 Wh would be a minimum o 178 GW Since these estimates

are based on a 10 year turbine lie the level o investment required rom 2013 onwards would

be 1650 MW per year or the bottom o the range and 2350 MW or the top end o the range

In the last 8 years total investment in onshore wind has only ever exceeded 600 MW in one

11 Re-powering is usually defined as the ull replacement o existing turbines whereas retrofitting covers the replace-

ment o various mechanical or electrical components In the data analysis re-powering involves the creation o a

new unit whereas retrofitting affects the perormance o a continuing unit Hence the potential benefits o retrofit-

ting are taken into account in constructing the age-perormance curves

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2252

20 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

year (883 MW in 2008) so it seems rather unlikely that even the lower end o DECCrsquos range or

onshore wind generation will actually be realised

34 O course this assessment would be different i UK onshore wind operators were able to match

the Danish age-perormance curve here are many actors which differentiate the UK and

Danish wind industries he average scale o wind arms in the UK is larger they are moreremote and there appears to be significantly greater variation between operating and potential

new sites in achieved perormance It would thereore require a very optimistic observer to

conclude that the difference in age-perormance curves between the two countries will disap-

pear in the short or medium term

Figure 7 Projected load factors for UK offshore wind in 2020

20

22

24

26

28

30

32

34

18 GW ndash life 15 yrs 18 GW ndash life 20 yrs 24 GW ndash life 15 yrs 24 GW ndash life 20 yrs

L o a d f a c t o r ( )

2007ndash11 2015 2020

Note Based on Danish age-perormance curves Source Authorrsquos estimates

35 Figure 7 shows the results rom carrying out a similar exercise or offshore wind arms in the

UK he only evidence on the age-perormance curves or offshore installations is based upon

the Danish experience so these have been used to construct the projections Further the profile

o investment assumes a progressive increase in additions to capacity rom 2013 to 2016 and

then a stable level o investment rom 2017 to 2020 he DECC road map involves a very large

commitment to offshore wind investment ndash in the range 25-35 GW o new capacity annually

rom 2016 onwards As a consequence the average age o offshore capacity in 2020 will be quite

low and the ull effects o the age-perormance curves will only be elt during the decade ollow-

ing 2020 Even so the average normalised load actor in 2020 will be about 15 lower than or

2007ndash11 while by 2025 it will be about 30 below the base level

36 On these projections 232 GW o offshore capacity is the minimum necessary to generate 58

Wh o electricity rom offshore wind in 2020 his requirement will increase to 298 GW by

2025 simply to offset the effect o the ageing o offshore generation capacity on total output

37 Combining the estimates or onshore and offshore wind generation the total amount o wind

capacity required in 2020 to produce 90 Wh o electricity rom wind will be in the range 39

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 21

to 41 GW once allowance is made or the observed age-perormance curves or both onshore

and offshore installations hese estimates are roughly a third higher than official predictions o

28ndash31 GW o wind capacity in 2020

38 urning the figures round gives a slightly different picture o the reality o the governmentrsquos

strategy on renewable electricity generation A total o 30 GW o wind capacity in 2020 (12GW onshore 18 GW offshore) will generate about 64 Wh o electricity in a normal year

Since electricity generation rom wind and biomass are expected to account or 50 o renew-

able energy in 2020 the consequence is that biomass is expected to account or 53 Wh In

2011 landfill gas and sewage sludge digestion accounted or 44 o electricity rom bioenergy

(DUKES ables 64) here are limited opportunities to expand production rom these sources

so a continuation o historic rates o growth would mean that renewable generation rom other

sources o bioenergy would have to increase rom 72 Wh in 2011 to 46 Wh in 2020 here

are severe constraints on increasing the amounts o electricity generated rom animal biomass

biodegradable municipal solid waste and anaerobic digestion so it would be optimistic to

assume that these will contribute more than 6-8 Wh in 2020 (up rom 26 Wh in 2011)

39 In practice i the projections or renewable energy are to be believed they rest upon an assump-

tion that there will be a large increase in the amount o electricity that is generated rom plant

biomass ndash either on its own or through co-firing with coal his increase will involve increas-

ing the amount o electricity generated in these ways rom 46 Wh in 2011 to 38-40 Wh in

2020 Since the amounts o straw and UK-grown timber available or this purpose are strictly

limited this will come down to the amount o wood chips that can be imported and the cost o

doing this40 he historic average load actor or biomass plants is just over 60 so about 7500 MW o

biomass capacity will be required to supply 40 Wh o electricity he weight o wood chips per

MWh o electricity depends upon moisture content storage boiler design and other actors but

an indicative range is 085 to 115 tonnes per MWh o wood chips with a moisture content o

30 According to DECC and Forestry Commission estimates the UK imported 265 million

tonnes o steam coal and 26 million tonnes o other wood products (including wood chips)

in 2011 It seems that the net effect o the governmentrsquos strategy will be to replace 15 million

tonnes o coal imported in 2011 by 32 to 44 million tonnes o imported wood chips his is a

somewhat strange way o increasing energy security

Conclusion

41 Wind power is a highly capital-intensive technology or generating electricity Its merits rely

entirely upon a substitution o capital or uel inputs he same is true or hydro or tidal or wave

power In comparison with hydro power wind is a low quality resource because o its variability

and because it cannot be stored So the economic case or wind power must rest on obtaining

the most out o the wind that is available

42 While the decline in the achieved perormance o onshore wind turbines in Denmark is

much less than that or the UK or offshore nonetheless the decline in expected output under

standardised wind conditions over 10 years is 10 unweighted and 13 capacity weighted

hese declines accelerate afer age 10 so that the reductions in perormance are 17 and 20

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2452

22 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

respectively afer 15 years For UK onshore wind arms the reduction in perormance due to

age is much worse at 27 unweighted and 69 capacity weighted by age 10

43 Evidence on the perormance o Danish offshore installations is both restricted and so poor

that there may be concern that the results are affected by a small number o outliers Still the

sample contains a reasonable number o sites with at least 5 years o operating experience andthe decline in perormance by age 5 is 38 unweighted and 26 capacity weighted

44 In addition to these results there is strong evidence that the average normalised load actor or

new onshore wind installations in the UK has allen significantly over the period rom 2000 to

2011 his is consistent with a pattern in which the most avourable sites are developed first

Equally it could mean that wind developers have been unable to keep up with the rate o new

investment while maintaining the quality o development and operations For example the site

design or selection o turbine characteristics may make less effective use o the available wind

resources or the sites available than was the case in the past

45 Whatever the reasons the deterioration in initial perormance means that the expected returns

rom the expansion in wind capacity both or investors and in terms o the reduction in CO983090

emissions have been alling without a concomitant decrease in the private and social costs

that are borne by customers and the general public Clearly this is unsatisactory at best and it

suggests that the benefits claimed or current policies cannot be taken at ace value

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 23

Appendix Data and Methods

A Data for the United Kingdom

he raw data used or this study was extracted by the Renewable Energy Foundation rom the

Renewables and CHP Register database compiled by Ogem12 his inormation is used in the

administration o the market in Renewable Obligation Certificates (ROCs) he Renewables

Obligation is the primary mechanism by which large scale generators o renewable electric-

ity receive subsidies so the operators o wind arms have a very strong incentive to submit

complete and up-to-date inormation to the database he data extracted covered all onshore

wind generators with at least 05 MW o generating capacity which are eligible or ROCs or the

10 year period April 2002 to March 2012 he key variables are the monthly output o electric-

ity (used or the purpose o allocating ROCs) and nominal generating capacity here are 282

reporting units in the final dataset A reporting unit may correspond to an entire wind arm orto separate phases o development Complications arise when an existing wind arm is re-pow-

ered ie when old wind turbines are replaced by newer and usually more powerul turbines

he database reports this as a change in the generating capacity o the recording unit but in

this analysis repowering is treated as the termination o the old recording unit and the creation

o a new recording unit

A considerable amount o data checking and cleaning was required beore the data could

be analysed he first step was to add the age o each reporting unit his was calculated by

reerence to the month in which the turbines were commissioned For wind arms in operation

in or beore June 2002 the commissioning date was obtained rom various online sources ndash

notably the thewindpowernet database o UK wind arms13 which was cross-checked against

the inormation provided on the websites o wind operators and other sources For recording

units whose first data relates to dates afer June 2002 the commissioning date was initially

assumed to be the month or which data is first reported Cross-checks were carried out or

all recording units which commenced reporting in the period July 2002 to December 2003 In

more than 80 o cases the externally-reported commissioning date is within one month o the

first month or which data is reported he exceptions ndash Bu Farm and Mablethorpe ndash appear to

be due to delays by small operators in reporting data to Ogem

In a number o cases it is clear rom the data that plants were not in ull operation at the

time when data was first reported to the database here are two indicators o partial or delayed

commissioning

bull For an initial series o months the calculated load actor is very low ndash usually below 5 ndash

and then increases abruptly In such cases the data or the early months is dropped and the

commissioning date is treated as the month immediately preceding the first month with a

lsquonormalrsquo load actor14

12 httpswwwrenewablesandchpogemgovuk

13 See httpwwwthewindpowernetwindarms_list_enphp

14 he month in which a plant is reported as having been commissioned is not included in the analysis

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2652

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 25

the decline in the plantrsquos load actor as it ages As an example consider a plant whose reported

capacity was increased rom 99 MW to 10 MW at age 3 I this was simply a change in report-

ing practice this would reduce the estimated load actor rom age 3 onwards By increasing

the initial capacity rom 99 MW to 10 MW ndash or by reducing the capacity rom 10 MW to 99

MW rom age 3 onwards ndash this bias is removed It is not important which adjustment is madebecause the difference between the alternative assumptions is captured by the unit fixed effect

he numbers o separate sites and o observations per site mean that it is not possible to

provide a simple visual representation o the data Figure 8 provides a summary by illustrating

the distributions o load actors by the age o the wind arm pooled across all years he ranges

o the load actors or each age are wide while the medians show no clear trend by age hus a

naiumlve analysis might conclude that there is nothing to investigate However as will be explained

below the distributions mask crucial differences between the perormance o wind arms as

they age because they do not control or the differences in location and wind availability over

time

B Data for Denmark

he data or Danish wind arms used in this study comes rom a database compiled by the

Danish Energy Agency covering the characteristics and perormance o all wind turbines rom

2002 up to the end o August 201216 he basic recording unit or the register is the individual

wind turbine but where a number o turbines are connected to a single meter an identical

average output per turbine is recorded or all turbines in the group he presence o what are in

effect repeated observations increases the noise in the data In addition the model that is used

or the analysis assumes that site operator and turbine characteristics are common or all o theunits that make up a single wind arm o address this point separate wind arms have been

identified by grouping all turbines which have identical values or turbine manuacturer and

type local authority date o commissioning and date o decommissioning (where applicable)

he register contains inormation on 7569 turbines commissioned rom 1977 onwards his

includes a large number o small turbines which would not all within the scope o the ROC

scheme in the UK Hence or comparability with the UK data all turbines commissioned

prior to 1992 were dropped as well as all wind arms with a generating capacity o less than 1

MW he resulting sample covered 823 onshore wind arms and 30 offshore wind arms with

monthly load actors or various periods rom January 2002 to August 2012 Given the cut-off

or wind arms commissioned prior to 1992 the maximum age or an onshore wind arm was 20

years and the maximum age or an offshore wind arm was 17 years

he total capacity o the onshore wind arms recorded in the Danish database was 2757 MW

he average size o an onshore wind arm in Denmark is small ndash only 33 MW as compared

with 161 MW or the UK ndash because 84 o the sample consists o installations with between

1 and 4 turbines he largest onshore installation has 39 turbines with a capacity o 234 MW

while the onshore installation with the greatest capacity has 19 turbines with a capacity o 584

MW Overall the sample o Danish onshore wind arms is older and smaller than the sample

o UK onshore wind arms reflecting the much longer history o wind generation in Denmark

16 httpwwwensdkEN-USINFOFACSANDFIGURESENERGY_SAISICS_AND_INDICAORS

OVERVIEWOFHEENERGYSECORREGISEROFWINDURBINESSiderForsideaspx

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2852

26 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

he larger scale o UK wind arms may account or some o the differences in perormance

curves or the two countries

he sample o offshore wind arms in Denmark is quite small with 30 installations in

total Still it is useul to analyse the perormance o these installations since there is very little

comparable data or the UK he sample covers 394 turbines with an average o just over 13turbines per installation and an average capacity o 288 MW he largest installation which

was commissioned in 2003 has 72 turbines with a total capacity o 1656 MW Again offshore

installations are smaller and older in Denmark than is the current pattern in the UK he first

two wind arms in the sample were commissioned in the years 1995 and 2000 but the sample

increases significantly rom 2002 onwards o avoid reliance upon a sample o 1 or 2 wind

arms to determine specific coefficients or ages gt 10 years all observations with an age greater

than 10 were omitted rom the analysis17

A somewhat different problem with the offshore wind arm data arose because the coeffi-

cients estimated using capacity weights (see below) were heavily influenced by the largest windarm his has had an extremely erratic profile o perormance over time with a very low average

load actor he treatment o this wind arm is discussed in more detail in Section F below

C Specication and estimation methods

As Figure 8 illustrates it is necessary to go beyond simple summary statistics in order to assess

whether there is any systematic relationship between age and the average perormance o wind

arms he perormance o wind arms varies over both time and space some months and years

have more or less wind than the long term average while specific characteristics o wind arms

ndash including location type o turbine and operating regime ndash will influence the perormance oeach plant under identical wind conditions

he combination o (a) site-specific characteristics which are constant over time and (b)

period-specific characteristics which are constant across wind arms can be represented by

what statisticians call an error components model with fixed effects or each wind arm and

each time period he dependent variable in the model consists o a sequence o load actors ndash

denoted by or unit i in month t ndash or each wind arm (the panel unit) over time he load

actors are calculated by dividing the total output in MWh by (24 x number o days in month x

reported generating capacity in MW) and are multiplied by 100 to convert to percentages

he additive version o the error components model may be written as

(1)

in which denotes the age o plant i in period t and is some unction that is either a

representation o or an approximation to the normalised age-perormance relationship he

multiplicative version is similar except that is replaced by where ln() denotes the

natural logarithm he terms are wind arm or unit fixed effects which capture actors such

as location type o turbine operating regimes etc he terms are period fixed effects which

capture the influence o wind conditions over the UK or Denmark as a whole as well as seasonal

maintenance and demand Without loss o generality it may be assumed that the fixed effects are

17 For the avoidance o doubt this means that no attempt was made to estimate the effect o age on the perormance o

offshore wind arms beyond age 10 while the data or the oldest wind arms was included in the analysis up to and

including age 10

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 27

normalised so that ie all o the error components have a zero mean while the

mean o the random error is zero by definition he terms are random errors which capture

random variations in wind conditions that are not specific to the site turbine breakdowns and

other actors that are uncorrelated with either the site or the time period It is assumed that the

mean o the random error is zero while additional assumptions depend upon the method o

estimation that is adopted

he model specified in equation (1) can be estimated in a variety o ways18 A crucial issue is

whether the are or may be significantly correlated with the age terms or any other independent

regressors I it is assumed that there is no such correlation the model is usually reerred to as

a random effects model In that case the equation can be estimated by a variant o least squares

(LS) generalised least squares (GLS) generalised methods o moments (GMM) or maximum

likelihood (ML) ndash all o which should be yield parameter estimates that will converge in prob-

ability to the true parameter values as the sample size increases An alternative but less restric-

tive specification is known as the fixed effects model in which no assumption is made about

the correlation between the unit fixed effects and other regressors he limitation o the fixed

effects model is that it cannot be used to estimate parameters or regressors that vary across

panel units but are constant over time or a panel unit (time-invariant regressors) ndash eg the

year in which a wind arm was commissioned its location or its rated generating capacity he

influence o such variables is captured by the unit fixed effects

In order to avoid the risk o obtaining biased estimates o the age effects the results reported

in this note are based upon estimates using the fixed effects model he influence o time-invar-

iant regressors is examined in a second stage by regressing the estimates o the unit fixed effectson these variables he simplest method o estimating the fixed effects model is known as the

lsquowithinrsquo estimator in which the mean value or each unit is subtracted rom all observations or

that unit For example i is a general polynomial o order M in A equation (1) becomes

(2)

so that the model can be estimated by least square with the inclusion o dummy variables or

each period t he variance-covariance matrix o the coefficients have been estimated by using

(a) a robust sandwich estimator adjusted or clustering by panel unit and (b) a bootstrap esti-

mator with 400 repetitions he estimates o the standard errors will be consistent even i the

random error component is serially correlated within panel units andor has different variances

across panel units19 Since the number o panel units and the average number o time periods

per panel units are both large (other than or Danish offshore wind arms) the least square esti-

mates o the parameters will be unbiased under these assumptions

he bootstrap method is expensive to carry out but it provides a useul cross-check on

the standard errors generated by other methods Unortunately the bootstrap method o

18 See Chapter 21 o AC Cameron amp PK rivedi ndash Microeconometrics Methods and Applications (Cambridge

Cambridge University Press 2005) or a standard textbook treatment o estimation methods or panel data models

19 he estimation was carried out using the xtreg procedure in Stata Version 12 he methods o constructing the

sandwich and bootstrap estimates o the variance-covariance matrix are described in Chapters 8 amp 13 o AC

Cameron amp PK rivedi ndash Microeconometrics using Stata (Revised Edition College Station exas Stata Press 2010)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 15: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 13

13 o give a sense o the numbers unit fixed effects o (a) 02 or (b) -02 mean that the base load

actor given the age o the plant is multiplied by (a) exp(02) = 122 or (b) exp(-02) = 082 Since

the base load actor or a plant o age 0 is 231 these fixed effects translate to load actors in

the first year o operation o (a) 282 and (b) 189 For a wind arm with 50 MW o gener-

ating capacity the difference amounts to an additional 40700 MWh o electricity output peryear Using the standard figures cited by RenewableUK this would be sufficient to supply 12300

homes or a year his range (040) is slightly greater than the interquartile range o unit fixed

effects or UK onshore wind arms (035) Again or comparison operators o thermal power

plants would regard any difference o even 10 up or down in efficiency relative to what is

expected as a striking and perhaps worrying indication o either good or bad perormance

Figure 3 Range of unit fixed effects by category

- 1

- 5

0

5

1

A d j f i x e d e f f e c t ndash q u a d r a t i c

UK onshore DK onshore DK offshore

Source Authorrsquos estimates

14 A notable eature o the distributions o the unit fixed effects that are summarised in the box

plots shown in Figure 3 is the range o variation between the best and worst perormers espe-

cially or UK onshore and Danish offshore wind arms Box plots are constructed so that the

top and bottom o the box correspond to the upper and lower quartiles while the bars at topand bottom correspond to the maximum and minimum values excluding outliers7 A plant at

the upper quartile o the distribution o UK onshore wind arms will generate an annual output

that is about 40 higher than a plant at the lower quartile he equivalent figure or Danish

onshore wind plants is an increase o 30 while or Danish offshore wind arms the difference

between the upper and lower quartiles is equivalent to 25 times the output o a lower quartile

plant By any standard such differences are important

7 he upper quartile is the point in a distribution such that the 25 o values exceed it while the lower quartile is the

point in a distribution such that 75 o values exceed it he line in the middle o the box marks the median the

point exceeded by 50 o values Outliers are defined as values that lie outside the range o ukey adjacent values

which are defined as the upperlower quartiles +- 15 x interquartile range

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1652

14 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

15 Outside the range between the upper and lower quartiles the top and bottom segments o

the distributions extend urther up and down or UK onshore wind arms than or those in

Denmark he rather different pattern or Danish offshore wind installations is partly a conse-

quence o the relatively small size o the sample but the evidence points to considerably greater

variability in the perormance o offshore wind installations than or onshore installations16 It is somewhat surprising to observe the magnitude o the differences in the perormance in

onshore wind arms in the UK Given the nature o the subsidy regimes and the high capital

cost o developing new installations it might be expected that operators have a strong incen-

tive to identiy the best locations and then choose equipment that will deliver the maximum

amount o electricity output at a high level o reliability For UK onshore operators it seems that

good or bad perormance is somewhat o a lottery However i the subsidies provided by ROCs

are sufficient to underwrite investment in inefficient plants ndash as appears to be the case ndash then

those subsidies are extremely generous or plants that operate close to the efficient rontier

As location is likely to be the main actor that determines the perormance o a specific plant

relative to all other plants the inerence must be that many wind plants have been developed on

sites with poor wind characteristics

Figure 4 Initial load factors by year of commissioning for onshore wind

10

15

20

25

30

35

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

I n i t i a l l o a d f a c t o r ( )

Year commissioned

UK onshore DK onshore Linear (UK onshore) Linear (DK onshore)

Source Authorrsquos estimates

17 here is another disturbing characteristic o UK wind arms which is revealed by examination

o the unit fixed effects and illustrated in Figure 4 It might be expected that the average peror-

mance o recently installed units would be higher than that o older units his should reflect

improvements in turbine design and reliability as well as in the identification o good sites or

wind generation However that is not the pattern or onshore wind arms in the UK he figure

shows the average values o the initial load actors calculated using equally weighted data by

year o commissioning or onshore wind arms in the UK and Denmark together with fitted

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 15

trend lines8 here has been a clear decline in the initial load actors in the UK which has not

occurred in Denmark his pattern is confirmed by more detailed statistical results shown in

the Appendix which take account o location and size o installation

18 he decline in initial perormance is particularly obvious afer 2005 when the rate o commis-

sioning new capacity increased sharply rom less than 500 MW in the 5 years 2000ndash04 to morethan 500 MW per year rom 2005 onwards In contrast the peak years or commissioning new

onshore wind arms in Denmark were 2000ndash02 when 920 MW was installed he rate o new

building (or replacement) picked up again in 2008 but it is still much lower than at the turn o

century Not only is the perormance o the onshore wind plants in the UK commissioned in

recent years significantly worse than that or wind arms commissioned beore 2005 but it is

possible that this is a direct consequence o an overly rapid expansion o capacity In essence

the evidence suggests that the industry does not have the capability to identiy develop and

operate new onshore wind arms at the rate envisaged by UK government targets while main-

taining a satisactory average level o perormance

Implications for future policy and performance

19 he pattern o perormance degradation with age and over time identified in this analysis has

implications or the assessment o new investment in wind arms and or the design o the

proposed Electricity Market Reorm (EMR) here are two related issues that are affected by

the results

a) What are the implications or the design o policies intended to promote the adoption

and expansion o wind generation in the UK his includes the structure o current

subsidies as well as the specification o the contracts that are proposed under the EMR A

related question concerns the expected lie o new wind arms

b) How might the average perormance o wind generation develop over the next decade

his is absolutely critical to any assessment o the amount o wind capacity that will be

required to meet the UK governmentrsquos targets or the share o electricity ndash and more

generally energy ndash rom renewable sources he government has assumed that the

average load actors or both onshore and offshore wind arms will either remain stable

or increase in uture he prospects will be very different i these assumptions are not

borne out

20 he first question can be addressed by considering the discounted sum o cumulative net

output rom a new wind arm on the assumption that its perormance matches the averages

identified in the analysis his takes account o both perormance degradation over time plus

the annual cost o maintaining the plant which may be expressed as a deduction rom gross

production so as to calculate the contribution o expected output at age s to recovering the

original capital cost o building the plant hese figures have to be discounted back to the date

o commissioning in order to assess the net present value o the initial investment in the plant

he curves or constant and observed perormance using a discount rate o 9 (a typical cost o

capital or wind generation) are shown in Figure 5 he black line is based on the ull set o age

8 he initial load actors are calculated as the normalised load actor or UKDanish wind arms at age 0 multiplied by

where is the unit fixed effect or installation i

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1852

16 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

effects while the red dashed line is based on the smoothed quadratic perormance curve and

the blue dashed line shows what the pattern would be i operating perormance was not affected

by age he calculation is extended out to age 25 since DECCrsquos standard comparisons assume

that wind turbines have a lie o 25 years

Figure 5 Impact of performance degradation on discounted cumulative output

0

10

20

30

40

50

60

70

80

90

100

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

o

f d i s c o u n t e d c u m u l a t i v e o u t p u t

Age of plant (years)

Constant Quadratic Full age effects

Source Authorrsquos estimates

21 he key points in the figure are that 80 o the discounted cumulative output o a new wind

plant is likely to be produced in the first 10 years o its lie and 90 in the first 14 years his

is consistent with the structure o Danish subsidies or onshore wind arms which extend over

a typical period o 125 years Since sites or wind arms are scarce and involve the payment o

significant rents that may be linked to output it is very unlikely that any new installation o

wind turbines will have an expected lie o more than 15 years Instead as has happened in the

past wind operators will have a strong incentive to decommission plants afer no more than 15

years and replace the turbines with newer equipment

22 As a consequence any economic assessment o wind generation should not be based on an

expected lie which is longer than 15 years In recent work reported in evidence to the House

o Commons Select Committee on Energy and Climate Change I assumed that wind plants

would have a residual value equal to 20 o their initial cost in real terms at the end o 15 years

he analysis in this paper suggests that this is too avourable an assumption Given the costs o

decommissioning old turbines the residual value is likely to be well below 10 o their initial

cost and the decision point may be at 10 rather than 15 years

23 he corollary o this observation is that it makes little sense to offer long term contracts or 20

years or more that guarantee prices or eed-in tariffs (Fis) to wind operators A contract length

o 10 to 12 years would be sufficient to remove most o the market risk associated with invest-

ment in wind generation In this respect the subsidy arrangements implemented in Denmark

are better designed

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 17

24 At the same time the offer o subsidies andor guaranteed prices may have serious adverse

consequences or the efficiency o wind generation Returning to the unit fixed effects measur-

ing the perormance o wind plants commissioned afer 2005 only 28 out o 159 units have an

operating perormance that exceeds the average or the period hose 28 units account or 360

MW o capacity whereas the remaining 131 units account or 2810 MW o capacity Not onlyare recent plants less efficient than the average or the whole period but the plants which are

below-average in efficiency are typically larger than the more efficient ones and account or a

disproportionate share o the recent additions to generating capacity

25 In mid-2012 there was approximately 4500 MW o onshore wind capacity operating in the

UK By 2020 it is expected that there will be at least 10000 MW (based on proposed wind

arms with planning consent) and perhaps as much 15000 MW o wind capacity in operation

he government is relying upon a substantial contribution rom onshore wind arms towards

meeting its targets or renewable energy in 2020 o provide context the governmentrsquos targets

or renewable energy in 2020 speciy a total o 234 Wh o energy rom all renewable sources

o which 90ndash140 Wh is expected to come rom wind and biomass electricity generation9

he projections or biomass electricity are extremely vague and have to be reconciled with

an assumed increase o 3ndash4 times in the amount o biomass used separately or heat So in

concrete terms it is reasonable to assume that wind generation will account or at least 90 Wh

with 24ndash32 Wh rom onshore generation and the remainder rom offshore

26 he average load actor or onshore wind over the 10 year period rom April 2002 to March

2012 was 256 which is influenced by the large amounts o new capacity added rom 2005

onwards I this load actor were achieved in 2020 the total amount o onshore wind capacityrequired to meet the onshore generation target would be 107 to 143 GW he average load

actor or offshore wind installations (on an unchanged configuration basis) or the five year

period 2007ndash11 was 32010 I this load actor were achieved in 2020 the total amount o

offshore wind capacity required to meet the remainder o the 90 Wh target would be 206

to 235 GW though these estimates would all to 180 to 205 GW i the average load actor

matched the historic average value o 367 or Danish offshore wind installations

27 he question that has to be addressed is whether these load actors are consistent with (a) the

investment programmes that would be required to achieve the total amounts o generation

capacity implied by the governmentrsquos targets and (b) the age profiles o the perormance o

wind arms discussed in this paper he short answer is that this is simply not possible i the

initial load actor or new wind arms continues to decline at the rate observed the past decade

So even as a starting point it is necessary to assume that this underlying deterioration in wind

arm perormance ceases his would be a major change in itsel and it is not obvious what

actors might lead to such a result

28 he results illustrated below are based on a vintage model o perormance that takes account

o the actual expansion o capacity up to the end o 2011 plus whatever investment in new or

replacement turbines is required to achieve the target levels o wind capacity in 2020 o high-light the ull range o potential outcomes these targets are either 10 GW or 15 GW o onshore

9 Department o Energy and Climate Change ndash UK Renewable Energy Road Map July 2011 Figure 2

10 he equivalent average load actor or onshore wind arms was 264

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2052

18 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

wind and either 18 or 24 GW o offshore wind he levels o investment required to meet these

targets depend upon the expected lie o wind turbines For onshore turbines an expected lie

o 10 years is consistent with re-powering decisions over the last decade while an expected lie

o 15 years seems to be the maximum consistent with the age-perormance profile For offshore

turbines the analysis uses expected lives o 15 and 20 years

Figure 6A Projected load factors for UK onshore wind (total capacity of 10 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

Figure 6B Projected load factors for UK onshore wind (total capacity of 15 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

29 Figure 6A shows projected values o the average normalised load actors or UK onshore wind

arms in 2015 and 2020 using a target capacity in 2020 equal to 10 GW plus the age-peror-

mance profiles or the UK and Denmark (DK) Since vintage effects are important in this

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 19

calculation the profile o new or replacement investment affects the results In this case the

level o gross investment is assumed to be constant rom 2013 onwards at either 930 or 1220

MW per year ndash the range is defined by the assumed lie o turbines Achieving a target o 15

GW in 2020 (Figure 6B) would require a level o investment rom 2013 onwards o either 1540

or 1930 MW per year Such figures are 3ndash4 times the average level o investment over theperiod 2005ndash11 so the lower target may provide a better indication o what will happen

30 All o the estimates are calibrated to match the observed average load actor in 2007ndash11 so that

they should be interpreted as relative rather than absolute values Using the age-perormance

profile or UK onshore wind installations the average normalised load actor is likely to all by

about 19 rom 2007ndash11 to 2015 and by 23ndash28 to 2020 I the age-perormance curves or

Danish onshore installations were to apply the reduction in the average normalised load actor

would be considerably less but there would still be a all o up to 5 by 2020

31 In the medium term the all is greater i turbines operate or 15 years because the average age

o all turbines will be higher and thus the degradation in perormance due to age will have a

larger effect By 2020 this effect reduces the average normalised load actor rom 203 or a 10

year lie to 195 or a 15 year lie here is a complex economic and commercial trade-off here

o meet a target o 10 GW in 2020 it is necessary to increase the level o investment in onshore

capacity by 30 rom 2013 Almost all o this will take the orm o re-powering existing wind

arms11 he costs are airly high because this will usually involve new development consents

decommissioning the original turbines and replacing them with a smaller number o larger and

more powerul turbines

32 Estimates o re-powering costs are generally treated as being commercially sensitive but broadproject costs cited in specialist magazines suggest that the unit costs are likely to be at least

pound1000 per kW or more than pound300 million per year in aggregate On the other hand the increase

in the amount o energy generated would be about 12 Wh or a 10 year turbine lie rather

than a 15 year turbine lie though this does not allow or the loss o production during re-pow-

ering Under the best circumstances the payback period may be quite short but the best option

rom a broader economic and social perspective is not obvious

33 Even under the best scenario 10 GW o onshore wind capacity will only generate 178 Wh o

electricity in 2020 well below the lower end o the target range or onshore wind generation

specified in DECCrsquos road map In act allowing or interactions between the age-perormance

profile and investment the amount o onshore wind capacity required in 2020 to generate 24

Wh would be a minimum o 134 GW At the top end o DECCrsquos target range the capacity

required in 2020 to generate 32 Wh would be a minimum o 178 GW Since these estimates

are based on a 10 year turbine lie the level o investment required rom 2013 onwards would

be 1650 MW per year or the bottom o the range and 2350 MW or the top end o the range

In the last 8 years total investment in onshore wind has only ever exceeded 600 MW in one

11 Re-powering is usually defined as the ull replacement o existing turbines whereas retrofitting covers the replace-

ment o various mechanical or electrical components In the data analysis re-powering involves the creation o a

new unit whereas retrofitting affects the perormance o a continuing unit Hence the potential benefits o retrofit-

ting are taken into account in constructing the age-perormance curves

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2252

20 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

year (883 MW in 2008) so it seems rather unlikely that even the lower end o DECCrsquos range or

onshore wind generation will actually be realised

34 O course this assessment would be different i UK onshore wind operators were able to match

the Danish age-perormance curve here are many actors which differentiate the UK and

Danish wind industries he average scale o wind arms in the UK is larger they are moreremote and there appears to be significantly greater variation between operating and potential

new sites in achieved perormance It would thereore require a very optimistic observer to

conclude that the difference in age-perormance curves between the two countries will disap-

pear in the short or medium term

Figure 7 Projected load factors for UK offshore wind in 2020

20

22

24

26

28

30

32

34

18 GW ndash life 15 yrs 18 GW ndash life 20 yrs 24 GW ndash life 15 yrs 24 GW ndash life 20 yrs

L o a d f a c t o r ( )

2007ndash11 2015 2020

Note Based on Danish age-perormance curves Source Authorrsquos estimates

35 Figure 7 shows the results rom carrying out a similar exercise or offshore wind arms in the

UK he only evidence on the age-perormance curves or offshore installations is based upon

the Danish experience so these have been used to construct the projections Further the profile

o investment assumes a progressive increase in additions to capacity rom 2013 to 2016 and

then a stable level o investment rom 2017 to 2020 he DECC road map involves a very large

commitment to offshore wind investment ndash in the range 25-35 GW o new capacity annually

rom 2016 onwards As a consequence the average age o offshore capacity in 2020 will be quite

low and the ull effects o the age-perormance curves will only be elt during the decade ollow-

ing 2020 Even so the average normalised load actor in 2020 will be about 15 lower than or

2007ndash11 while by 2025 it will be about 30 below the base level

36 On these projections 232 GW o offshore capacity is the minimum necessary to generate 58

Wh o electricity rom offshore wind in 2020 his requirement will increase to 298 GW by

2025 simply to offset the effect o the ageing o offshore generation capacity on total output

37 Combining the estimates or onshore and offshore wind generation the total amount o wind

capacity required in 2020 to produce 90 Wh o electricity rom wind will be in the range 39

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 21

to 41 GW once allowance is made or the observed age-perormance curves or both onshore

and offshore installations hese estimates are roughly a third higher than official predictions o

28ndash31 GW o wind capacity in 2020

38 urning the figures round gives a slightly different picture o the reality o the governmentrsquos

strategy on renewable electricity generation A total o 30 GW o wind capacity in 2020 (12GW onshore 18 GW offshore) will generate about 64 Wh o electricity in a normal year

Since electricity generation rom wind and biomass are expected to account or 50 o renew-

able energy in 2020 the consequence is that biomass is expected to account or 53 Wh In

2011 landfill gas and sewage sludge digestion accounted or 44 o electricity rom bioenergy

(DUKES ables 64) here are limited opportunities to expand production rom these sources

so a continuation o historic rates o growth would mean that renewable generation rom other

sources o bioenergy would have to increase rom 72 Wh in 2011 to 46 Wh in 2020 here

are severe constraints on increasing the amounts o electricity generated rom animal biomass

biodegradable municipal solid waste and anaerobic digestion so it would be optimistic to

assume that these will contribute more than 6-8 Wh in 2020 (up rom 26 Wh in 2011)

39 In practice i the projections or renewable energy are to be believed they rest upon an assump-

tion that there will be a large increase in the amount o electricity that is generated rom plant

biomass ndash either on its own or through co-firing with coal his increase will involve increas-

ing the amount o electricity generated in these ways rom 46 Wh in 2011 to 38-40 Wh in

2020 Since the amounts o straw and UK-grown timber available or this purpose are strictly

limited this will come down to the amount o wood chips that can be imported and the cost o

doing this40 he historic average load actor or biomass plants is just over 60 so about 7500 MW o

biomass capacity will be required to supply 40 Wh o electricity he weight o wood chips per

MWh o electricity depends upon moisture content storage boiler design and other actors but

an indicative range is 085 to 115 tonnes per MWh o wood chips with a moisture content o

30 According to DECC and Forestry Commission estimates the UK imported 265 million

tonnes o steam coal and 26 million tonnes o other wood products (including wood chips)

in 2011 It seems that the net effect o the governmentrsquos strategy will be to replace 15 million

tonnes o coal imported in 2011 by 32 to 44 million tonnes o imported wood chips his is a

somewhat strange way o increasing energy security

Conclusion

41 Wind power is a highly capital-intensive technology or generating electricity Its merits rely

entirely upon a substitution o capital or uel inputs he same is true or hydro or tidal or wave

power In comparison with hydro power wind is a low quality resource because o its variability

and because it cannot be stored So the economic case or wind power must rest on obtaining

the most out o the wind that is available

42 While the decline in the achieved perormance o onshore wind turbines in Denmark is

much less than that or the UK or offshore nonetheless the decline in expected output under

standardised wind conditions over 10 years is 10 unweighted and 13 capacity weighted

hese declines accelerate afer age 10 so that the reductions in perormance are 17 and 20

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2452

22 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

respectively afer 15 years For UK onshore wind arms the reduction in perormance due to

age is much worse at 27 unweighted and 69 capacity weighted by age 10

43 Evidence on the perormance o Danish offshore installations is both restricted and so poor

that there may be concern that the results are affected by a small number o outliers Still the

sample contains a reasonable number o sites with at least 5 years o operating experience andthe decline in perormance by age 5 is 38 unweighted and 26 capacity weighted

44 In addition to these results there is strong evidence that the average normalised load actor or

new onshore wind installations in the UK has allen significantly over the period rom 2000 to

2011 his is consistent with a pattern in which the most avourable sites are developed first

Equally it could mean that wind developers have been unable to keep up with the rate o new

investment while maintaining the quality o development and operations For example the site

design or selection o turbine characteristics may make less effective use o the available wind

resources or the sites available than was the case in the past

45 Whatever the reasons the deterioration in initial perormance means that the expected returns

rom the expansion in wind capacity both or investors and in terms o the reduction in CO983090

emissions have been alling without a concomitant decrease in the private and social costs

that are borne by customers and the general public Clearly this is unsatisactory at best and it

suggests that the benefits claimed or current policies cannot be taken at ace value

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 23

Appendix Data and Methods

A Data for the United Kingdom

he raw data used or this study was extracted by the Renewable Energy Foundation rom the

Renewables and CHP Register database compiled by Ogem12 his inormation is used in the

administration o the market in Renewable Obligation Certificates (ROCs) he Renewables

Obligation is the primary mechanism by which large scale generators o renewable electric-

ity receive subsidies so the operators o wind arms have a very strong incentive to submit

complete and up-to-date inormation to the database he data extracted covered all onshore

wind generators with at least 05 MW o generating capacity which are eligible or ROCs or the

10 year period April 2002 to March 2012 he key variables are the monthly output o electric-

ity (used or the purpose o allocating ROCs) and nominal generating capacity here are 282

reporting units in the final dataset A reporting unit may correspond to an entire wind arm orto separate phases o development Complications arise when an existing wind arm is re-pow-

ered ie when old wind turbines are replaced by newer and usually more powerul turbines

he database reports this as a change in the generating capacity o the recording unit but in

this analysis repowering is treated as the termination o the old recording unit and the creation

o a new recording unit

A considerable amount o data checking and cleaning was required beore the data could

be analysed he first step was to add the age o each reporting unit his was calculated by

reerence to the month in which the turbines were commissioned For wind arms in operation

in or beore June 2002 the commissioning date was obtained rom various online sources ndash

notably the thewindpowernet database o UK wind arms13 which was cross-checked against

the inormation provided on the websites o wind operators and other sources For recording

units whose first data relates to dates afer June 2002 the commissioning date was initially

assumed to be the month or which data is first reported Cross-checks were carried out or

all recording units which commenced reporting in the period July 2002 to December 2003 In

more than 80 o cases the externally-reported commissioning date is within one month o the

first month or which data is reported he exceptions ndash Bu Farm and Mablethorpe ndash appear to

be due to delays by small operators in reporting data to Ogem

In a number o cases it is clear rom the data that plants were not in ull operation at the

time when data was first reported to the database here are two indicators o partial or delayed

commissioning

bull For an initial series o months the calculated load actor is very low ndash usually below 5 ndash

and then increases abruptly In such cases the data or the early months is dropped and the

commissioning date is treated as the month immediately preceding the first month with a

lsquonormalrsquo load actor14

12 httpswwwrenewablesandchpogemgovuk

13 See httpwwwthewindpowernetwindarms_list_enphp

14 he month in which a plant is reported as having been commissioned is not included in the analysis

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2652

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 25

the decline in the plantrsquos load actor as it ages As an example consider a plant whose reported

capacity was increased rom 99 MW to 10 MW at age 3 I this was simply a change in report-

ing practice this would reduce the estimated load actor rom age 3 onwards By increasing

the initial capacity rom 99 MW to 10 MW ndash or by reducing the capacity rom 10 MW to 99

MW rom age 3 onwards ndash this bias is removed It is not important which adjustment is madebecause the difference between the alternative assumptions is captured by the unit fixed effect

he numbers o separate sites and o observations per site mean that it is not possible to

provide a simple visual representation o the data Figure 8 provides a summary by illustrating

the distributions o load actors by the age o the wind arm pooled across all years he ranges

o the load actors or each age are wide while the medians show no clear trend by age hus a

naiumlve analysis might conclude that there is nothing to investigate However as will be explained

below the distributions mask crucial differences between the perormance o wind arms as

they age because they do not control or the differences in location and wind availability over

time

B Data for Denmark

he data or Danish wind arms used in this study comes rom a database compiled by the

Danish Energy Agency covering the characteristics and perormance o all wind turbines rom

2002 up to the end o August 201216 he basic recording unit or the register is the individual

wind turbine but where a number o turbines are connected to a single meter an identical

average output per turbine is recorded or all turbines in the group he presence o what are in

effect repeated observations increases the noise in the data In addition the model that is used

or the analysis assumes that site operator and turbine characteristics are common or all o theunits that make up a single wind arm o address this point separate wind arms have been

identified by grouping all turbines which have identical values or turbine manuacturer and

type local authority date o commissioning and date o decommissioning (where applicable)

he register contains inormation on 7569 turbines commissioned rom 1977 onwards his

includes a large number o small turbines which would not all within the scope o the ROC

scheme in the UK Hence or comparability with the UK data all turbines commissioned

prior to 1992 were dropped as well as all wind arms with a generating capacity o less than 1

MW he resulting sample covered 823 onshore wind arms and 30 offshore wind arms with

monthly load actors or various periods rom January 2002 to August 2012 Given the cut-off

or wind arms commissioned prior to 1992 the maximum age or an onshore wind arm was 20

years and the maximum age or an offshore wind arm was 17 years

he total capacity o the onshore wind arms recorded in the Danish database was 2757 MW

he average size o an onshore wind arm in Denmark is small ndash only 33 MW as compared

with 161 MW or the UK ndash because 84 o the sample consists o installations with between

1 and 4 turbines he largest onshore installation has 39 turbines with a capacity o 234 MW

while the onshore installation with the greatest capacity has 19 turbines with a capacity o 584

MW Overall the sample o Danish onshore wind arms is older and smaller than the sample

o UK onshore wind arms reflecting the much longer history o wind generation in Denmark

16 httpwwwensdkEN-USINFOFACSANDFIGURESENERGY_SAISICS_AND_INDICAORS

OVERVIEWOFHEENERGYSECORREGISEROFWINDURBINESSiderForsideaspx

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2852

26 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

he larger scale o UK wind arms may account or some o the differences in perormance

curves or the two countries

he sample o offshore wind arms in Denmark is quite small with 30 installations in

total Still it is useul to analyse the perormance o these installations since there is very little

comparable data or the UK he sample covers 394 turbines with an average o just over 13turbines per installation and an average capacity o 288 MW he largest installation which

was commissioned in 2003 has 72 turbines with a total capacity o 1656 MW Again offshore

installations are smaller and older in Denmark than is the current pattern in the UK he first

two wind arms in the sample were commissioned in the years 1995 and 2000 but the sample

increases significantly rom 2002 onwards o avoid reliance upon a sample o 1 or 2 wind

arms to determine specific coefficients or ages gt 10 years all observations with an age greater

than 10 were omitted rom the analysis17

A somewhat different problem with the offshore wind arm data arose because the coeffi-

cients estimated using capacity weights (see below) were heavily influenced by the largest windarm his has had an extremely erratic profile o perormance over time with a very low average

load actor he treatment o this wind arm is discussed in more detail in Section F below

C Specication and estimation methods

As Figure 8 illustrates it is necessary to go beyond simple summary statistics in order to assess

whether there is any systematic relationship between age and the average perormance o wind

arms he perormance o wind arms varies over both time and space some months and years

have more or less wind than the long term average while specific characteristics o wind arms

ndash including location type o turbine and operating regime ndash will influence the perormance oeach plant under identical wind conditions

he combination o (a) site-specific characteristics which are constant over time and (b)

period-specific characteristics which are constant across wind arms can be represented by

what statisticians call an error components model with fixed effects or each wind arm and

each time period he dependent variable in the model consists o a sequence o load actors ndash

denoted by or unit i in month t ndash or each wind arm (the panel unit) over time he load

actors are calculated by dividing the total output in MWh by (24 x number o days in month x

reported generating capacity in MW) and are multiplied by 100 to convert to percentages

he additive version o the error components model may be written as

(1)

in which denotes the age o plant i in period t and is some unction that is either a

representation o or an approximation to the normalised age-perormance relationship he

multiplicative version is similar except that is replaced by where ln() denotes the

natural logarithm he terms are wind arm or unit fixed effects which capture actors such

as location type o turbine operating regimes etc he terms are period fixed effects which

capture the influence o wind conditions over the UK or Denmark as a whole as well as seasonal

maintenance and demand Without loss o generality it may be assumed that the fixed effects are

17 For the avoidance o doubt this means that no attempt was made to estimate the effect o age on the perormance o

offshore wind arms beyond age 10 while the data or the oldest wind arms was included in the analysis up to and

including age 10

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 27

normalised so that ie all o the error components have a zero mean while the

mean o the random error is zero by definition he terms are random errors which capture

random variations in wind conditions that are not specific to the site turbine breakdowns and

other actors that are uncorrelated with either the site or the time period It is assumed that the

mean o the random error is zero while additional assumptions depend upon the method o

estimation that is adopted

he model specified in equation (1) can be estimated in a variety o ways18 A crucial issue is

whether the are or may be significantly correlated with the age terms or any other independent

regressors I it is assumed that there is no such correlation the model is usually reerred to as

a random effects model In that case the equation can be estimated by a variant o least squares

(LS) generalised least squares (GLS) generalised methods o moments (GMM) or maximum

likelihood (ML) ndash all o which should be yield parameter estimates that will converge in prob-

ability to the true parameter values as the sample size increases An alternative but less restric-

tive specification is known as the fixed effects model in which no assumption is made about

the correlation between the unit fixed effects and other regressors he limitation o the fixed

effects model is that it cannot be used to estimate parameters or regressors that vary across

panel units but are constant over time or a panel unit (time-invariant regressors) ndash eg the

year in which a wind arm was commissioned its location or its rated generating capacity he

influence o such variables is captured by the unit fixed effects

In order to avoid the risk o obtaining biased estimates o the age effects the results reported

in this note are based upon estimates using the fixed effects model he influence o time-invar-

iant regressors is examined in a second stage by regressing the estimates o the unit fixed effectson these variables he simplest method o estimating the fixed effects model is known as the

lsquowithinrsquo estimator in which the mean value or each unit is subtracted rom all observations or

that unit For example i is a general polynomial o order M in A equation (1) becomes

(2)

so that the model can be estimated by least square with the inclusion o dummy variables or

each period t he variance-covariance matrix o the coefficients have been estimated by using

(a) a robust sandwich estimator adjusted or clustering by panel unit and (b) a bootstrap esti-

mator with 400 repetitions he estimates o the standard errors will be consistent even i the

random error component is serially correlated within panel units andor has different variances

across panel units19 Since the number o panel units and the average number o time periods

per panel units are both large (other than or Danish offshore wind arms) the least square esti-

mates o the parameters will be unbiased under these assumptions

he bootstrap method is expensive to carry out but it provides a useul cross-check on

the standard errors generated by other methods Unortunately the bootstrap method o

18 See Chapter 21 o AC Cameron amp PK rivedi ndash Microeconometrics Methods and Applications (Cambridge

Cambridge University Press 2005) or a standard textbook treatment o estimation methods or panel data models

19 he estimation was carried out using the xtreg procedure in Stata Version 12 he methods o constructing the

sandwich and bootstrap estimates o the variance-covariance matrix are described in Chapters 8 amp 13 o AC

Cameron amp PK rivedi ndash Microeconometrics using Stata (Revised Edition College Station exas Stata Press 2010)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 16: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1652

14 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

15 Outside the range between the upper and lower quartiles the top and bottom segments o

the distributions extend urther up and down or UK onshore wind arms than or those in

Denmark he rather different pattern or Danish offshore wind installations is partly a conse-

quence o the relatively small size o the sample but the evidence points to considerably greater

variability in the perormance o offshore wind installations than or onshore installations16 It is somewhat surprising to observe the magnitude o the differences in the perormance in

onshore wind arms in the UK Given the nature o the subsidy regimes and the high capital

cost o developing new installations it might be expected that operators have a strong incen-

tive to identiy the best locations and then choose equipment that will deliver the maximum

amount o electricity output at a high level o reliability For UK onshore operators it seems that

good or bad perormance is somewhat o a lottery However i the subsidies provided by ROCs

are sufficient to underwrite investment in inefficient plants ndash as appears to be the case ndash then

those subsidies are extremely generous or plants that operate close to the efficient rontier

As location is likely to be the main actor that determines the perormance o a specific plant

relative to all other plants the inerence must be that many wind plants have been developed on

sites with poor wind characteristics

Figure 4 Initial load factors by year of commissioning for onshore wind

10

15

20

25

30

35

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

I n i t i a l l o a d f a c t o r ( )

Year commissioned

UK onshore DK onshore Linear (UK onshore) Linear (DK onshore)

Source Authorrsquos estimates

17 here is another disturbing characteristic o UK wind arms which is revealed by examination

o the unit fixed effects and illustrated in Figure 4 It might be expected that the average peror-

mance o recently installed units would be higher than that o older units his should reflect

improvements in turbine design and reliability as well as in the identification o good sites or

wind generation However that is not the pattern or onshore wind arms in the UK he figure

shows the average values o the initial load actors calculated using equally weighted data by

year o commissioning or onshore wind arms in the UK and Denmark together with fitted

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 15

trend lines8 here has been a clear decline in the initial load actors in the UK which has not

occurred in Denmark his pattern is confirmed by more detailed statistical results shown in

the Appendix which take account o location and size o installation

18 he decline in initial perormance is particularly obvious afer 2005 when the rate o commis-

sioning new capacity increased sharply rom less than 500 MW in the 5 years 2000ndash04 to morethan 500 MW per year rom 2005 onwards In contrast the peak years or commissioning new

onshore wind arms in Denmark were 2000ndash02 when 920 MW was installed he rate o new

building (or replacement) picked up again in 2008 but it is still much lower than at the turn o

century Not only is the perormance o the onshore wind plants in the UK commissioned in

recent years significantly worse than that or wind arms commissioned beore 2005 but it is

possible that this is a direct consequence o an overly rapid expansion o capacity In essence

the evidence suggests that the industry does not have the capability to identiy develop and

operate new onshore wind arms at the rate envisaged by UK government targets while main-

taining a satisactory average level o perormance

Implications for future policy and performance

19 he pattern o perormance degradation with age and over time identified in this analysis has

implications or the assessment o new investment in wind arms and or the design o the

proposed Electricity Market Reorm (EMR) here are two related issues that are affected by

the results

a) What are the implications or the design o policies intended to promote the adoption

and expansion o wind generation in the UK his includes the structure o current

subsidies as well as the specification o the contracts that are proposed under the EMR A

related question concerns the expected lie o new wind arms

b) How might the average perormance o wind generation develop over the next decade

his is absolutely critical to any assessment o the amount o wind capacity that will be

required to meet the UK governmentrsquos targets or the share o electricity ndash and more

generally energy ndash rom renewable sources he government has assumed that the

average load actors or both onshore and offshore wind arms will either remain stable

or increase in uture he prospects will be very different i these assumptions are not

borne out

20 he first question can be addressed by considering the discounted sum o cumulative net

output rom a new wind arm on the assumption that its perormance matches the averages

identified in the analysis his takes account o both perormance degradation over time plus

the annual cost o maintaining the plant which may be expressed as a deduction rom gross

production so as to calculate the contribution o expected output at age s to recovering the

original capital cost o building the plant hese figures have to be discounted back to the date

o commissioning in order to assess the net present value o the initial investment in the plant

he curves or constant and observed perormance using a discount rate o 9 (a typical cost o

capital or wind generation) are shown in Figure 5 he black line is based on the ull set o age

8 he initial load actors are calculated as the normalised load actor or UKDanish wind arms at age 0 multiplied by

where is the unit fixed effect or installation i

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1852

16 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

effects while the red dashed line is based on the smoothed quadratic perormance curve and

the blue dashed line shows what the pattern would be i operating perormance was not affected

by age he calculation is extended out to age 25 since DECCrsquos standard comparisons assume

that wind turbines have a lie o 25 years

Figure 5 Impact of performance degradation on discounted cumulative output

0

10

20

30

40

50

60

70

80

90

100

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

o

f d i s c o u n t e d c u m u l a t i v e o u t p u t

Age of plant (years)

Constant Quadratic Full age effects

Source Authorrsquos estimates

21 he key points in the figure are that 80 o the discounted cumulative output o a new wind

plant is likely to be produced in the first 10 years o its lie and 90 in the first 14 years his

is consistent with the structure o Danish subsidies or onshore wind arms which extend over

a typical period o 125 years Since sites or wind arms are scarce and involve the payment o

significant rents that may be linked to output it is very unlikely that any new installation o

wind turbines will have an expected lie o more than 15 years Instead as has happened in the

past wind operators will have a strong incentive to decommission plants afer no more than 15

years and replace the turbines with newer equipment

22 As a consequence any economic assessment o wind generation should not be based on an

expected lie which is longer than 15 years In recent work reported in evidence to the House

o Commons Select Committee on Energy and Climate Change I assumed that wind plants

would have a residual value equal to 20 o their initial cost in real terms at the end o 15 years

he analysis in this paper suggests that this is too avourable an assumption Given the costs o

decommissioning old turbines the residual value is likely to be well below 10 o their initial

cost and the decision point may be at 10 rather than 15 years

23 he corollary o this observation is that it makes little sense to offer long term contracts or 20

years or more that guarantee prices or eed-in tariffs (Fis) to wind operators A contract length

o 10 to 12 years would be sufficient to remove most o the market risk associated with invest-

ment in wind generation In this respect the subsidy arrangements implemented in Denmark

are better designed

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 17

24 At the same time the offer o subsidies andor guaranteed prices may have serious adverse

consequences or the efficiency o wind generation Returning to the unit fixed effects measur-

ing the perormance o wind plants commissioned afer 2005 only 28 out o 159 units have an

operating perormance that exceeds the average or the period hose 28 units account or 360

MW o capacity whereas the remaining 131 units account or 2810 MW o capacity Not onlyare recent plants less efficient than the average or the whole period but the plants which are

below-average in efficiency are typically larger than the more efficient ones and account or a

disproportionate share o the recent additions to generating capacity

25 In mid-2012 there was approximately 4500 MW o onshore wind capacity operating in the

UK By 2020 it is expected that there will be at least 10000 MW (based on proposed wind

arms with planning consent) and perhaps as much 15000 MW o wind capacity in operation

he government is relying upon a substantial contribution rom onshore wind arms towards

meeting its targets or renewable energy in 2020 o provide context the governmentrsquos targets

or renewable energy in 2020 speciy a total o 234 Wh o energy rom all renewable sources

o which 90ndash140 Wh is expected to come rom wind and biomass electricity generation9

he projections or biomass electricity are extremely vague and have to be reconciled with

an assumed increase o 3ndash4 times in the amount o biomass used separately or heat So in

concrete terms it is reasonable to assume that wind generation will account or at least 90 Wh

with 24ndash32 Wh rom onshore generation and the remainder rom offshore

26 he average load actor or onshore wind over the 10 year period rom April 2002 to March

2012 was 256 which is influenced by the large amounts o new capacity added rom 2005

onwards I this load actor were achieved in 2020 the total amount o onshore wind capacityrequired to meet the onshore generation target would be 107 to 143 GW he average load

actor or offshore wind installations (on an unchanged configuration basis) or the five year

period 2007ndash11 was 32010 I this load actor were achieved in 2020 the total amount o

offshore wind capacity required to meet the remainder o the 90 Wh target would be 206

to 235 GW though these estimates would all to 180 to 205 GW i the average load actor

matched the historic average value o 367 or Danish offshore wind installations

27 he question that has to be addressed is whether these load actors are consistent with (a) the

investment programmes that would be required to achieve the total amounts o generation

capacity implied by the governmentrsquos targets and (b) the age profiles o the perormance o

wind arms discussed in this paper he short answer is that this is simply not possible i the

initial load actor or new wind arms continues to decline at the rate observed the past decade

So even as a starting point it is necessary to assume that this underlying deterioration in wind

arm perormance ceases his would be a major change in itsel and it is not obvious what

actors might lead to such a result

28 he results illustrated below are based on a vintage model o perormance that takes account

o the actual expansion o capacity up to the end o 2011 plus whatever investment in new or

replacement turbines is required to achieve the target levels o wind capacity in 2020 o high-light the ull range o potential outcomes these targets are either 10 GW or 15 GW o onshore

9 Department o Energy and Climate Change ndash UK Renewable Energy Road Map July 2011 Figure 2

10 he equivalent average load actor or onshore wind arms was 264

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2052

18 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

wind and either 18 or 24 GW o offshore wind he levels o investment required to meet these

targets depend upon the expected lie o wind turbines For onshore turbines an expected lie

o 10 years is consistent with re-powering decisions over the last decade while an expected lie

o 15 years seems to be the maximum consistent with the age-perormance profile For offshore

turbines the analysis uses expected lives o 15 and 20 years

Figure 6A Projected load factors for UK onshore wind (total capacity of 10 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

Figure 6B Projected load factors for UK onshore wind (total capacity of 15 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

29 Figure 6A shows projected values o the average normalised load actors or UK onshore wind

arms in 2015 and 2020 using a target capacity in 2020 equal to 10 GW plus the age-peror-

mance profiles or the UK and Denmark (DK) Since vintage effects are important in this

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 19

calculation the profile o new or replacement investment affects the results In this case the

level o gross investment is assumed to be constant rom 2013 onwards at either 930 or 1220

MW per year ndash the range is defined by the assumed lie o turbines Achieving a target o 15

GW in 2020 (Figure 6B) would require a level o investment rom 2013 onwards o either 1540

or 1930 MW per year Such figures are 3ndash4 times the average level o investment over theperiod 2005ndash11 so the lower target may provide a better indication o what will happen

30 All o the estimates are calibrated to match the observed average load actor in 2007ndash11 so that

they should be interpreted as relative rather than absolute values Using the age-perormance

profile or UK onshore wind installations the average normalised load actor is likely to all by

about 19 rom 2007ndash11 to 2015 and by 23ndash28 to 2020 I the age-perormance curves or

Danish onshore installations were to apply the reduction in the average normalised load actor

would be considerably less but there would still be a all o up to 5 by 2020

31 In the medium term the all is greater i turbines operate or 15 years because the average age

o all turbines will be higher and thus the degradation in perormance due to age will have a

larger effect By 2020 this effect reduces the average normalised load actor rom 203 or a 10

year lie to 195 or a 15 year lie here is a complex economic and commercial trade-off here

o meet a target o 10 GW in 2020 it is necessary to increase the level o investment in onshore

capacity by 30 rom 2013 Almost all o this will take the orm o re-powering existing wind

arms11 he costs are airly high because this will usually involve new development consents

decommissioning the original turbines and replacing them with a smaller number o larger and

more powerul turbines

32 Estimates o re-powering costs are generally treated as being commercially sensitive but broadproject costs cited in specialist magazines suggest that the unit costs are likely to be at least

pound1000 per kW or more than pound300 million per year in aggregate On the other hand the increase

in the amount o energy generated would be about 12 Wh or a 10 year turbine lie rather

than a 15 year turbine lie though this does not allow or the loss o production during re-pow-

ering Under the best circumstances the payback period may be quite short but the best option

rom a broader economic and social perspective is not obvious

33 Even under the best scenario 10 GW o onshore wind capacity will only generate 178 Wh o

electricity in 2020 well below the lower end o the target range or onshore wind generation

specified in DECCrsquos road map In act allowing or interactions between the age-perormance

profile and investment the amount o onshore wind capacity required in 2020 to generate 24

Wh would be a minimum o 134 GW At the top end o DECCrsquos target range the capacity

required in 2020 to generate 32 Wh would be a minimum o 178 GW Since these estimates

are based on a 10 year turbine lie the level o investment required rom 2013 onwards would

be 1650 MW per year or the bottom o the range and 2350 MW or the top end o the range

In the last 8 years total investment in onshore wind has only ever exceeded 600 MW in one

11 Re-powering is usually defined as the ull replacement o existing turbines whereas retrofitting covers the replace-

ment o various mechanical or electrical components In the data analysis re-powering involves the creation o a

new unit whereas retrofitting affects the perormance o a continuing unit Hence the potential benefits o retrofit-

ting are taken into account in constructing the age-perormance curves

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2252

20 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

year (883 MW in 2008) so it seems rather unlikely that even the lower end o DECCrsquos range or

onshore wind generation will actually be realised

34 O course this assessment would be different i UK onshore wind operators were able to match

the Danish age-perormance curve here are many actors which differentiate the UK and

Danish wind industries he average scale o wind arms in the UK is larger they are moreremote and there appears to be significantly greater variation between operating and potential

new sites in achieved perormance It would thereore require a very optimistic observer to

conclude that the difference in age-perormance curves between the two countries will disap-

pear in the short or medium term

Figure 7 Projected load factors for UK offshore wind in 2020

20

22

24

26

28

30

32

34

18 GW ndash life 15 yrs 18 GW ndash life 20 yrs 24 GW ndash life 15 yrs 24 GW ndash life 20 yrs

L o a d f a c t o r ( )

2007ndash11 2015 2020

Note Based on Danish age-perormance curves Source Authorrsquos estimates

35 Figure 7 shows the results rom carrying out a similar exercise or offshore wind arms in the

UK he only evidence on the age-perormance curves or offshore installations is based upon

the Danish experience so these have been used to construct the projections Further the profile

o investment assumes a progressive increase in additions to capacity rom 2013 to 2016 and

then a stable level o investment rom 2017 to 2020 he DECC road map involves a very large

commitment to offshore wind investment ndash in the range 25-35 GW o new capacity annually

rom 2016 onwards As a consequence the average age o offshore capacity in 2020 will be quite

low and the ull effects o the age-perormance curves will only be elt during the decade ollow-

ing 2020 Even so the average normalised load actor in 2020 will be about 15 lower than or

2007ndash11 while by 2025 it will be about 30 below the base level

36 On these projections 232 GW o offshore capacity is the minimum necessary to generate 58

Wh o electricity rom offshore wind in 2020 his requirement will increase to 298 GW by

2025 simply to offset the effect o the ageing o offshore generation capacity on total output

37 Combining the estimates or onshore and offshore wind generation the total amount o wind

capacity required in 2020 to produce 90 Wh o electricity rom wind will be in the range 39

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 21

to 41 GW once allowance is made or the observed age-perormance curves or both onshore

and offshore installations hese estimates are roughly a third higher than official predictions o

28ndash31 GW o wind capacity in 2020

38 urning the figures round gives a slightly different picture o the reality o the governmentrsquos

strategy on renewable electricity generation A total o 30 GW o wind capacity in 2020 (12GW onshore 18 GW offshore) will generate about 64 Wh o electricity in a normal year

Since electricity generation rom wind and biomass are expected to account or 50 o renew-

able energy in 2020 the consequence is that biomass is expected to account or 53 Wh In

2011 landfill gas and sewage sludge digestion accounted or 44 o electricity rom bioenergy

(DUKES ables 64) here are limited opportunities to expand production rom these sources

so a continuation o historic rates o growth would mean that renewable generation rom other

sources o bioenergy would have to increase rom 72 Wh in 2011 to 46 Wh in 2020 here

are severe constraints on increasing the amounts o electricity generated rom animal biomass

biodegradable municipal solid waste and anaerobic digestion so it would be optimistic to

assume that these will contribute more than 6-8 Wh in 2020 (up rom 26 Wh in 2011)

39 In practice i the projections or renewable energy are to be believed they rest upon an assump-

tion that there will be a large increase in the amount o electricity that is generated rom plant

biomass ndash either on its own or through co-firing with coal his increase will involve increas-

ing the amount o electricity generated in these ways rom 46 Wh in 2011 to 38-40 Wh in

2020 Since the amounts o straw and UK-grown timber available or this purpose are strictly

limited this will come down to the amount o wood chips that can be imported and the cost o

doing this40 he historic average load actor or biomass plants is just over 60 so about 7500 MW o

biomass capacity will be required to supply 40 Wh o electricity he weight o wood chips per

MWh o electricity depends upon moisture content storage boiler design and other actors but

an indicative range is 085 to 115 tonnes per MWh o wood chips with a moisture content o

30 According to DECC and Forestry Commission estimates the UK imported 265 million

tonnes o steam coal and 26 million tonnes o other wood products (including wood chips)

in 2011 It seems that the net effect o the governmentrsquos strategy will be to replace 15 million

tonnes o coal imported in 2011 by 32 to 44 million tonnes o imported wood chips his is a

somewhat strange way o increasing energy security

Conclusion

41 Wind power is a highly capital-intensive technology or generating electricity Its merits rely

entirely upon a substitution o capital or uel inputs he same is true or hydro or tidal or wave

power In comparison with hydro power wind is a low quality resource because o its variability

and because it cannot be stored So the economic case or wind power must rest on obtaining

the most out o the wind that is available

42 While the decline in the achieved perormance o onshore wind turbines in Denmark is

much less than that or the UK or offshore nonetheless the decline in expected output under

standardised wind conditions over 10 years is 10 unweighted and 13 capacity weighted

hese declines accelerate afer age 10 so that the reductions in perormance are 17 and 20

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2452

22 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

respectively afer 15 years For UK onshore wind arms the reduction in perormance due to

age is much worse at 27 unweighted and 69 capacity weighted by age 10

43 Evidence on the perormance o Danish offshore installations is both restricted and so poor

that there may be concern that the results are affected by a small number o outliers Still the

sample contains a reasonable number o sites with at least 5 years o operating experience andthe decline in perormance by age 5 is 38 unweighted and 26 capacity weighted

44 In addition to these results there is strong evidence that the average normalised load actor or

new onshore wind installations in the UK has allen significantly over the period rom 2000 to

2011 his is consistent with a pattern in which the most avourable sites are developed first

Equally it could mean that wind developers have been unable to keep up with the rate o new

investment while maintaining the quality o development and operations For example the site

design or selection o turbine characteristics may make less effective use o the available wind

resources or the sites available than was the case in the past

45 Whatever the reasons the deterioration in initial perormance means that the expected returns

rom the expansion in wind capacity both or investors and in terms o the reduction in CO983090

emissions have been alling without a concomitant decrease in the private and social costs

that are borne by customers and the general public Clearly this is unsatisactory at best and it

suggests that the benefits claimed or current policies cannot be taken at ace value

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 23

Appendix Data and Methods

A Data for the United Kingdom

he raw data used or this study was extracted by the Renewable Energy Foundation rom the

Renewables and CHP Register database compiled by Ogem12 his inormation is used in the

administration o the market in Renewable Obligation Certificates (ROCs) he Renewables

Obligation is the primary mechanism by which large scale generators o renewable electric-

ity receive subsidies so the operators o wind arms have a very strong incentive to submit

complete and up-to-date inormation to the database he data extracted covered all onshore

wind generators with at least 05 MW o generating capacity which are eligible or ROCs or the

10 year period April 2002 to March 2012 he key variables are the monthly output o electric-

ity (used or the purpose o allocating ROCs) and nominal generating capacity here are 282

reporting units in the final dataset A reporting unit may correspond to an entire wind arm orto separate phases o development Complications arise when an existing wind arm is re-pow-

ered ie when old wind turbines are replaced by newer and usually more powerul turbines

he database reports this as a change in the generating capacity o the recording unit but in

this analysis repowering is treated as the termination o the old recording unit and the creation

o a new recording unit

A considerable amount o data checking and cleaning was required beore the data could

be analysed he first step was to add the age o each reporting unit his was calculated by

reerence to the month in which the turbines were commissioned For wind arms in operation

in or beore June 2002 the commissioning date was obtained rom various online sources ndash

notably the thewindpowernet database o UK wind arms13 which was cross-checked against

the inormation provided on the websites o wind operators and other sources For recording

units whose first data relates to dates afer June 2002 the commissioning date was initially

assumed to be the month or which data is first reported Cross-checks were carried out or

all recording units which commenced reporting in the period July 2002 to December 2003 In

more than 80 o cases the externally-reported commissioning date is within one month o the

first month or which data is reported he exceptions ndash Bu Farm and Mablethorpe ndash appear to

be due to delays by small operators in reporting data to Ogem

In a number o cases it is clear rom the data that plants were not in ull operation at the

time when data was first reported to the database here are two indicators o partial or delayed

commissioning

bull For an initial series o months the calculated load actor is very low ndash usually below 5 ndash

and then increases abruptly In such cases the data or the early months is dropped and the

commissioning date is treated as the month immediately preceding the first month with a

lsquonormalrsquo load actor14

12 httpswwwrenewablesandchpogemgovuk

13 See httpwwwthewindpowernetwindarms_list_enphp

14 he month in which a plant is reported as having been commissioned is not included in the analysis

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2652

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 25

the decline in the plantrsquos load actor as it ages As an example consider a plant whose reported

capacity was increased rom 99 MW to 10 MW at age 3 I this was simply a change in report-

ing practice this would reduce the estimated load actor rom age 3 onwards By increasing

the initial capacity rom 99 MW to 10 MW ndash or by reducing the capacity rom 10 MW to 99

MW rom age 3 onwards ndash this bias is removed It is not important which adjustment is madebecause the difference between the alternative assumptions is captured by the unit fixed effect

he numbers o separate sites and o observations per site mean that it is not possible to

provide a simple visual representation o the data Figure 8 provides a summary by illustrating

the distributions o load actors by the age o the wind arm pooled across all years he ranges

o the load actors or each age are wide while the medians show no clear trend by age hus a

naiumlve analysis might conclude that there is nothing to investigate However as will be explained

below the distributions mask crucial differences between the perormance o wind arms as

they age because they do not control or the differences in location and wind availability over

time

B Data for Denmark

he data or Danish wind arms used in this study comes rom a database compiled by the

Danish Energy Agency covering the characteristics and perormance o all wind turbines rom

2002 up to the end o August 201216 he basic recording unit or the register is the individual

wind turbine but where a number o turbines are connected to a single meter an identical

average output per turbine is recorded or all turbines in the group he presence o what are in

effect repeated observations increases the noise in the data In addition the model that is used

or the analysis assumes that site operator and turbine characteristics are common or all o theunits that make up a single wind arm o address this point separate wind arms have been

identified by grouping all turbines which have identical values or turbine manuacturer and

type local authority date o commissioning and date o decommissioning (where applicable)

he register contains inormation on 7569 turbines commissioned rom 1977 onwards his

includes a large number o small turbines which would not all within the scope o the ROC

scheme in the UK Hence or comparability with the UK data all turbines commissioned

prior to 1992 were dropped as well as all wind arms with a generating capacity o less than 1

MW he resulting sample covered 823 onshore wind arms and 30 offshore wind arms with

monthly load actors or various periods rom January 2002 to August 2012 Given the cut-off

or wind arms commissioned prior to 1992 the maximum age or an onshore wind arm was 20

years and the maximum age or an offshore wind arm was 17 years

he total capacity o the onshore wind arms recorded in the Danish database was 2757 MW

he average size o an onshore wind arm in Denmark is small ndash only 33 MW as compared

with 161 MW or the UK ndash because 84 o the sample consists o installations with between

1 and 4 turbines he largest onshore installation has 39 turbines with a capacity o 234 MW

while the onshore installation with the greatest capacity has 19 turbines with a capacity o 584

MW Overall the sample o Danish onshore wind arms is older and smaller than the sample

o UK onshore wind arms reflecting the much longer history o wind generation in Denmark

16 httpwwwensdkEN-USINFOFACSANDFIGURESENERGY_SAISICS_AND_INDICAORS

OVERVIEWOFHEENERGYSECORREGISEROFWINDURBINESSiderForsideaspx

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2852

26 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

he larger scale o UK wind arms may account or some o the differences in perormance

curves or the two countries

he sample o offshore wind arms in Denmark is quite small with 30 installations in

total Still it is useul to analyse the perormance o these installations since there is very little

comparable data or the UK he sample covers 394 turbines with an average o just over 13turbines per installation and an average capacity o 288 MW he largest installation which

was commissioned in 2003 has 72 turbines with a total capacity o 1656 MW Again offshore

installations are smaller and older in Denmark than is the current pattern in the UK he first

two wind arms in the sample were commissioned in the years 1995 and 2000 but the sample

increases significantly rom 2002 onwards o avoid reliance upon a sample o 1 or 2 wind

arms to determine specific coefficients or ages gt 10 years all observations with an age greater

than 10 were omitted rom the analysis17

A somewhat different problem with the offshore wind arm data arose because the coeffi-

cients estimated using capacity weights (see below) were heavily influenced by the largest windarm his has had an extremely erratic profile o perormance over time with a very low average

load actor he treatment o this wind arm is discussed in more detail in Section F below

C Specication and estimation methods

As Figure 8 illustrates it is necessary to go beyond simple summary statistics in order to assess

whether there is any systematic relationship between age and the average perormance o wind

arms he perormance o wind arms varies over both time and space some months and years

have more or less wind than the long term average while specific characteristics o wind arms

ndash including location type o turbine and operating regime ndash will influence the perormance oeach plant under identical wind conditions

he combination o (a) site-specific characteristics which are constant over time and (b)

period-specific characteristics which are constant across wind arms can be represented by

what statisticians call an error components model with fixed effects or each wind arm and

each time period he dependent variable in the model consists o a sequence o load actors ndash

denoted by or unit i in month t ndash or each wind arm (the panel unit) over time he load

actors are calculated by dividing the total output in MWh by (24 x number o days in month x

reported generating capacity in MW) and are multiplied by 100 to convert to percentages

he additive version o the error components model may be written as

(1)

in which denotes the age o plant i in period t and is some unction that is either a

representation o or an approximation to the normalised age-perormance relationship he

multiplicative version is similar except that is replaced by where ln() denotes the

natural logarithm he terms are wind arm or unit fixed effects which capture actors such

as location type o turbine operating regimes etc he terms are period fixed effects which

capture the influence o wind conditions over the UK or Denmark as a whole as well as seasonal

maintenance and demand Without loss o generality it may be assumed that the fixed effects are

17 For the avoidance o doubt this means that no attempt was made to estimate the effect o age on the perormance o

offshore wind arms beyond age 10 while the data or the oldest wind arms was included in the analysis up to and

including age 10

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 27

normalised so that ie all o the error components have a zero mean while the

mean o the random error is zero by definition he terms are random errors which capture

random variations in wind conditions that are not specific to the site turbine breakdowns and

other actors that are uncorrelated with either the site or the time period It is assumed that the

mean o the random error is zero while additional assumptions depend upon the method o

estimation that is adopted

he model specified in equation (1) can be estimated in a variety o ways18 A crucial issue is

whether the are or may be significantly correlated with the age terms or any other independent

regressors I it is assumed that there is no such correlation the model is usually reerred to as

a random effects model In that case the equation can be estimated by a variant o least squares

(LS) generalised least squares (GLS) generalised methods o moments (GMM) or maximum

likelihood (ML) ndash all o which should be yield parameter estimates that will converge in prob-

ability to the true parameter values as the sample size increases An alternative but less restric-

tive specification is known as the fixed effects model in which no assumption is made about

the correlation between the unit fixed effects and other regressors he limitation o the fixed

effects model is that it cannot be used to estimate parameters or regressors that vary across

panel units but are constant over time or a panel unit (time-invariant regressors) ndash eg the

year in which a wind arm was commissioned its location or its rated generating capacity he

influence o such variables is captured by the unit fixed effects

In order to avoid the risk o obtaining biased estimates o the age effects the results reported

in this note are based upon estimates using the fixed effects model he influence o time-invar-

iant regressors is examined in a second stage by regressing the estimates o the unit fixed effectson these variables he simplest method o estimating the fixed effects model is known as the

lsquowithinrsquo estimator in which the mean value or each unit is subtracted rom all observations or

that unit For example i is a general polynomial o order M in A equation (1) becomes

(2)

so that the model can be estimated by least square with the inclusion o dummy variables or

each period t he variance-covariance matrix o the coefficients have been estimated by using

(a) a robust sandwich estimator adjusted or clustering by panel unit and (b) a bootstrap esti-

mator with 400 repetitions he estimates o the standard errors will be consistent even i the

random error component is serially correlated within panel units andor has different variances

across panel units19 Since the number o panel units and the average number o time periods

per panel units are both large (other than or Danish offshore wind arms) the least square esti-

mates o the parameters will be unbiased under these assumptions

he bootstrap method is expensive to carry out but it provides a useul cross-check on

the standard errors generated by other methods Unortunately the bootstrap method o

18 See Chapter 21 o AC Cameron amp PK rivedi ndash Microeconometrics Methods and Applications (Cambridge

Cambridge University Press 2005) or a standard textbook treatment o estimation methods or panel data models

19 he estimation was carried out using the xtreg procedure in Stata Version 12 he methods o constructing the

sandwich and bootstrap estimates o the variance-covariance matrix are described in Chapters 8 amp 13 o AC

Cameron amp PK rivedi ndash Microeconometrics using Stata (Revised Edition College Station exas Stata Press 2010)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 17: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 15

trend lines8 here has been a clear decline in the initial load actors in the UK which has not

occurred in Denmark his pattern is confirmed by more detailed statistical results shown in

the Appendix which take account o location and size o installation

18 he decline in initial perormance is particularly obvious afer 2005 when the rate o commis-

sioning new capacity increased sharply rom less than 500 MW in the 5 years 2000ndash04 to morethan 500 MW per year rom 2005 onwards In contrast the peak years or commissioning new

onshore wind arms in Denmark were 2000ndash02 when 920 MW was installed he rate o new

building (or replacement) picked up again in 2008 but it is still much lower than at the turn o

century Not only is the perormance o the onshore wind plants in the UK commissioned in

recent years significantly worse than that or wind arms commissioned beore 2005 but it is

possible that this is a direct consequence o an overly rapid expansion o capacity In essence

the evidence suggests that the industry does not have the capability to identiy develop and

operate new onshore wind arms at the rate envisaged by UK government targets while main-

taining a satisactory average level o perormance

Implications for future policy and performance

19 he pattern o perormance degradation with age and over time identified in this analysis has

implications or the assessment o new investment in wind arms and or the design o the

proposed Electricity Market Reorm (EMR) here are two related issues that are affected by

the results

a) What are the implications or the design o policies intended to promote the adoption

and expansion o wind generation in the UK his includes the structure o current

subsidies as well as the specification o the contracts that are proposed under the EMR A

related question concerns the expected lie o new wind arms

b) How might the average perormance o wind generation develop over the next decade

his is absolutely critical to any assessment o the amount o wind capacity that will be

required to meet the UK governmentrsquos targets or the share o electricity ndash and more

generally energy ndash rom renewable sources he government has assumed that the

average load actors or both onshore and offshore wind arms will either remain stable

or increase in uture he prospects will be very different i these assumptions are not

borne out

20 he first question can be addressed by considering the discounted sum o cumulative net

output rom a new wind arm on the assumption that its perormance matches the averages

identified in the analysis his takes account o both perormance degradation over time plus

the annual cost o maintaining the plant which may be expressed as a deduction rom gross

production so as to calculate the contribution o expected output at age s to recovering the

original capital cost o building the plant hese figures have to be discounted back to the date

o commissioning in order to assess the net present value o the initial investment in the plant

he curves or constant and observed perormance using a discount rate o 9 (a typical cost o

capital or wind generation) are shown in Figure 5 he black line is based on the ull set o age

8 he initial load actors are calculated as the normalised load actor or UKDanish wind arms at age 0 multiplied by

where is the unit fixed effect or installation i

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1852

16 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

effects while the red dashed line is based on the smoothed quadratic perormance curve and

the blue dashed line shows what the pattern would be i operating perormance was not affected

by age he calculation is extended out to age 25 since DECCrsquos standard comparisons assume

that wind turbines have a lie o 25 years

Figure 5 Impact of performance degradation on discounted cumulative output

0

10

20

30

40

50

60

70

80

90

100

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

o

f d i s c o u n t e d c u m u l a t i v e o u t p u t

Age of plant (years)

Constant Quadratic Full age effects

Source Authorrsquos estimates

21 he key points in the figure are that 80 o the discounted cumulative output o a new wind

plant is likely to be produced in the first 10 years o its lie and 90 in the first 14 years his

is consistent with the structure o Danish subsidies or onshore wind arms which extend over

a typical period o 125 years Since sites or wind arms are scarce and involve the payment o

significant rents that may be linked to output it is very unlikely that any new installation o

wind turbines will have an expected lie o more than 15 years Instead as has happened in the

past wind operators will have a strong incentive to decommission plants afer no more than 15

years and replace the turbines with newer equipment

22 As a consequence any economic assessment o wind generation should not be based on an

expected lie which is longer than 15 years In recent work reported in evidence to the House

o Commons Select Committee on Energy and Climate Change I assumed that wind plants

would have a residual value equal to 20 o their initial cost in real terms at the end o 15 years

he analysis in this paper suggests that this is too avourable an assumption Given the costs o

decommissioning old turbines the residual value is likely to be well below 10 o their initial

cost and the decision point may be at 10 rather than 15 years

23 he corollary o this observation is that it makes little sense to offer long term contracts or 20

years or more that guarantee prices or eed-in tariffs (Fis) to wind operators A contract length

o 10 to 12 years would be sufficient to remove most o the market risk associated with invest-

ment in wind generation In this respect the subsidy arrangements implemented in Denmark

are better designed

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 17

24 At the same time the offer o subsidies andor guaranteed prices may have serious adverse

consequences or the efficiency o wind generation Returning to the unit fixed effects measur-

ing the perormance o wind plants commissioned afer 2005 only 28 out o 159 units have an

operating perormance that exceeds the average or the period hose 28 units account or 360

MW o capacity whereas the remaining 131 units account or 2810 MW o capacity Not onlyare recent plants less efficient than the average or the whole period but the plants which are

below-average in efficiency are typically larger than the more efficient ones and account or a

disproportionate share o the recent additions to generating capacity

25 In mid-2012 there was approximately 4500 MW o onshore wind capacity operating in the

UK By 2020 it is expected that there will be at least 10000 MW (based on proposed wind

arms with planning consent) and perhaps as much 15000 MW o wind capacity in operation

he government is relying upon a substantial contribution rom onshore wind arms towards

meeting its targets or renewable energy in 2020 o provide context the governmentrsquos targets

or renewable energy in 2020 speciy a total o 234 Wh o energy rom all renewable sources

o which 90ndash140 Wh is expected to come rom wind and biomass electricity generation9

he projections or biomass electricity are extremely vague and have to be reconciled with

an assumed increase o 3ndash4 times in the amount o biomass used separately or heat So in

concrete terms it is reasonable to assume that wind generation will account or at least 90 Wh

with 24ndash32 Wh rom onshore generation and the remainder rom offshore

26 he average load actor or onshore wind over the 10 year period rom April 2002 to March

2012 was 256 which is influenced by the large amounts o new capacity added rom 2005

onwards I this load actor were achieved in 2020 the total amount o onshore wind capacityrequired to meet the onshore generation target would be 107 to 143 GW he average load

actor or offshore wind installations (on an unchanged configuration basis) or the five year

period 2007ndash11 was 32010 I this load actor were achieved in 2020 the total amount o

offshore wind capacity required to meet the remainder o the 90 Wh target would be 206

to 235 GW though these estimates would all to 180 to 205 GW i the average load actor

matched the historic average value o 367 or Danish offshore wind installations

27 he question that has to be addressed is whether these load actors are consistent with (a) the

investment programmes that would be required to achieve the total amounts o generation

capacity implied by the governmentrsquos targets and (b) the age profiles o the perormance o

wind arms discussed in this paper he short answer is that this is simply not possible i the

initial load actor or new wind arms continues to decline at the rate observed the past decade

So even as a starting point it is necessary to assume that this underlying deterioration in wind

arm perormance ceases his would be a major change in itsel and it is not obvious what

actors might lead to such a result

28 he results illustrated below are based on a vintage model o perormance that takes account

o the actual expansion o capacity up to the end o 2011 plus whatever investment in new or

replacement turbines is required to achieve the target levels o wind capacity in 2020 o high-light the ull range o potential outcomes these targets are either 10 GW or 15 GW o onshore

9 Department o Energy and Climate Change ndash UK Renewable Energy Road Map July 2011 Figure 2

10 he equivalent average load actor or onshore wind arms was 264

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2052

18 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

wind and either 18 or 24 GW o offshore wind he levels o investment required to meet these

targets depend upon the expected lie o wind turbines For onshore turbines an expected lie

o 10 years is consistent with re-powering decisions over the last decade while an expected lie

o 15 years seems to be the maximum consistent with the age-perormance profile For offshore

turbines the analysis uses expected lives o 15 and 20 years

Figure 6A Projected load factors for UK onshore wind (total capacity of 10 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

Figure 6B Projected load factors for UK onshore wind (total capacity of 15 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

29 Figure 6A shows projected values o the average normalised load actors or UK onshore wind

arms in 2015 and 2020 using a target capacity in 2020 equal to 10 GW plus the age-peror-

mance profiles or the UK and Denmark (DK) Since vintage effects are important in this

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 19

calculation the profile o new or replacement investment affects the results In this case the

level o gross investment is assumed to be constant rom 2013 onwards at either 930 or 1220

MW per year ndash the range is defined by the assumed lie o turbines Achieving a target o 15

GW in 2020 (Figure 6B) would require a level o investment rom 2013 onwards o either 1540

or 1930 MW per year Such figures are 3ndash4 times the average level o investment over theperiod 2005ndash11 so the lower target may provide a better indication o what will happen

30 All o the estimates are calibrated to match the observed average load actor in 2007ndash11 so that

they should be interpreted as relative rather than absolute values Using the age-perormance

profile or UK onshore wind installations the average normalised load actor is likely to all by

about 19 rom 2007ndash11 to 2015 and by 23ndash28 to 2020 I the age-perormance curves or

Danish onshore installations were to apply the reduction in the average normalised load actor

would be considerably less but there would still be a all o up to 5 by 2020

31 In the medium term the all is greater i turbines operate or 15 years because the average age

o all turbines will be higher and thus the degradation in perormance due to age will have a

larger effect By 2020 this effect reduces the average normalised load actor rom 203 or a 10

year lie to 195 or a 15 year lie here is a complex economic and commercial trade-off here

o meet a target o 10 GW in 2020 it is necessary to increase the level o investment in onshore

capacity by 30 rom 2013 Almost all o this will take the orm o re-powering existing wind

arms11 he costs are airly high because this will usually involve new development consents

decommissioning the original turbines and replacing them with a smaller number o larger and

more powerul turbines

32 Estimates o re-powering costs are generally treated as being commercially sensitive but broadproject costs cited in specialist magazines suggest that the unit costs are likely to be at least

pound1000 per kW or more than pound300 million per year in aggregate On the other hand the increase

in the amount o energy generated would be about 12 Wh or a 10 year turbine lie rather

than a 15 year turbine lie though this does not allow or the loss o production during re-pow-

ering Under the best circumstances the payback period may be quite short but the best option

rom a broader economic and social perspective is not obvious

33 Even under the best scenario 10 GW o onshore wind capacity will only generate 178 Wh o

electricity in 2020 well below the lower end o the target range or onshore wind generation

specified in DECCrsquos road map In act allowing or interactions between the age-perormance

profile and investment the amount o onshore wind capacity required in 2020 to generate 24

Wh would be a minimum o 134 GW At the top end o DECCrsquos target range the capacity

required in 2020 to generate 32 Wh would be a minimum o 178 GW Since these estimates

are based on a 10 year turbine lie the level o investment required rom 2013 onwards would

be 1650 MW per year or the bottom o the range and 2350 MW or the top end o the range

In the last 8 years total investment in onshore wind has only ever exceeded 600 MW in one

11 Re-powering is usually defined as the ull replacement o existing turbines whereas retrofitting covers the replace-

ment o various mechanical or electrical components In the data analysis re-powering involves the creation o a

new unit whereas retrofitting affects the perormance o a continuing unit Hence the potential benefits o retrofit-

ting are taken into account in constructing the age-perormance curves

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2252

20 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

year (883 MW in 2008) so it seems rather unlikely that even the lower end o DECCrsquos range or

onshore wind generation will actually be realised

34 O course this assessment would be different i UK onshore wind operators were able to match

the Danish age-perormance curve here are many actors which differentiate the UK and

Danish wind industries he average scale o wind arms in the UK is larger they are moreremote and there appears to be significantly greater variation between operating and potential

new sites in achieved perormance It would thereore require a very optimistic observer to

conclude that the difference in age-perormance curves between the two countries will disap-

pear in the short or medium term

Figure 7 Projected load factors for UK offshore wind in 2020

20

22

24

26

28

30

32

34

18 GW ndash life 15 yrs 18 GW ndash life 20 yrs 24 GW ndash life 15 yrs 24 GW ndash life 20 yrs

L o a d f a c t o r ( )

2007ndash11 2015 2020

Note Based on Danish age-perormance curves Source Authorrsquos estimates

35 Figure 7 shows the results rom carrying out a similar exercise or offshore wind arms in the

UK he only evidence on the age-perormance curves or offshore installations is based upon

the Danish experience so these have been used to construct the projections Further the profile

o investment assumes a progressive increase in additions to capacity rom 2013 to 2016 and

then a stable level o investment rom 2017 to 2020 he DECC road map involves a very large

commitment to offshore wind investment ndash in the range 25-35 GW o new capacity annually

rom 2016 onwards As a consequence the average age o offshore capacity in 2020 will be quite

low and the ull effects o the age-perormance curves will only be elt during the decade ollow-

ing 2020 Even so the average normalised load actor in 2020 will be about 15 lower than or

2007ndash11 while by 2025 it will be about 30 below the base level

36 On these projections 232 GW o offshore capacity is the minimum necessary to generate 58

Wh o electricity rom offshore wind in 2020 his requirement will increase to 298 GW by

2025 simply to offset the effect o the ageing o offshore generation capacity on total output

37 Combining the estimates or onshore and offshore wind generation the total amount o wind

capacity required in 2020 to produce 90 Wh o electricity rom wind will be in the range 39

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 21

to 41 GW once allowance is made or the observed age-perormance curves or both onshore

and offshore installations hese estimates are roughly a third higher than official predictions o

28ndash31 GW o wind capacity in 2020

38 urning the figures round gives a slightly different picture o the reality o the governmentrsquos

strategy on renewable electricity generation A total o 30 GW o wind capacity in 2020 (12GW onshore 18 GW offshore) will generate about 64 Wh o electricity in a normal year

Since electricity generation rom wind and biomass are expected to account or 50 o renew-

able energy in 2020 the consequence is that biomass is expected to account or 53 Wh In

2011 landfill gas and sewage sludge digestion accounted or 44 o electricity rom bioenergy

(DUKES ables 64) here are limited opportunities to expand production rom these sources

so a continuation o historic rates o growth would mean that renewable generation rom other

sources o bioenergy would have to increase rom 72 Wh in 2011 to 46 Wh in 2020 here

are severe constraints on increasing the amounts o electricity generated rom animal biomass

biodegradable municipal solid waste and anaerobic digestion so it would be optimistic to

assume that these will contribute more than 6-8 Wh in 2020 (up rom 26 Wh in 2011)

39 In practice i the projections or renewable energy are to be believed they rest upon an assump-

tion that there will be a large increase in the amount o electricity that is generated rom plant

biomass ndash either on its own or through co-firing with coal his increase will involve increas-

ing the amount o electricity generated in these ways rom 46 Wh in 2011 to 38-40 Wh in

2020 Since the amounts o straw and UK-grown timber available or this purpose are strictly

limited this will come down to the amount o wood chips that can be imported and the cost o

doing this40 he historic average load actor or biomass plants is just over 60 so about 7500 MW o

biomass capacity will be required to supply 40 Wh o electricity he weight o wood chips per

MWh o electricity depends upon moisture content storage boiler design and other actors but

an indicative range is 085 to 115 tonnes per MWh o wood chips with a moisture content o

30 According to DECC and Forestry Commission estimates the UK imported 265 million

tonnes o steam coal and 26 million tonnes o other wood products (including wood chips)

in 2011 It seems that the net effect o the governmentrsquos strategy will be to replace 15 million

tonnes o coal imported in 2011 by 32 to 44 million tonnes o imported wood chips his is a

somewhat strange way o increasing energy security

Conclusion

41 Wind power is a highly capital-intensive technology or generating electricity Its merits rely

entirely upon a substitution o capital or uel inputs he same is true or hydro or tidal or wave

power In comparison with hydro power wind is a low quality resource because o its variability

and because it cannot be stored So the economic case or wind power must rest on obtaining

the most out o the wind that is available

42 While the decline in the achieved perormance o onshore wind turbines in Denmark is

much less than that or the UK or offshore nonetheless the decline in expected output under

standardised wind conditions over 10 years is 10 unweighted and 13 capacity weighted

hese declines accelerate afer age 10 so that the reductions in perormance are 17 and 20

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2452

22 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

respectively afer 15 years For UK onshore wind arms the reduction in perormance due to

age is much worse at 27 unweighted and 69 capacity weighted by age 10

43 Evidence on the perormance o Danish offshore installations is both restricted and so poor

that there may be concern that the results are affected by a small number o outliers Still the

sample contains a reasonable number o sites with at least 5 years o operating experience andthe decline in perormance by age 5 is 38 unweighted and 26 capacity weighted

44 In addition to these results there is strong evidence that the average normalised load actor or

new onshore wind installations in the UK has allen significantly over the period rom 2000 to

2011 his is consistent with a pattern in which the most avourable sites are developed first

Equally it could mean that wind developers have been unable to keep up with the rate o new

investment while maintaining the quality o development and operations For example the site

design or selection o turbine characteristics may make less effective use o the available wind

resources or the sites available than was the case in the past

45 Whatever the reasons the deterioration in initial perormance means that the expected returns

rom the expansion in wind capacity both or investors and in terms o the reduction in CO983090

emissions have been alling without a concomitant decrease in the private and social costs

that are borne by customers and the general public Clearly this is unsatisactory at best and it

suggests that the benefits claimed or current policies cannot be taken at ace value

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 23

Appendix Data and Methods

A Data for the United Kingdom

he raw data used or this study was extracted by the Renewable Energy Foundation rom the

Renewables and CHP Register database compiled by Ogem12 his inormation is used in the

administration o the market in Renewable Obligation Certificates (ROCs) he Renewables

Obligation is the primary mechanism by which large scale generators o renewable electric-

ity receive subsidies so the operators o wind arms have a very strong incentive to submit

complete and up-to-date inormation to the database he data extracted covered all onshore

wind generators with at least 05 MW o generating capacity which are eligible or ROCs or the

10 year period April 2002 to March 2012 he key variables are the monthly output o electric-

ity (used or the purpose o allocating ROCs) and nominal generating capacity here are 282

reporting units in the final dataset A reporting unit may correspond to an entire wind arm orto separate phases o development Complications arise when an existing wind arm is re-pow-

ered ie when old wind turbines are replaced by newer and usually more powerul turbines

he database reports this as a change in the generating capacity o the recording unit but in

this analysis repowering is treated as the termination o the old recording unit and the creation

o a new recording unit

A considerable amount o data checking and cleaning was required beore the data could

be analysed he first step was to add the age o each reporting unit his was calculated by

reerence to the month in which the turbines were commissioned For wind arms in operation

in or beore June 2002 the commissioning date was obtained rom various online sources ndash

notably the thewindpowernet database o UK wind arms13 which was cross-checked against

the inormation provided on the websites o wind operators and other sources For recording

units whose first data relates to dates afer June 2002 the commissioning date was initially

assumed to be the month or which data is first reported Cross-checks were carried out or

all recording units which commenced reporting in the period July 2002 to December 2003 In

more than 80 o cases the externally-reported commissioning date is within one month o the

first month or which data is reported he exceptions ndash Bu Farm and Mablethorpe ndash appear to

be due to delays by small operators in reporting data to Ogem

In a number o cases it is clear rom the data that plants were not in ull operation at the

time when data was first reported to the database here are two indicators o partial or delayed

commissioning

bull For an initial series o months the calculated load actor is very low ndash usually below 5 ndash

and then increases abruptly In such cases the data or the early months is dropped and the

commissioning date is treated as the month immediately preceding the first month with a

lsquonormalrsquo load actor14

12 httpswwwrenewablesandchpogemgovuk

13 See httpwwwthewindpowernetwindarms_list_enphp

14 he month in which a plant is reported as having been commissioned is not included in the analysis

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2652

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 25

the decline in the plantrsquos load actor as it ages As an example consider a plant whose reported

capacity was increased rom 99 MW to 10 MW at age 3 I this was simply a change in report-

ing practice this would reduce the estimated load actor rom age 3 onwards By increasing

the initial capacity rom 99 MW to 10 MW ndash or by reducing the capacity rom 10 MW to 99

MW rom age 3 onwards ndash this bias is removed It is not important which adjustment is madebecause the difference between the alternative assumptions is captured by the unit fixed effect

he numbers o separate sites and o observations per site mean that it is not possible to

provide a simple visual representation o the data Figure 8 provides a summary by illustrating

the distributions o load actors by the age o the wind arm pooled across all years he ranges

o the load actors or each age are wide while the medians show no clear trend by age hus a

naiumlve analysis might conclude that there is nothing to investigate However as will be explained

below the distributions mask crucial differences between the perormance o wind arms as

they age because they do not control or the differences in location and wind availability over

time

B Data for Denmark

he data or Danish wind arms used in this study comes rom a database compiled by the

Danish Energy Agency covering the characteristics and perormance o all wind turbines rom

2002 up to the end o August 201216 he basic recording unit or the register is the individual

wind turbine but where a number o turbines are connected to a single meter an identical

average output per turbine is recorded or all turbines in the group he presence o what are in

effect repeated observations increases the noise in the data In addition the model that is used

or the analysis assumes that site operator and turbine characteristics are common or all o theunits that make up a single wind arm o address this point separate wind arms have been

identified by grouping all turbines which have identical values or turbine manuacturer and

type local authority date o commissioning and date o decommissioning (where applicable)

he register contains inormation on 7569 turbines commissioned rom 1977 onwards his

includes a large number o small turbines which would not all within the scope o the ROC

scheme in the UK Hence or comparability with the UK data all turbines commissioned

prior to 1992 were dropped as well as all wind arms with a generating capacity o less than 1

MW he resulting sample covered 823 onshore wind arms and 30 offshore wind arms with

monthly load actors or various periods rom January 2002 to August 2012 Given the cut-off

or wind arms commissioned prior to 1992 the maximum age or an onshore wind arm was 20

years and the maximum age or an offshore wind arm was 17 years

he total capacity o the onshore wind arms recorded in the Danish database was 2757 MW

he average size o an onshore wind arm in Denmark is small ndash only 33 MW as compared

with 161 MW or the UK ndash because 84 o the sample consists o installations with between

1 and 4 turbines he largest onshore installation has 39 turbines with a capacity o 234 MW

while the onshore installation with the greatest capacity has 19 turbines with a capacity o 584

MW Overall the sample o Danish onshore wind arms is older and smaller than the sample

o UK onshore wind arms reflecting the much longer history o wind generation in Denmark

16 httpwwwensdkEN-USINFOFACSANDFIGURESENERGY_SAISICS_AND_INDICAORS

OVERVIEWOFHEENERGYSECORREGISEROFWINDURBINESSiderForsideaspx

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2852

26 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

he larger scale o UK wind arms may account or some o the differences in perormance

curves or the two countries

he sample o offshore wind arms in Denmark is quite small with 30 installations in

total Still it is useul to analyse the perormance o these installations since there is very little

comparable data or the UK he sample covers 394 turbines with an average o just over 13turbines per installation and an average capacity o 288 MW he largest installation which

was commissioned in 2003 has 72 turbines with a total capacity o 1656 MW Again offshore

installations are smaller and older in Denmark than is the current pattern in the UK he first

two wind arms in the sample were commissioned in the years 1995 and 2000 but the sample

increases significantly rom 2002 onwards o avoid reliance upon a sample o 1 or 2 wind

arms to determine specific coefficients or ages gt 10 years all observations with an age greater

than 10 were omitted rom the analysis17

A somewhat different problem with the offshore wind arm data arose because the coeffi-

cients estimated using capacity weights (see below) were heavily influenced by the largest windarm his has had an extremely erratic profile o perormance over time with a very low average

load actor he treatment o this wind arm is discussed in more detail in Section F below

C Specication and estimation methods

As Figure 8 illustrates it is necessary to go beyond simple summary statistics in order to assess

whether there is any systematic relationship between age and the average perormance o wind

arms he perormance o wind arms varies over both time and space some months and years

have more or less wind than the long term average while specific characteristics o wind arms

ndash including location type o turbine and operating regime ndash will influence the perormance oeach plant under identical wind conditions

he combination o (a) site-specific characteristics which are constant over time and (b)

period-specific characteristics which are constant across wind arms can be represented by

what statisticians call an error components model with fixed effects or each wind arm and

each time period he dependent variable in the model consists o a sequence o load actors ndash

denoted by or unit i in month t ndash or each wind arm (the panel unit) over time he load

actors are calculated by dividing the total output in MWh by (24 x number o days in month x

reported generating capacity in MW) and are multiplied by 100 to convert to percentages

he additive version o the error components model may be written as

(1)

in which denotes the age o plant i in period t and is some unction that is either a

representation o or an approximation to the normalised age-perormance relationship he

multiplicative version is similar except that is replaced by where ln() denotes the

natural logarithm he terms are wind arm or unit fixed effects which capture actors such

as location type o turbine operating regimes etc he terms are period fixed effects which

capture the influence o wind conditions over the UK or Denmark as a whole as well as seasonal

maintenance and demand Without loss o generality it may be assumed that the fixed effects are

17 For the avoidance o doubt this means that no attempt was made to estimate the effect o age on the perormance o

offshore wind arms beyond age 10 while the data or the oldest wind arms was included in the analysis up to and

including age 10

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 27

normalised so that ie all o the error components have a zero mean while the

mean o the random error is zero by definition he terms are random errors which capture

random variations in wind conditions that are not specific to the site turbine breakdowns and

other actors that are uncorrelated with either the site or the time period It is assumed that the

mean o the random error is zero while additional assumptions depend upon the method o

estimation that is adopted

he model specified in equation (1) can be estimated in a variety o ways18 A crucial issue is

whether the are or may be significantly correlated with the age terms or any other independent

regressors I it is assumed that there is no such correlation the model is usually reerred to as

a random effects model In that case the equation can be estimated by a variant o least squares

(LS) generalised least squares (GLS) generalised methods o moments (GMM) or maximum

likelihood (ML) ndash all o which should be yield parameter estimates that will converge in prob-

ability to the true parameter values as the sample size increases An alternative but less restric-

tive specification is known as the fixed effects model in which no assumption is made about

the correlation between the unit fixed effects and other regressors he limitation o the fixed

effects model is that it cannot be used to estimate parameters or regressors that vary across

panel units but are constant over time or a panel unit (time-invariant regressors) ndash eg the

year in which a wind arm was commissioned its location or its rated generating capacity he

influence o such variables is captured by the unit fixed effects

In order to avoid the risk o obtaining biased estimates o the age effects the results reported

in this note are based upon estimates using the fixed effects model he influence o time-invar-

iant regressors is examined in a second stage by regressing the estimates o the unit fixed effectson these variables he simplest method o estimating the fixed effects model is known as the

lsquowithinrsquo estimator in which the mean value or each unit is subtracted rom all observations or

that unit For example i is a general polynomial o order M in A equation (1) becomes

(2)

so that the model can be estimated by least square with the inclusion o dummy variables or

each period t he variance-covariance matrix o the coefficients have been estimated by using

(a) a robust sandwich estimator adjusted or clustering by panel unit and (b) a bootstrap esti-

mator with 400 repetitions he estimates o the standard errors will be consistent even i the

random error component is serially correlated within panel units andor has different variances

across panel units19 Since the number o panel units and the average number o time periods

per panel units are both large (other than or Danish offshore wind arms) the least square esti-

mates o the parameters will be unbiased under these assumptions

he bootstrap method is expensive to carry out but it provides a useul cross-check on

the standard errors generated by other methods Unortunately the bootstrap method o

18 See Chapter 21 o AC Cameron amp PK rivedi ndash Microeconometrics Methods and Applications (Cambridge

Cambridge University Press 2005) or a standard textbook treatment o estimation methods or panel data models

19 he estimation was carried out using the xtreg procedure in Stata Version 12 he methods o constructing the

sandwich and bootstrap estimates o the variance-covariance matrix are described in Chapters 8 amp 13 o AC

Cameron amp PK rivedi ndash Microeconometrics using Stata (Revised Edition College Station exas Stata Press 2010)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 18: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1852

16 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

effects while the red dashed line is based on the smoothed quadratic perormance curve and

the blue dashed line shows what the pattern would be i operating perormance was not affected

by age he calculation is extended out to age 25 since DECCrsquos standard comparisons assume

that wind turbines have a lie o 25 years

Figure 5 Impact of performance degradation on discounted cumulative output

0

10

20

30

40

50

60

70

80

90

100

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

o

f d i s c o u n t e d c u m u l a t i v e o u t p u t

Age of plant (years)

Constant Quadratic Full age effects

Source Authorrsquos estimates

21 he key points in the figure are that 80 o the discounted cumulative output o a new wind

plant is likely to be produced in the first 10 years o its lie and 90 in the first 14 years his

is consistent with the structure o Danish subsidies or onshore wind arms which extend over

a typical period o 125 years Since sites or wind arms are scarce and involve the payment o

significant rents that may be linked to output it is very unlikely that any new installation o

wind turbines will have an expected lie o more than 15 years Instead as has happened in the

past wind operators will have a strong incentive to decommission plants afer no more than 15

years and replace the turbines with newer equipment

22 As a consequence any economic assessment o wind generation should not be based on an

expected lie which is longer than 15 years In recent work reported in evidence to the House

o Commons Select Committee on Energy and Climate Change I assumed that wind plants

would have a residual value equal to 20 o their initial cost in real terms at the end o 15 years

he analysis in this paper suggests that this is too avourable an assumption Given the costs o

decommissioning old turbines the residual value is likely to be well below 10 o their initial

cost and the decision point may be at 10 rather than 15 years

23 he corollary o this observation is that it makes little sense to offer long term contracts or 20

years or more that guarantee prices or eed-in tariffs (Fis) to wind operators A contract length

o 10 to 12 years would be sufficient to remove most o the market risk associated with invest-

ment in wind generation In this respect the subsidy arrangements implemented in Denmark

are better designed

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 17

24 At the same time the offer o subsidies andor guaranteed prices may have serious adverse

consequences or the efficiency o wind generation Returning to the unit fixed effects measur-

ing the perormance o wind plants commissioned afer 2005 only 28 out o 159 units have an

operating perormance that exceeds the average or the period hose 28 units account or 360

MW o capacity whereas the remaining 131 units account or 2810 MW o capacity Not onlyare recent plants less efficient than the average or the whole period but the plants which are

below-average in efficiency are typically larger than the more efficient ones and account or a

disproportionate share o the recent additions to generating capacity

25 In mid-2012 there was approximately 4500 MW o onshore wind capacity operating in the

UK By 2020 it is expected that there will be at least 10000 MW (based on proposed wind

arms with planning consent) and perhaps as much 15000 MW o wind capacity in operation

he government is relying upon a substantial contribution rom onshore wind arms towards

meeting its targets or renewable energy in 2020 o provide context the governmentrsquos targets

or renewable energy in 2020 speciy a total o 234 Wh o energy rom all renewable sources

o which 90ndash140 Wh is expected to come rom wind and biomass electricity generation9

he projections or biomass electricity are extremely vague and have to be reconciled with

an assumed increase o 3ndash4 times in the amount o biomass used separately or heat So in

concrete terms it is reasonable to assume that wind generation will account or at least 90 Wh

with 24ndash32 Wh rom onshore generation and the remainder rom offshore

26 he average load actor or onshore wind over the 10 year period rom April 2002 to March

2012 was 256 which is influenced by the large amounts o new capacity added rom 2005

onwards I this load actor were achieved in 2020 the total amount o onshore wind capacityrequired to meet the onshore generation target would be 107 to 143 GW he average load

actor or offshore wind installations (on an unchanged configuration basis) or the five year

period 2007ndash11 was 32010 I this load actor were achieved in 2020 the total amount o

offshore wind capacity required to meet the remainder o the 90 Wh target would be 206

to 235 GW though these estimates would all to 180 to 205 GW i the average load actor

matched the historic average value o 367 or Danish offshore wind installations

27 he question that has to be addressed is whether these load actors are consistent with (a) the

investment programmes that would be required to achieve the total amounts o generation

capacity implied by the governmentrsquos targets and (b) the age profiles o the perormance o

wind arms discussed in this paper he short answer is that this is simply not possible i the

initial load actor or new wind arms continues to decline at the rate observed the past decade

So even as a starting point it is necessary to assume that this underlying deterioration in wind

arm perormance ceases his would be a major change in itsel and it is not obvious what

actors might lead to such a result

28 he results illustrated below are based on a vintage model o perormance that takes account

o the actual expansion o capacity up to the end o 2011 plus whatever investment in new or

replacement turbines is required to achieve the target levels o wind capacity in 2020 o high-light the ull range o potential outcomes these targets are either 10 GW or 15 GW o onshore

9 Department o Energy and Climate Change ndash UK Renewable Energy Road Map July 2011 Figure 2

10 he equivalent average load actor or onshore wind arms was 264

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2052

18 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

wind and either 18 or 24 GW o offshore wind he levels o investment required to meet these

targets depend upon the expected lie o wind turbines For onshore turbines an expected lie

o 10 years is consistent with re-powering decisions over the last decade while an expected lie

o 15 years seems to be the maximum consistent with the age-perormance profile For offshore

turbines the analysis uses expected lives o 15 and 20 years

Figure 6A Projected load factors for UK onshore wind (total capacity of 10 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

Figure 6B Projected load factors for UK onshore wind (total capacity of 15 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

29 Figure 6A shows projected values o the average normalised load actors or UK onshore wind

arms in 2015 and 2020 using a target capacity in 2020 equal to 10 GW plus the age-peror-

mance profiles or the UK and Denmark (DK) Since vintage effects are important in this

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 19

calculation the profile o new or replacement investment affects the results In this case the

level o gross investment is assumed to be constant rom 2013 onwards at either 930 or 1220

MW per year ndash the range is defined by the assumed lie o turbines Achieving a target o 15

GW in 2020 (Figure 6B) would require a level o investment rom 2013 onwards o either 1540

or 1930 MW per year Such figures are 3ndash4 times the average level o investment over theperiod 2005ndash11 so the lower target may provide a better indication o what will happen

30 All o the estimates are calibrated to match the observed average load actor in 2007ndash11 so that

they should be interpreted as relative rather than absolute values Using the age-perormance

profile or UK onshore wind installations the average normalised load actor is likely to all by

about 19 rom 2007ndash11 to 2015 and by 23ndash28 to 2020 I the age-perormance curves or

Danish onshore installations were to apply the reduction in the average normalised load actor

would be considerably less but there would still be a all o up to 5 by 2020

31 In the medium term the all is greater i turbines operate or 15 years because the average age

o all turbines will be higher and thus the degradation in perormance due to age will have a

larger effect By 2020 this effect reduces the average normalised load actor rom 203 or a 10

year lie to 195 or a 15 year lie here is a complex economic and commercial trade-off here

o meet a target o 10 GW in 2020 it is necessary to increase the level o investment in onshore

capacity by 30 rom 2013 Almost all o this will take the orm o re-powering existing wind

arms11 he costs are airly high because this will usually involve new development consents

decommissioning the original turbines and replacing them with a smaller number o larger and

more powerul turbines

32 Estimates o re-powering costs are generally treated as being commercially sensitive but broadproject costs cited in specialist magazines suggest that the unit costs are likely to be at least

pound1000 per kW or more than pound300 million per year in aggregate On the other hand the increase

in the amount o energy generated would be about 12 Wh or a 10 year turbine lie rather

than a 15 year turbine lie though this does not allow or the loss o production during re-pow-

ering Under the best circumstances the payback period may be quite short but the best option

rom a broader economic and social perspective is not obvious

33 Even under the best scenario 10 GW o onshore wind capacity will only generate 178 Wh o

electricity in 2020 well below the lower end o the target range or onshore wind generation

specified in DECCrsquos road map In act allowing or interactions between the age-perormance

profile and investment the amount o onshore wind capacity required in 2020 to generate 24

Wh would be a minimum o 134 GW At the top end o DECCrsquos target range the capacity

required in 2020 to generate 32 Wh would be a minimum o 178 GW Since these estimates

are based on a 10 year turbine lie the level o investment required rom 2013 onwards would

be 1650 MW per year or the bottom o the range and 2350 MW or the top end o the range

In the last 8 years total investment in onshore wind has only ever exceeded 600 MW in one

11 Re-powering is usually defined as the ull replacement o existing turbines whereas retrofitting covers the replace-

ment o various mechanical or electrical components In the data analysis re-powering involves the creation o a

new unit whereas retrofitting affects the perormance o a continuing unit Hence the potential benefits o retrofit-

ting are taken into account in constructing the age-perormance curves

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2252

20 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

year (883 MW in 2008) so it seems rather unlikely that even the lower end o DECCrsquos range or

onshore wind generation will actually be realised

34 O course this assessment would be different i UK onshore wind operators were able to match

the Danish age-perormance curve here are many actors which differentiate the UK and

Danish wind industries he average scale o wind arms in the UK is larger they are moreremote and there appears to be significantly greater variation between operating and potential

new sites in achieved perormance It would thereore require a very optimistic observer to

conclude that the difference in age-perormance curves between the two countries will disap-

pear in the short or medium term

Figure 7 Projected load factors for UK offshore wind in 2020

20

22

24

26

28

30

32

34

18 GW ndash life 15 yrs 18 GW ndash life 20 yrs 24 GW ndash life 15 yrs 24 GW ndash life 20 yrs

L o a d f a c t o r ( )

2007ndash11 2015 2020

Note Based on Danish age-perormance curves Source Authorrsquos estimates

35 Figure 7 shows the results rom carrying out a similar exercise or offshore wind arms in the

UK he only evidence on the age-perormance curves or offshore installations is based upon

the Danish experience so these have been used to construct the projections Further the profile

o investment assumes a progressive increase in additions to capacity rom 2013 to 2016 and

then a stable level o investment rom 2017 to 2020 he DECC road map involves a very large

commitment to offshore wind investment ndash in the range 25-35 GW o new capacity annually

rom 2016 onwards As a consequence the average age o offshore capacity in 2020 will be quite

low and the ull effects o the age-perormance curves will only be elt during the decade ollow-

ing 2020 Even so the average normalised load actor in 2020 will be about 15 lower than or

2007ndash11 while by 2025 it will be about 30 below the base level

36 On these projections 232 GW o offshore capacity is the minimum necessary to generate 58

Wh o electricity rom offshore wind in 2020 his requirement will increase to 298 GW by

2025 simply to offset the effect o the ageing o offshore generation capacity on total output

37 Combining the estimates or onshore and offshore wind generation the total amount o wind

capacity required in 2020 to produce 90 Wh o electricity rom wind will be in the range 39

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 21

to 41 GW once allowance is made or the observed age-perormance curves or both onshore

and offshore installations hese estimates are roughly a third higher than official predictions o

28ndash31 GW o wind capacity in 2020

38 urning the figures round gives a slightly different picture o the reality o the governmentrsquos

strategy on renewable electricity generation A total o 30 GW o wind capacity in 2020 (12GW onshore 18 GW offshore) will generate about 64 Wh o electricity in a normal year

Since electricity generation rom wind and biomass are expected to account or 50 o renew-

able energy in 2020 the consequence is that biomass is expected to account or 53 Wh In

2011 landfill gas and sewage sludge digestion accounted or 44 o electricity rom bioenergy

(DUKES ables 64) here are limited opportunities to expand production rom these sources

so a continuation o historic rates o growth would mean that renewable generation rom other

sources o bioenergy would have to increase rom 72 Wh in 2011 to 46 Wh in 2020 here

are severe constraints on increasing the amounts o electricity generated rom animal biomass

biodegradable municipal solid waste and anaerobic digestion so it would be optimistic to

assume that these will contribute more than 6-8 Wh in 2020 (up rom 26 Wh in 2011)

39 In practice i the projections or renewable energy are to be believed they rest upon an assump-

tion that there will be a large increase in the amount o electricity that is generated rom plant

biomass ndash either on its own or through co-firing with coal his increase will involve increas-

ing the amount o electricity generated in these ways rom 46 Wh in 2011 to 38-40 Wh in

2020 Since the amounts o straw and UK-grown timber available or this purpose are strictly

limited this will come down to the amount o wood chips that can be imported and the cost o

doing this40 he historic average load actor or biomass plants is just over 60 so about 7500 MW o

biomass capacity will be required to supply 40 Wh o electricity he weight o wood chips per

MWh o electricity depends upon moisture content storage boiler design and other actors but

an indicative range is 085 to 115 tonnes per MWh o wood chips with a moisture content o

30 According to DECC and Forestry Commission estimates the UK imported 265 million

tonnes o steam coal and 26 million tonnes o other wood products (including wood chips)

in 2011 It seems that the net effect o the governmentrsquos strategy will be to replace 15 million

tonnes o coal imported in 2011 by 32 to 44 million tonnes o imported wood chips his is a

somewhat strange way o increasing energy security

Conclusion

41 Wind power is a highly capital-intensive technology or generating electricity Its merits rely

entirely upon a substitution o capital or uel inputs he same is true or hydro or tidal or wave

power In comparison with hydro power wind is a low quality resource because o its variability

and because it cannot be stored So the economic case or wind power must rest on obtaining

the most out o the wind that is available

42 While the decline in the achieved perormance o onshore wind turbines in Denmark is

much less than that or the UK or offshore nonetheless the decline in expected output under

standardised wind conditions over 10 years is 10 unweighted and 13 capacity weighted

hese declines accelerate afer age 10 so that the reductions in perormance are 17 and 20

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2452

22 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

respectively afer 15 years For UK onshore wind arms the reduction in perormance due to

age is much worse at 27 unweighted and 69 capacity weighted by age 10

43 Evidence on the perormance o Danish offshore installations is both restricted and so poor

that there may be concern that the results are affected by a small number o outliers Still the

sample contains a reasonable number o sites with at least 5 years o operating experience andthe decline in perormance by age 5 is 38 unweighted and 26 capacity weighted

44 In addition to these results there is strong evidence that the average normalised load actor or

new onshore wind installations in the UK has allen significantly over the period rom 2000 to

2011 his is consistent with a pattern in which the most avourable sites are developed first

Equally it could mean that wind developers have been unable to keep up with the rate o new

investment while maintaining the quality o development and operations For example the site

design or selection o turbine characteristics may make less effective use o the available wind

resources or the sites available than was the case in the past

45 Whatever the reasons the deterioration in initial perormance means that the expected returns

rom the expansion in wind capacity both or investors and in terms o the reduction in CO983090

emissions have been alling without a concomitant decrease in the private and social costs

that are borne by customers and the general public Clearly this is unsatisactory at best and it

suggests that the benefits claimed or current policies cannot be taken at ace value

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 23

Appendix Data and Methods

A Data for the United Kingdom

he raw data used or this study was extracted by the Renewable Energy Foundation rom the

Renewables and CHP Register database compiled by Ogem12 his inormation is used in the

administration o the market in Renewable Obligation Certificates (ROCs) he Renewables

Obligation is the primary mechanism by which large scale generators o renewable electric-

ity receive subsidies so the operators o wind arms have a very strong incentive to submit

complete and up-to-date inormation to the database he data extracted covered all onshore

wind generators with at least 05 MW o generating capacity which are eligible or ROCs or the

10 year period April 2002 to March 2012 he key variables are the monthly output o electric-

ity (used or the purpose o allocating ROCs) and nominal generating capacity here are 282

reporting units in the final dataset A reporting unit may correspond to an entire wind arm orto separate phases o development Complications arise when an existing wind arm is re-pow-

ered ie when old wind turbines are replaced by newer and usually more powerul turbines

he database reports this as a change in the generating capacity o the recording unit but in

this analysis repowering is treated as the termination o the old recording unit and the creation

o a new recording unit

A considerable amount o data checking and cleaning was required beore the data could

be analysed he first step was to add the age o each reporting unit his was calculated by

reerence to the month in which the turbines were commissioned For wind arms in operation

in or beore June 2002 the commissioning date was obtained rom various online sources ndash

notably the thewindpowernet database o UK wind arms13 which was cross-checked against

the inormation provided on the websites o wind operators and other sources For recording

units whose first data relates to dates afer June 2002 the commissioning date was initially

assumed to be the month or which data is first reported Cross-checks were carried out or

all recording units which commenced reporting in the period July 2002 to December 2003 In

more than 80 o cases the externally-reported commissioning date is within one month o the

first month or which data is reported he exceptions ndash Bu Farm and Mablethorpe ndash appear to

be due to delays by small operators in reporting data to Ogem

In a number o cases it is clear rom the data that plants were not in ull operation at the

time when data was first reported to the database here are two indicators o partial or delayed

commissioning

bull For an initial series o months the calculated load actor is very low ndash usually below 5 ndash

and then increases abruptly In such cases the data or the early months is dropped and the

commissioning date is treated as the month immediately preceding the first month with a

lsquonormalrsquo load actor14

12 httpswwwrenewablesandchpogemgovuk

13 See httpwwwthewindpowernetwindarms_list_enphp

14 he month in which a plant is reported as having been commissioned is not included in the analysis

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2652

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 25

the decline in the plantrsquos load actor as it ages As an example consider a plant whose reported

capacity was increased rom 99 MW to 10 MW at age 3 I this was simply a change in report-

ing practice this would reduce the estimated load actor rom age 3 onwards By increasing

the initial capacity rom 99 MW to 10 MW ndash or by reducing the capacity rom 10 MW to 99

MW rom age 3 onwards ndash this bias is removed It is not important which adjustment is madebecause the difference between the alternative assumptions is captured by the unit fixed effect

he numbers o separate sites and o observations per site mean that it is not possible to

provide a simple visual representation o the data Figure 8 provides a summary by illustrating

the distributions o load actors by the age o the wind arm pooled across all years he ranges

o the load actors or each age are wide while the medians show no clear trend by age hus a

naiumlve analysis might conclude that there is nothing to investigate However as will be explained

below the distributions mask crucial differences between the perormance o wind arms as

they age because they do not control or the differences in location and wind availability over

time

B Data for Denmark

he data or Danish wind arms used in this study comes rom a database compiled by the

Danish Energy Agency covering the characteristics and perormance o all wind turbines rom

2002 up to the end o August 201216 he basic recording unit or the register is the individual

wind turbine but where a number o turbines are connected to a single meter an identical

average output per turbine is recorded or all turbines in the group he presence o what are in

effect repeated observations increases the noise in the data In addition the model that is used

or the analysis assumes that site operator and turbine characteristics are common or all o theunits that make up a single wind arm o address this point separate wind arms have been

identified by grouping all turbines which have identical values or turbine manuacturer and

type local authority date o commissioning and date o decommissioning (where applicable)

he register contains inormation on 7569 turbines commissioned rom 1977 onwards his

includes a large number o small turbines which would not all within the scope o the ROC

scheme in the UK Hence or comparability with the UK data all turbines commissioned

prior to 1992 were dropped as well as all wind arms with a generating capacity o less than 1

MW he resulting sample covered 823 onshore wind arms and 30 offshore wind arms with

monthly load actors or various periods rom January 2002 to August 2012 Given the cut-off

or wind arms commissioned prior to 1992 the maximum age or an onshore wind arm was 20

years and the maximum age or an offshore wind arm was 17 years

he total capacity o the onshore wind arms recorded in the Danish database was 2757 MW

he average size o an onshore wind arm in Denmark is small ndash only 33 MW as compared

with 161 MW or the UK ndash because 84 o the sample consists o installations with between

1 and 4 turbines he largest onshore installation has 39 turbines with a capacity o 234 MW

while the onshore installation with the greatest capacity has 19 turbines with a capacity o 584

MW Overall the sample o Danish onshore wind arms is older and smaller than the sample

o UK onshore wind arms reflecting the much longer history o wind generation in Denmark

16 httpwwwensdkEN-USINFOFACSANDFIGURESENERGY_SAISICS_AND_INDICAORS

OVERVIEWOFHEENERGYSECORREGISEROFWINDURBINESSiderForsideaspx

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2852

26 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

he larger scale o UK wind arms may account or some o the differences in perormance

curves or the two countries

he sample o offshore wind arms in Denmark is quite small with 30 installations in

total Still it is useul to analyse the perormance o these installations since there is very little

comparable data or the UK he sample covers 394 turbines with an average o just over 13turbines per installation and an average capacity o 288 MW he largest installation which

was commissioned in 2003 has 72 turbines with a total capacity o 1656 MW Again offshore

installations are smaller and older in Denmark than is the current pattern in the UK he first

two wind arms in the sample were commissioned in the years 1995 and 2000 but the sample

increases significantly rom 2002 onwards o avoid reliance upon a sample o 1 or 2 wind

arms to determine specific coefficients or ages gt 10 years all observations with an age greater

than 10 were omitted rom the analysis17

A somewhat different problem with the offshore wind arm data arose because the coeffi-

cients estimated using capacity weights (see below) were heavily influenced by the largest windarm his has had an extremely erratic profile o perormance over time with a very low average

load actor he treatment o this wind arm is discussed in more detail in Section F below

C Specication and estimation methods

As Figure 8 illustrates it is necessary to go beyond simple summary statistics in order to assess

whether there is any systematic relationship between age and the average perormance o wind

arms he perormance o wind arms varies over both time and space some months and years

have more or less wind than the long term average while specific characteristics o wind arms

ndash including location type o turbine and operating regime ndash will influence the perormance oeach plant under identical wind conditions

he combination o (a) site-specific characteristics which are constant over time and (b)

period-specific characteristics which are constant across wind arms can be represented by

what statisticians call an error components model with fixed effects or each wind arm and

each time period he dependent variable in the model consists o a sequence o load actors ndash

denoted by or unit i in month t ndash or each wind arm (the panel unit) over time he load

actors are calculated by dividing the total output in MWh by (24 x number o days in month x

reported generating capacity in MW) and are multiplied by 100 to convert to percentages

he additive version o the error components model may be written as

(1)

in which denotes the age o plant i in period t and is some unction that is either a

representation o or an approximation to the normalised age-perormance relationship he

multiplicative version is similar except that is replaced by where ln() denotes the

natural logarithm he terms are wind arm or unit fixed effects which capture actors such

as location type o turbine operating regimes etc he terms are period fixed effects which

capture the influence o wind conditions over the UK or Denmark as a whole as well as seasonal

maintenance and demand Without loss o generality it may be assumed that the fixed effects are

17 For the avoidance o doubt this means that no attempt was made to estimate the effect o age on the perormance o

offshore wind arms beyond age 10 while the data or the oldest wind arms was included in the analysis up to and

including age 10

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 27

normalised so that ie all o the error components have a zero mean while the

mean o the random error is zero by definition he terms are random errors which capture

random variations in wind conditions that are not specific to the site turbine breakdowns and

other actors that are uncorrelated with either the site or the time period It is assumed that the

mean o the random error is zero while additional assumptions depend upon the method o

estimation that is adopted

he model specified in equation (1) can be estimated in a variety o ways18 A crucial issue is

whether the are or may be significantly correlated with the age terms or any other independent

regressors I it is assumed that there is no such correlation the model is usually reerred to as

a random effects model In that case the equation can be estimated by a variant o least squares

(LS) generalised least squares (GLS) generalised methods o moments (GMM) or maximum

likelihood (ML) ndash all o which should be yield parameter estimates that will converge in prob-

ability to the true parameter values as the sample size increases An alternative but less restric-

tive specification is known as the fixed effects model in which no assumption is made about

the correlation between the unit fixed effects and other regressors he limitation o the fixed

effects model is that it cannot be used to estimate parameters or regressors that vary across

panel units but are constant over time or a panel unit (time-invariant regressors) ndash eg the

year in which a wind arm was commissioned its location or its rated generating capacity he

influence o such variables is captured by the unit fixed effects

In order to avoid the risk o obtaining biased estimates o the age effects the results reported

in this note are based upon estimates using the fixed effects model he influence o time-invar-

iant regressors is examined in a second stage by regressing the estimates o the unit fixed effectson these variables he simplest method o estimating the fixed effects model is known as the

lsquowithinrsquo estimator in which the mean value or each unit is subtracted rom all observations or

that unit For example i is a general polynomial o order M in A equation (1) becomes

(2)

so that the model can be estimated by least square with the inclusion o dummy variables or

each period t he variance-covariance matrix o the coefficients have been estimated by using

(a) a robust sandwich estimator adjusted or clustering by panel unit and (b) a bootstrap esti-

mator with 400 repetitions he estimates o the standard errors will be consistent even i the

random error component is serially correlated within panel units andor has different variances

across panel units19 Since the number o panel units and the average number o time periods

per panel units are both large (other than or Danish offshore wind arms) the least square esti-

mates o the parameters will be unbiased under these assumptions

he bootstrap method is expensive to carry out but it provides a useul cross-check on

the standard errors generated by other methods Unortunately the bootstrap method o

18 See Chapter 21 o AC Cameron amp PK rivedi ndash Microeconometrics Methods and Applications (Cambridge

Cambridge University Press 2005) or a standard textbook treatment o estimation methods or panel data models

19 he estimation was carried out using the xtreg procedure in Stata Version 12 he methods o constructing the

sandwich and bootstrap estimates o the variance-covariance matrix are described in Chapters 8 amp 13 o AC

Cameron amp PK rivedi ndash Microeconometrics using Stata (Revised Edition College Station exas Stata Press 2010)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 19: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 1952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 17

24 At the same time the offer o subsidies andor guaranteed prices may have serious adverse

consequences or the efficiency o wind generation Returning to the unit fixed effects measur-

ing the perormance o wind plants commissioned afer 2005 only 28 out o 159 units have an

operating perormance that exceeds the average or the period hose 28 units account or 360

MW o capacity whereas the remaining 131 units account or 2810 MW o capacity Not onlyare recent plants less efficient than the average or the whole period but the plants which are

below-average in efficiency are typically larger than the more efficient ones and account or a

disproportionate share o the recent additions to generating capacity

25 In mid-2012 there was approximately 4500 MW o onshore wind capacity operating in the

UK By 2020 it is expected that there will be at least 10000 MW (based on proposed wind

arms with planning consent) and perhaps as much 15000 MW o wind capacity in operation

he government is relying upon a substantial contribution rom onshore wind arms towards

meeting its targets or renewable energy in 2020 o provide context the governmentrsquos targets

or renewable energy in 2020 speciy a total o 234 Wh o energy rom all renewable sources

o which 90ndash140 Wh is expected to come rom wind and biomass electricity generation9

he projections or biomass electricity are extremely vague and have to be reconciled with

an assumed increase o 3ndash4 times in the amount o biomass used separately or heat So in

concrete terms it is reasonable to assume that wind generation will account or at least 90 Wh

with 24ndash32 Wh rom onshore generation and the remainder rom offshore

26 he average load actor or onshore wind over the 10 year period rom April 2002 to March

2012 was 256 which is influenced by the large amounts o new capacity added rom 2005

onwards I this load actor were achieved in 2020 the total amount o onshore wind capacityrequired to meet the onshore generation target would be 107 to 143 GW he average load

actor or offshore wind installations (on an unchanged configuration basis) or the five year

period 2007ndash11 was 32010 I this load actor were achieved in 2020 the total amount o

offshore wind capacity required to meet the remainder o the 90 Wh target would be 206

to 235 GW though these estimates would all to 180 to 205 GW i the average load actor

matched the historic average value o 367 or Danish offshore wind installations

27 he question that has to be addressed is whether these load actors are consistent with (a) the

investment programmes that would be required to achieve the total amounts o generation

capacity implied by the governmentrsquos targets and (b) the age profiles o the perormance o

wind arms discussed in this paper he short answer is that this is simply not possible i the

initial load actor or new wind arms continues to decline at the rate observed the past decade

So even as a starting point it is necessary to assume that this underlying deterioration in wind

arm perormance ceases his would be a major change in itsel and it is not obvious what

actors might lead to such a result

28 he results illustrated below are based on a vintage model o perormance that takes account

o the actual expansion o capacity up to the end o 2011 plus whatever investment in new or

replacement turbines is required to achieve the target levels o wind capacity in 2020 o high-light the ull range o potential outcomes these targets are either 10 GW or 15 GW o onshore

9 Department o Energy and Climate Change ndash UK Renewable Energy Road Map July 2011 Figure 2

10 he equivalent average load actor or onshore wind arms was 264

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2052

18 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

wind and either 18 or 24 GW o offshore wind he levels o investment required to meet these

targets depend upon the expected lie o wind turbines For onshore turbines an expected lie

o 10 years is consistent with re-powering decisions over the last decade while an expected lie

o 15 years seems to be the maximum consistent with the age-perormance profile For offshore

turbines the analysis uses expected lives o 15 and 20 years

Figure 6A Projected load factors for UK onshore wind (total capacity of 10 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

Figure 6B Projected load factors for UK onshore wind (total capacity of 15 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

29 Figure 6A shows projected values o the average normalised load actors or UK onshore wind

arms in 2015 and 2020 using a target capacity in 2020 equal to 10 GW plus the age-peror-

mance profiles or the UK and Denmark (DK) Since vintage effects are important in this

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 19

calculation the profile o new or replacement investment affects the results In this case the

level o gross investment is assumed to be constant rom 2013 onwards at either 930 or 1220

MW per year ndash the range is defined by the assumed lie o turbines Achieving a target o 15

GW in 2020 (Figure 6B) would require a level o investment rom 2013 onwards o either 1540

or 1930 MW per year Such figures are 3ndash4 times the average level o investment over theperiod 2005ndash11 so the lower target may provide a better indication o what will happen

30 All o the estimates are calibrated to match the observed average load actor in 2007ndash11 so that

they should be interpreted as relative rather than absolute values Using the age-perormance

profile or UK onshore wind installations the average normalised load actor is likely to all by

about 19 rom 2007ndash11 to 2015 and by 23ndash28 to 2020 I the age-perormance curves or

Danish onshore installations were to apply the reduction in the average normalised load actor

would be considerably less but there would still be a all o up to 5 by 2020

31 In the medium term the all is greater i turbines operate or 15 years because the average age

o all turbines will be higher and thus the degradation in perormance due to age will have a

larger effect By 2020 this effect reduces the average normalised load actor rom 203 or a 10

year lie to 195 or a 15 year lie here is a complex economic and commercial trade-off here

o meet a target o 10 GW in 2020 it is necessary to increase the level o investment in onshore

capacity by 30 rom 2013 Almost all o this will take the orm o re-powering existing wind

arms11 he costs are airly high because this will usually involve new development consents

decommissioning the original turbines and replacing them with a smaller number o larger and

more powerul turbines

32 Estimates o re-powering costs are generally treated as being commercially sensitive but broadproject costs cited in specialist magazines suggest that the unit costs are likely to be at least

pound1000 per kW or more than pound300 million per year in aggregate On the other hand the increase

in the amount o energy generated would be about 12 Wh or a 10 year turbine lie rather

than a 15 year turbine lie though this does not allow or the loss o production during re-pow-

ering Under the best circumstances the payback period may be quite short but the best option

rom a broader economic and social perspective is not obvious

33 Even under the best scenario 10 GW o onshore wind capacity will only generate 178 Wh o

electricity in 2020 well below the lower end o the target range or onshore wind generation

specified in DECCrsquos road map In act allowing or interactions between the age-perormance

profile and investment the amount o onshore wind capacity required in 2020 to generate 24

Wh would be a minimum o 134 GW At the top end o DECCrsquos target range the capacity

required in 2020 to generate 32 Wh would be a minimum o 178 GW Since these estimates

are based on a 10 year turbine lie the level o investment required rom 2013 onwards would

be 1650 MW per year or the bottom o the range and 2350 MW or the top end o the range

In the last 8 years total investment in onshore wind has only ever exceeded 600 MW in one

11 Re-powering is usually defined as the ull replacement o existing turbines whereas retrofitting covers the replace-

ment o various mechanical or electrical components In the data analysis re-powering involves the creation o a

new unit whereas retrofitting affects the perormance o a continuing unit Hence the potential benefits o retrofit-

ting are taken into account in constructing the age-perormance curves

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2252

20 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

year (883 MW in 2008) so it seems rather unlikely that even the lower end o DECCrsquos range or

onshore wind generation will actually be realised

34 O course this assessment would be different i UK onshore wind operators were able to match

the Danish age-perormance curve here are many actors which differentiate the UK and

Danish wind industries he average scale o wind arms in the UK is larger they are moreremote and there appears to be significantly greater variation between operating and potential

new sites in achieved perormance It would thereore require a very optimistic observer to

conclude that the difference in age-perormance curves between the two countries will disap-

pear in the short or medium term

Figure 7 Projected load factors for UK offshore wind in 2020

20

22

24

26

28

30

32

34

18 GW ndash life 15 yrs 18 GW ndash life 20 yrs 24 GW ndash life 15 yrs 24 GW ndash life 20 yrs

L o a d f a c t o r ( )

2007ndash11 2015 2020

Note Based on Danish age-perormance curves Source Authorrsquos estimates

35 Figure 7 shows the results rom carrying out a similar exercise or offshore wind arms in the

UK he only evidence on the age-perormance curves or offshore installations is based upon

the Danish experience so these have been used to construct the projections Further the profile

o investment assumes a progressive increase in additions to capacity rom 2013 to 2016 and

then a stable level o investment rom 2017 to 2020 he DECC road map involves a very large

commitment to offshore wind investment ndash in the range 25-35 GW o new capacity annually

rom 2016 onwards As a consequence the average age o offshore capacity in 2020 will be quite

low and the ull effects o the age-perormance curves will only be elt during the decade ollow-

ing 2020 Even so the average normalised load actor in 2020 will be about 15 lower than or

2007ndash11 while by 2025 it will be about 30 below the base level

36 On these projections 232 GW o offshore capacity is the minimum necessary to generate 58

Wh o electricity rom offshore wind in 2020 his requirement will increase to 298 GW by

2025 simply to offset the effect o the ageing o offshore generation capacity on total output

37 Combining the estimates or onshore and offshore wind generation the total amount o wind

capacity required in 2020 to produce 90 Wh o electricity rom wind will be in the range 39

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 21

to 41 GW once allowance is made or the observed age-perormance curves or both onshore

and offshore installations hese estimates are roughly a third higher than official predictions o

28ndash31 GW o wind capacity in 2020

38 urning the figures round gives a slightly different picture o the reality o the governmentrsquos

strategy on renewable electricity generation A total o 30 GW o wind capacity in 2020 (12GW onshore 18 GW offshore) will generate about 64 Wh o electricity in a normal year

Since electricity generation rom wind and biomass are expected to account or 50 o renew-

able energy in 2020 the consequence is that biomass is expected to account or 53 Wh In

2011 landfill gas and sewage sludge digestion accounted or 44 o electricity rom bioenergy

(DUKES ables 64) here are limited opportunities to expand production rom these sources

so a continuation o historic rates o growth would mean that renewable generation rom other

sources o bioenergy would have to increase rom 72 Wh in 2011 to 46 Wh in 2020 here

are severe constraints on increasing the amounts o electricity generated rom animal biomass

biodegradable municipal solid waste and anaerobic digestion so it would be optimistic to

assume that these will contribute more than 6-8 Wh in 2020 (up rom 26 Wh in 2011)

39 In practice i the projections or renewable energy are to be believed they rest upon an assump-

tion that there will be a large increase in the amount o electricity that is generated rom plant

biomass ndash either on its own or through co-firing with coal his increase will involve increas-

ing the amount o electricity generated in these ways rom 46 Wh in 2011 to 38-40 Wh in

2020 Since the amounts o straw and UK-grown timber available or this purpose are strictly

limited this will come down to the amount o wood chips that can be imported and the cost o

doing this40 he historic average load actor or biomass plants is just over 60 so about 7500 MW o

biomass capacity will be required to supply 40 Wh o electricity he weight o wood chips per

MWh o electricity depends upon moisture content storage boiler design and other actors but

an indicative range is 085 to 115 tonnes per MWh o wood chips with a moisture content o

30 According to DECC and Forestry Commission estimates the UK imported 265 million

tonnes o steam coal and 26 million tonnes o other wood products (including wood chips)

in 2011 It seems that the net effect o the governmentrsquos strategy will be to replace 15 million

tonnes o coal imported in 2011 by 32 to 44 million tonnes o imported wood chips his is a

somewhat strange way o increasing energy security

Conclusion

41 Wind power is a highly capital-intensive technology or generating electricity Its merits rely

entirely upon a substitution o capital or uel inputs he same is true or hydro or tidal or wave

power In comparison with hydro power wind is a low quality resource because o its variability

and because it cannot be stored So the economic case or wind power must rest on obtaining

the most out o the wind that is available

42 While the decline in the achieved perormance o onshore wind turbines in Denmark is

much less than that or the UK or offshore nonetheless the decline in expected output under

standardised wind conditions over 10 years is 10 unweighted and 13 capacity weighted

hese declines accelerate afer age 10 so that the reductions in perormance are 17 and 20

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2452

22 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

respectively afer 15 years For UK onshore wind arms the reduction in perormance due to

age is much worse at 27 unweighted and 69 capacity weighted by age 10

43 Evidence on the perormance o Danish offshore installations is both restricted and so poor

that there may be concern that the results are affected by a small number o outliers Still the

sample contains a reasonable number o sites with at least 5 years o operating experience andthe decline in perormance by age 5 is 38 unweighted and 26 capacity weighted

44 In addition to these results there is strong evidence that the average normalised load actor or

new onshore wind installations in the UK has allen significantly over the period rom 2000 to

2011 his is consistent with a pattern in which the most avourable sites are developed first

Equally it could mean that wind developers have been unable to keep up with the rate o new

investment while maintaining the quality o development and operations For example the site

design or selection o turbine characteristics may make less effective use o the available wind

resources or the sites available than was the case in the past

45 Whatever the reasons the deterioration in initial perormance means that the expected returns

rom the expansion in wind capacity both or investors and in terms o the reduction in CO983090

emissions have been alling without a concomitant decrease in the private and social costs

that are borne by customers and the general public Clearly this is unsatisactory at best and it

suggests that the benefits claimed or current policies cannot be taken at ace value

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 23

Appendix Data and Methods

A Data for the United Kingdom

he raw data used or this study was extracted by the Renewable Energy Foundation rom the

Renewables and CHP Register database compiled by Ogem12 his inormation is used in the

administration o the market in Renewable Obligation Certificates (ROCs) he Renewables

Obligation is the primary mechanism by which large scale generators o renewable electric-

ity receive subsidies so the operators o wind arms have a very strong incentive to submit

complete and up-to-date inormation to the database he data extracted covered all onshore

wind generators with at least 05 MW o generating capacity which are eligible or ROCs or the

10 year period April 2002 to March 2012 he key variables are the monthly output o electric-

ity (used or the purpose o allocating ROCs) and nominal generating capacity here are 282

reporting units in the final dataset A reporting unit may correspond to an entire wind arm orto separate phases o development Complications arise when an existing wind arm is re-pow-

ered ie when old wind turbines are replaced by newer and usually more powerul turbines

he database reports this as a change in the generating capacity o the recording unit but in

this analysis repowering is treated as the termination o the old recording unit and the creation

o a new recording unit

A considerable amount o data checking and cleaning was required beore the data could

be analysed he first step was to add the age o each reporting unit his was calculated by

reerence to the month in which the turbines were commissioned For wind arms in operation

in or beore June 2002 the commissioning date was obtained rom various online sources ndash

notably the thewindpowernet database o UK wind arms13 which was cross-checked against

the inormation provided on the websites o wind operators and other sources For recording

units whose first data relates to dates afer June 2002 the commissioning date was initially

assumed to be the month or which data is first reported Cross-checks were carried out or

all recording units which commenced reporting in the period July 2002 to December 2003 In

more than 80 o cases the externally-reported commissioning date is within one month o the

first month or which data is reported he exceptions ndash Bu Farm and Mablethorpe ndash appear to

be due to delays by small operators in reporting data to Ogem

In a number o cases it is clear rom the data that plants were not in ull operation at the

time when data was first reported to the database here are two indicators o partial or delayed

commissioning

bull For an initial series o months the calculated load actor is very low ndash usually below 5 ndash

and then increases abruptly In such cases the data or the early months is dropped and the

commissioning date is treated as the month immediately preceding the first month with a

lsquonormalrsquo load actor14

12 httpswwwrenewablesandchpogemgovuk

13 See httpwwwthewindpowernetwindarms_list_enphp

14 he month in which a plant is reported as having been commissioned is not included in the analysis

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2652

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 25

the decline in the plantrsquos load actor as it ages As an example consider a plant whose reported

capacity was increased rom 99 MW to 10 MW at age 3 I this was simply a change in report-

ing practice this would reduce the estimated load actor rom age 3 onwards By increasing

the initial capacity rom 99 MW to 10 MW ndash or by reducing the capacity rom 10 MW to 99

MW rom age 3 onwards ndash this bias is removed It is not important which adjustment is madebecause the difference between the alternative assumptions is captured by the unit fixed effect

he numbers o separate sites and o observations per site mean that it is not possible to

provide a simple visual representation o the data Figure 8 provides a summary by illustrating

the distributions o load actors by the age o the wind arm pooled across all years he ranges

o the load actors or each age are wide while the medians show no clear trend by age hus a

naiumlve analysis might conclude that there is nothing to investigate However as will be explained

below the distributions mask crucial differences between the perormance o wind arms as

they age because they do not control or the differences in location and wind availability over

time

B Data for Denmark

he data or Danish wind arms used in this study comes rom a database compiled by the

Danish Energy Agency covering the characteristics and perormance o all wind turbines rom

2002 up to the end o August 201216 he basic recording unit or the register is the individual

wind turbine but where a number o turbines are connected to a single meter an identical

average output per turbine is recorded or all turbines in the group he presence o what are in

effect repeated observations increases the noise in the data In addition the model that is used

or the analysis assumes that site operator and turbine characteristics are common or all o theunits that make up a single wind arm o address this point separate wind arms have been

identified by grouping all turbines which have identical values or turbine manuacturer and

type local authority date o commissioning and date o decommissioning (where applicable)

he register contains inormation on 7569 turbines commissioned rom 1977 onwards his

includes a large number o small turbines which would not all within the scope o the ROC

scheme in the UK Hence or comparability with the UK data all turbines commissioned

prior to 1992 were dropped as well as all wind arms with a generating capacity o less than 1

MW he resulting sample covered 823 onshore wind arms and 30 offshore wind arms with

monthly load actors or various periods rom January 2002 to August 2012 Given the cut-off

or wind arms commissioned prior to 1992 the maximum age or an onshore wind arm was 20

years and the maximum age or an offshore wind arm was 17 years

he total capacity o the onshore wind arms recorded in the Danish database was 2757 MW

he average size o an onshore wind arm in Denmark is small ndash only 33 MW as compared

with 161 MW or the UK ndash because 84 o the sample consists o installations with between

1 and 4 turbines he largest onshore installation has 39 turbines with a capacity o 234 MW

while the onshore installation with the greatest capacity has 19 turbines with a capacity o 584

MW Overall the sample o Danish onshore wind arms is older and smaller than the sample

o UK onshore wind arms reflecting the much longer history o wind generation in Denmark

16 httpwwwensdkEN-USINFOFACSANDFIGURESENERGY_SAISICS_AND_INDICAORS

OVERVIEWOFHEENERGYSECORREGISEROFWINDURBINESSiderForsideaspx

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2852

26 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

he larger scale o UK wind arms may account or some o the differences in perormance

curves or the two countries

he sample o offshore wind arms in Denmark is quite small with 30 installations in

total Still it is useul to analyse the perormance o these installations since there is very little

comparable data or the UK he sample covers 394 turbines with an average o just over 13turbines per installation and an average capacity o 288 MW he largest installation which

was commissioned in 2003 has 72 turbines with a total capacity o 1656 MW Again offshore

installations are smaller and older in Denmark than is the current pattern in the UK he first

two wind arms in the sample were commissioned in the years 1995 and 2000 but the sample

increases significantly rom 2002 onwards o avoid reliance upon a sample o 1 or 2 wind

arms to determine specific coefficients or ages gt 10 years all observations with an age greater

than 10 were omitted rom the analysis17

A somewhat different problem with the offshore wind arm data arose because the coeffi-

cients estimated using capacity weights (see below) were heavily influenced by the largest windarm his has had an extremely erratic profile o perormance over time with a very low average

load actor he treatment o this wind arm is discussed in more detail in Section F below

C Specication and estimation methods

As Figure 8 illustrates it is necessary to go beyond simple summary statistics in order to assess

whether there is any systematic relationship between age and the average perormance o wind

arms he perormance o wind arms varies over both time and space some months and years

have more or less wind than the long term average while specific characteristics o wind arms

ndash including location type o turbine and operating regime ndash will influence the perormance oeach plant under identical wind conditions

he combination o (a) site-specific characteristics which are constant over time and (b)

period-specific characteristics which are constant across wind arms can be represented by

what statisticians call an error components model with fixed effects or each wind arm and

each time period he dependent variable in the model consists o a sequence o load actors ndash

denoted by or unit i in month t ndash or each wind arm (the panel unit) over time he load

actors are calculated by dividing the total output in MWh by (24 x number o days in month x

reported generating capacity in MW) and are multiplied by 100 to convert to percentages

he additive version o the error components model may be written as

(1)

in which denotes the age o plant i in period t and is some unction that is either a

representation o or an approximation to the normalised age-perormance relationship he

multiplicative version is similar except that is replaced by where ln() denotes the

natural logarithm he terms are wind arm or unit fixed effects which capture actors such

as location type o turbine operating regimes etc he terms are period fixed effects which

capture the influence o wind conditions over the UK or Denmark as a whole as well as seasonal

maintenance and demand Without loss o generality it may be assumed that the fixed effects are

17 For the avoidance o doubt this means that no attempt was made to estimate the effect o age on the perormance o

offshore wind arms beyond age 10 while the data or the oldest wind arms was included in the analysis up to and

including age 10

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 27

normalised so that ie all o the error components have a zero mean while the

mean o the random error is zero by definition he terms are random errors which capture

random variations in wind conditions that are not specific to the site turbine breakdowns and

other actors that are uncorrelated with either the site or the time period It is assumed that the

mean o the random error is zero while additional assumptions depend upon the method o

estimation that is adopted

he model specified in equation (1) can be estimated in a variety o ways18 A crucial issue is

whether the are or may be significantly correlated with the age terms or any other independent

regressors I it is assumed that there is no such correlation the model is usually reerred to as

a random effects model In that case the equation can be estimated by a variant o least squares

(LS) generalised least squares (GLS) generalised methods o moments (GMM) or maximum

likelihood (ML) ndash all o which should be yield parameter estimates that will converge in prob-

ability to the true parameter values as the sample size increases An alternative but less restric-

tive specification is known as the fixed effects model in which no assumption is made about

the correlation between the unit fixed effects and other regressors he limitation o the fixed

effects model is that it cannot be used to estimate parameters or regressors that vary across

panel units but are constant over time or a panel unit (time-invariant regressors) ndash eg the

year in which a wind arm was commissioned its location or its rated generating capacity he

influence o such variables is captured by the unit fixed effects

In order to avoid the risk o obtaining biased estimates o the age effects the results reported

in this note are based upon estimates using the fixed effects model he influence o time-invar-

iant regressors is examined in a second stage by regressing the estimates o the unit fixed effectson these variables he simplest method o estimating the fixed effects model is known as the

lsquowithinrsquo estimator in which the mean value or each unit is subtracted rom all observations or

that unit For example i is a general polynomial o order M in A equation (1) becomes

(2)

so that the model can be estimated by least square with the inclusion o dummy variables or

each period t he variance-covariance matrix o the coefficients have been estimated by using

(a) a robust sandwich estimator adjusted or clustering by panel unit and (b) a bootstrap esti-

mator with 400 repetitions he estimates o the standard errors will be consistent even i the

random error component is serially correlated within panel units andor has different variances

across panel units19 Since the number o panel units and the average number o time periods

per panel units are both large (other than or Danish offshore wind arms) the least square esti-

mates o the parameters will be unbiased under these assumptions

he bootstrap method is expensive to carry out but it provides a useul cross-check on

the standard errors generated by other methods Unortunately the bootstrap method o

18 See Chapter 21 o AC Cameron amp PK rivedi ndash Microeconometrics Methods and Applications (Cambridge

Cambridge University Press 2005) or a standard textbook treatment o estimation methods or panel data models

19 he estimation was carried out using the xtreg procedure in Stata Version 12 he methods o constructing the

sandwich and bootstrap estimates o the variance-covariance matrix are described in Chapters 8 amp 13 o AC

Cameron amp PK rivedi ndash Microeconometrics using Stata (Revised Edition College Station exas Stata Press 2010)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 20: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2052

18 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

wind and either 18 or 24 GW o offshore wind he levels o investment required to meet these

targets depend upon the expected lie o wind turbines For onshore turbines an expected lie

o 10 years is consistent with re-powering decisions over the last decade while an expected lie

o 15 years seems to be the maximum consistent with the age-perormance profile For offshore

turbines the analysis uses expected lives o 15 and 20 years

Figure 6A Projected load factors for UK onshore wind (total capacity of 10 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

Figure 6B Projected load factors for UK onshore wind (total capacity of 15 GW in 2020)

10

12

14

16

18

20

22

24

26

28

UK ndash life 10 yrs UK ndash life 15 yrs DK ndash life 10 yrs DK ndash life 15 yrs

L o a d

f a c t o r ( )

2007ndash11 2015 2020

Source Authorrsquos estimates

29 Figure 6A shows projected values o the average normalised load actors or UK onshore wind

arms in 2015 and 2020 using a target capacity in 2020 equal to 10 GW plus the age-peror-

mance profiles or the UK and Denmark (DK) Since vintage effects are important in this

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 19

calculation the profile o new or replacement investment affects the results In this case the

level o gross investment is assumed to be constant rom 2013 onwards at either 930 or 1220

MW per year ndash the range is defined by the assumed lie o turbines Achieving a target o 15

GW in 2020 (Figure 6B) would require a level o investment rom 2013 onwards o either 1540

or 1930 MW per year Such figures are 3ndash4 times the average level o investment over theperiod 2005ndash11 so the lower target may provide a better indication o what will happen

30 All o the estimates are calibrated to match the observed average load actor in 2007ndash11 so that

they should be interpreted as relative rather than absolute values Using the age-perormance

profile or UK onshore wind installations the average normalised load actor is likely to all by

about 19 rom 2007ndash11 to 2015 and by 23ndash28 to 2020 I the age-perormance curves or

Danish onshore installations were to apply the reduction in the average normalised load actor

would be considerably less but there would still be a all o up to 5 by 2020

31 In the medium term the all is greater i turbines operate or 15 years because the average age

o all turbines will be higher and thus the degradation in perormance due to age will have a

larger effect By 2020 this effect reduces the average normalised load actor rom 203 or a 10

year lie to 195 or a 15 year lie here is a complex economic and commercial trade-off here

o meet a target o 10 GW in 2020 it is necessary to increase the level o investment in onshore

capacity by 30 rom 2013 Almost all o this will take the orm o re-powering existing wind

arms11 he costs are airly high because this will usually involve new development consents

decommissioning the original turbines and replacing them with a smaller number o larger and

more powerul turbines

32 Estimates o re-powering costs are generally treated as being commercially sensitive but broadproject costs cited in specialist magazines suggest that the unit costs are likely to be at least

pound1000 per kW or more than pound300 million per year in aggregate On the other hand the increase

in the amount o energy generated would be about 12 Wh or a 10 year turbine lie rather

than a 15 year turbine lie though this does not allow or the loss o production during re-pow-

ering Under the best circumstances the payback period may be quite short but the best option

rom a broader economic and social perspective is not obvious

33 Even under the best scenario 10 GW o onshore wind capacity will only generate 178 Wh o

electricity in 2020 well below the lower end o the target range or onshore wind generation

specified in DECCrsquos road map In act allowing or interactions between the age-perormance

profile and investment the amount o onshore wind capacity required in 2020 to generate 24

Wh would be a minimum o 134 GW At the top end o DECCrsquos target range the capacity

required in 2020 to generate 32 Wh would be a minimum o 178 GW Since these estimates

are based on a 10 year turbine lie the level o investment required rom 2013 onwards would

be 1650 MW per year or the bottom o the range and 2350 MW or the top end o the range

In the last 8 years total investment in onshore wind has only ever exceeded 600 MW in one

11 Re-powering is usually defined as the ull replacement o existing turbines whereas retrofitting covers the replace-

ment o various mechanical or electrical components In the data analysis re-powering involves the creation o a

new unit whereas retrofitting affects the perormance o a continuing unit Hence the potential benefits o retrofit-

ting are taken into account in constructing the age-perormance curves

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2252

20 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

year (883 MW in 2008) so it seems rather unlikely that even the lower end o DECCrsquos range or

onshore wind generation will actually be realised

34 O course this assessment would be different i UK onshore wind operators were able to match

the Danish age-perormance curve here are many actors which differentiate the UK and

Danish wind industries he average scale o wind arms in the UK is larger they are moreremote and there appears to be significantly greater variation between operating and potential

new sites in achieved perormance It would thereore require a very optimistic observer to

conclude that the difference in age-perormance curves between the two countries will disap-

pear in the short or medium term

Figure 7 Projected load factors for UK offshore wind in 2020

20

22

24

26

28

30

32

34

18 GW ndash life 15 yrs 18 GW ndash life 20 yrs 24 GW ndash life 15 yrs 24 GW ndash life 20 yrs

L o a d f a c t o r ( )

2007ndash11 2015 2020

Note Based on Danish age-perormance curves Source Authorrsquos estimates

35 Figure 7 shows the results rom carrying out a similar exercise or offshore wind arms in the

UK he only evidence on the age-perormance curves or offshore installations is based upon

the Danish experience so these have been used to construct the projections Further the profile

o investment assumes a progressive increase in additions to capacity rom 2013 to 2016 and

then a stable level o investment rom 2017 to 2020 he DECC road map involves a very large

commitment to offshore wind investment ndash in the range 25-35 GW o new capacity annually

rom 2016 onwards As a consequence the average age o offshore capacity in 2020 will be quite

low and the ull effects o the age-perormance curves will only be elt during the decade ollow-

ing 2020 Even so the average normalised load actor in 2020 will be about 15 lower than or

2007ndash11 while by 2025 it will be about 30 below the base level

36 On these projections 232 GW o offshore capacity is the minimum necessary to generate 58

Wh o electricity rom offshore wind in 2020 his requirement will increase to 298 GW by

2025 simply to offset the effect o the ageing o offshore generation capacity on total output

37 Combining the estimates or onshore and offshore wind generation the total amount o wind

capacity required in 2020 to produce 90 Wh o electricity rom wind will be in the range 39

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 21

to 41 GW once allowance is made or the observed age-perormance curves or both onshore

and offshore installations hese estimates are roughly a third higher than official predictions o

28ndash31 GW o wind capacity in 2020

38 urning the figures round gives a slightly different picture o the reality o the governmentrsquos

strategy on renewable electricity generation A total o 30 GW o wind capacity in 2020 (12GW onshore 18 GW offshore) will generate about 64 Wh o electricity in a normal year

Since electricity generation rom wind and biomass are expected to account or 50 o renew-

able energy in 2020 the consequence is that biomass is expected to account or 53 Wh In

2011 landfill gas and sewage sludge digestion accounted or 44 o electricity rom bioenergy

(DUKES ables 64) here are limited opportunities to expand production rom these sources

so a continuation o historic rates o growth would mean that renewable generation rom other

sources o bioenergy would have to increase rom 72 Wh in 2011 to 46 Wh in 2020 here

are severe constraints on increasing the amounts o electricity generated rom animal biomass

biodegradable municipal solid waste and anaerobic digestion so it would be optimistic to

assume that these will contribute more than 6-8 Wh in 2020 (up rom 26 Wh in 2011)

39 In practice i the projections or renewable energy are to be believed they rest upon an assump-

tion that there will be a large increase in the amount o electricity that is generated rom plant

biomass ndash either on its own or through co-firing with coal his increase will involve increas-

ing the amount o electricity generated in these ways rom 46 Wh in 2011 to 38-40 Wh in

2020 Since the amounts o straw and UK-grown timber available or this purpose are strictly

limited this will come down to the amount o wood chips that can be imported and the cost o

doing this40 he historic average load actor or biomass plants is just over 60 so about 7500 MW o

biomass capacity will be required to supply 40 Wh o electricity he weight o wood chips per

MWh o electricity depends upon moisture content storage boiler design and other actors but

an indicative range is 085 to 115 tonnes per MWh o wood chips with a moisture content o

30 According to DECC and Forestry Commission estimates the UK imported 265 million

tonnes o steam coal and 26 million tonnes o other wood products (including wood chips)

in 2011 It seems that the net effect o the governmentrsquos strategy will be to replace 15 million

tonnes o coal imported in 2011 by 32 to 44 million tonnes o imported wood chips his is a

somewhat strange way o increasing energy security

Conclusion

41 Wind power is a highly capital-intensive technology or generating electricity Its merits rely

entirely upon a substitution o capital or uel inputs he same is true or hydro or tidal or wave

power In comparison with hydro power wind is a low quality resource because o its variability

and because it cannot be stored So the economic case or wind power must rest on obtaining

the most out o the wind that is available

42 While the decline in the achieved perormance o onshore wind turbines in Denmark is

much less than that or the UK or offshore nonetheless the decline in expected output under

standardised wind conditions over 10 years is 10 unweighted and 13 capacity weighted

hese declines accelerate afer age 10 so that the reductions in perormance are 17 and 20

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2452

22 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

respectively afer 15 years For UK onshore wind arms the reduction in perormance due to

age is much worse at 27 unweighted and 69 capacity weighted by age 10

43 Evidence on the perormance o Danish offshore installations is both restricted and so poor

that there may be concern that the results are affected by a small number o outliers Still the

sample contains a reasonable number o sites with at least 5 years o operating experience andthe decline in perormance by age 5 is 38 unweighted and 26 capacity weighted

44 In addition to these results there is strong evidence that the average normalised load actor or

new onshore wind installations in the UK has allen significantly over the period rom 2000 to

2011 his is consistent with a pattern in which the most avourable sites are developed first

Equally it could mean that wind developers have been unable to keep up with the rate o new

investment while maintaining the quality o development and operations For example the site

design or selection o turbine characteristics may make less effective use o the available wind

resources or the sites available than was the case in the past

45 Whatever the reasons the deterioration in initial perormance means that the expected returns

rom the expansion in wind capacity both or investors and in terms o the reduction in CO983090

emissions have been alling without a concomitant decrease in the private and social costs

that are borne by customers and the general public Clearly this is unsatisactory at best and it

suggests that the benefits claimed or current policies cannot be taken at ace value

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 23

Appendix Data and Methods

A Data for the United Kingdom

he raw data used or this study was extracted by the Renewable Energy Foundation rom the

Renewables and CHP Register database compiled by Ogem12 his inormation is used in the

administration o the market in Renewable Obligation Certificates (ROCs) he Renewables

Obligation is the primary mechanism by which large scale generators o renewable electric-

ity receive subsidies so the operators o wind arms have a very strong incentive to submit

complete and up-to-date inormation to the database he data extracted covered all onshore

wind generators with at least 05 MW o generating capacity which are eligible or ROCs or the

10 year period April 2002 to March 2012 he key variables are the monthly output o electric-

ity (used or the purpose o allocating ROCs) and nominal generating capacity here are 282

reporting units in the final dataset A reporting unit may correspond to an entire wind arm orto separate phases o development Complications arise when an existing wind arm is re-pow-

ered ie when old wind turbines are replaced by newer and usually more powerul turbines

he database reports this as a change in the generating capacity o the recording unit but in

this analysis repowering is treated as the termination o the old recording unit and the creation

o a new recording unit

A considerable amount o data checking and cleaning was required beore the data could

be analysed he first step was to add the age o each reporting unit his was calculated by

reerence to the month in which the turbines were commissioned For wind arms in operation

in or beore June 2002 the commissioning date was obtained rom various online sources ndash

notably the thewindpowernet database o UK wind arms13 which was cross-checked against

the inormation provided on the websites o wind operators and other sources For recording

units whose first data relates to dates afer June 2002 the commissioning date was initially

assumed to be the month or which data is first reported Cross-checks were carried out or

all recording units which commenced reporting in the period July 2002 to December 2003 In

more than 80 o cases the externally-reported commissioning date is within one month o the

first month or which data is reported he exceptions ndash Bu Farm and Mablethorpe ndash appear to

be due to delays by small operators in reporting data to Ogem

In a number o cases it is clear rom the data that plants were not in ull operation at the

time when data was first reported to the database here are two indicators o partial or delayed

commissioning

bull For an initial series o months the calculated load actor is very low ndash usually below 5 ndash

and then increases abruptly In such cases the data or the early months is dropped and the

commissioning date is treated as the month immediately preceding the first month with a

lsquonormalrsquo load actor14

12 httpswwwrenewablesandchpogemgovuk

13 See httpwwwthewindpowernetwindarms_list_enphp

14 he month in which a plant is reported as having been commissioned is not included in the analysis

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2652

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 25

the decline in the plantrsquos load actor as it ages As an example consider a plant whose reported

capacity was increased rom 99 MW to 10 MW at age 3 I this was simply a change in report-

ing practice this would reduce the estimated load actor rom age 3 onwards By increasing

the initial capacity rom 99 MW to 10 MW ndash or by reducing the capacity rom 10 MW to 99

MW rom age 3 onwards ndash this bias is removed It is not important which adjustment is madebecause the difference between the alternative assumptions is captured by the unit fixed effect

he numbers o separate sites and o observations per site mean that it is not possible to

provide a simple visual representation o the data Figure 8 provides a summary by illustrating

the distributions o load actors by the age o the wind arm pooled across all years he ranges

o the load actors or each age are wide while the medians show no clear trend by age hus a

naiumlve analysis might conclude that there is nothing to investigate However as will be explained

below the distributions mask crucial differences between the perormance o wind arms as

they age because they do not control or the differences in location and wind availability over

time

B Data for Denmark

he data or Danish wind arms used in this study comes rom a database compiled by the

Danish Energy Agency covering the characteristics and perormance o all wind turbines rom

2002 up to the end o August 201216 he basic recording unit or the register is the individual

wind turbine but where a number o turbines are connected to a single meter an identical

average output per turbine is recorded or all turbines in the group he presence o what are in

effect repeated observations increases the noise in the data In addition the model that is used

or the analysis assumes that site operator and turbine characteristics are common or all o theunits that make up a single wind arm o address this point separate wind arms have been

identified by grouping all turbines which have identical values or turbine manuacturer and

type local authority date o commissioning and date o decommissioning (where applicable)

he register contains inormation on 7569 turbines commissioned rom 1977 onwards his

includes a large number o small turbines which would not all within the scope o the ROC

scheme in the UK Hence or comparability with the UK data all turbines commissioned

prior to 1992 were dropped as well as all wind arms with a generating capacity o less than 1

MW he resulting sample covered 823 onshore wind arms and 30 offshore wind arms with

monthly load actors or various periods rom January 2002 to August 2012 Given the cut-off

or wind arms commissioned prior to 1992 the maximum age or an onshore wind arm was 20

years and the maximum age or an offshore wind arm was 17 years

he total capacity o the onshore wind arms recorded in the Danish database was 2757 MW

he average size o an onshore wind arm in Denmark is small ndash only 33 MW as compared

with 161 MW or the UK ndash because 84 o the sample consists o installations with between

1 and 4 turbines he largest onshore installation has 39 turbines with a capacity o 234 MW

while the onshore installation with the greatest capacity has 19 turbines with a capacity o 584

MW Overall the sample o Danish onshore wind arms is older and smaller than the sample

o UK onshore wind arms reflecting the much longer history o wind generation in Denmark

16 httpwwwensdkEN-USINFOFACSANDFIGURESENERGY_SAISICS_AND_INDICAORS

OVERVIEWOFHEENERGYSECORREGISEROFWINDURBINESSiderForsideaspx

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2852

26 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

he larger scale o UK wind arms may account or some o the differences in perormance

curves or the two countries

he sample o offshore wind arms in Denmark is quite small with 30 installations in

total Still it is useul to analyse the perormance o these installations since there is very little

comparable data or the UK he sample covers 394 turbines with an average o just over 13turbines per installation and an average capacity o 288 MW he largest installation which

was commissioned in 2003 has 72 turbines with a total capacity o 1656 MW Again offshore

installations are smaller and older in Denmark than is the current pattern in the UK he first

two wind arms in the sample were commissioned in the years 1995 and 2000 but the sample

increases significantly rom 2002 onwards o avoid reliance upon a sample o 1 or 2 wind

arms to determine specific coefficients or ages gt 10 years all observations with an age greater

than 10 were omitted rom the analysis17

A somewhat different problem with the offshore wind arm data arose because the coeffi-

cients estimated using capacity weights (see below) were heavily influenced by the largest windarm his has had an extremely erratic profile o perormance over time with a very low average

load actor he treatment o this wind arm is discussed in more detail in Section F below

C Specication and estimation methods

As Figure 8 illustrates it is necessary to go beyond simple summary statistics in order to assess

whether there is any systematic relationship between age and the average perormance o wind

arms he perormance o wind arms varies over both time and space some months and years

have more or less wind than the long term average while specific characteristics o wind arms

ndash including location type o turbine and operating regime ndash will influence the perormance oeach plant under identical wind conditions

he combination o (a) site-specific characteristics which are constant over time and (b)

period-specific characteristics which are constant across wind arms can be represented by

what statisticians call an error components model with fixed effects or each wind arm and

each time period he dependent variable in the model consists o a sequence o load actors ndash

denoted by or unit i in month t ndash or each wind arm (the panel unit) over time he load

actors are calculated by dividing the total output in MWh by (24 x number o days in month x

reported generating capacity in MW) and are multiplied by 100 to convert to percentages

he additive version o the error components model may be written as

(1)

in which denotes the age o plant i in period t and is some unction that is either a

representation o or an approximation to the normalised age-perormance relationship he

multiplicative version is similar except that is replaced by where ln() denotes the

natural logarithm he terms are wind arm or unit fixed effects which capture actors such

as location type o turbine operating regimes etc he terms are period fixed effects which

capture the influence o wind conditions over the UK or Denmark as a whole as well as seasonal

maintenance and demand Without loss o generality it may be assumed that the fixed effects are

17 For the avoidance o doubt this means that no attempt was made to estimate the effect o age on the perormance o

offshore wind arms beyond age 10 while the data or the oldest wind arms was included in the analysis up to and

including age 10

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 27

normalised so that ie all o the error components have a zero mean while the

mean o the random error is zero by definition he terms are random errors which capture

random variations in wind conditions that are not specific to the site turbine breakdowns and

other actors that are uncorrelated with either the site or the time period It is assumed that the

mean o the random error is zero while additional assumptions depend upon the method o

estimation that is adopted

he model specified in equation (1) can be estimated in a variety o ways18 A crucial issue is

whether the are or may be significantly correlated with the age terms or any other independent

regressors I it is assumed that there is no such correlation the model is usually reerred to as

a random effects model In that case the equation can be estimated by a variant o least squares

(LS) generalised least squares (GLS) generalised methods o moments (GMM) or maximum

likelihood (ML) ndash all o which should be yield parameter estimates that will converge in prob-

ability to the true parameter values as the sample size increases An alternative but less restric-

tive specification is known as the fixed effects model in which no assumption is made about

the correlation between the unit fixed effects and other regressors he limitation o the fixed

effects model is that it cannot be used to estimate parameters or regressors that vary across

panel units but are constant over time or a panel unit (time-invariant regressors) ndash eg the

year in which a wind arm was commissioned its location or its rated generating capacity he

influence o such variables is captured by the unit fixed effects

In order to avoid the risk o obtaining biased estimates o the age effects the results reported

in this note are based upon estimates using the fixed effects model he influence o time-invar-

iant regressors is examined in a second stage by regressing the estimates o the unit fixed effectson these variables he simplest method o estimating the fixed effects model is known as the

lsquowithinrsquo estimator in which the mean value or each unit is subtracted rom all observations or

that unit For example i is a general polynomial o order M in A equation (1) becomes

(2)

so that the model can be estimated by least square with the inclusion o dummy variables or

each period t he variance-covariance matrix o the coefficients have been estimated by using

(a) a robust sandwich estimator adjusted or clustering by panel unit and (b) a bootstrap esti-

mator with 400 repetitions he estimates o the standard errors will be consistent even i the

random error component is serially correlated within panel units andor has different variances

across panel units19 Since the number o panel units and the average number o time periods

per panel units are both large (other than or Danish offshore wind arms) the least square esti-

mates o the parameters will be unbiased under these assumptions

he bootstrap method is expensive to carry out but it provides a useul cross-check on

the standard errors generated by other methods Unortunately the bootstrap method o

18 See Chapter 21 o AC Cameron amp PK rivedi ndash Microeconometrics Methods and Applications (Cambridge

Cambridge University Press 2005) or a standard textbook treatment o estimation methods or panel data models

19 he estimation was carried out using the xtreg procedure in Stata Version 12 he methods o constructing the

sandwich and bootstrap estimates o the variance-covariance matrix are described in Chapters 8 amp 13 o AC

Cameron amp PK rivedi ndash Microeconometrics using Stata (Revised Edition College Station exas Stata Press 2010)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 21: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 19

calculation the profile o new or replacement investment affects the results In this case the

level o gross investment is assumed to be constant rom 2013 onwards at either 930 or 1220

MW per year ndash the range is defined by the assumed lie o turbines Achieving a target o 15

GW in 2020 (Figure 6B) would require a level o investment rom 2013 onwards o either 1540

or 1930 MW per year Such figures are 3ndash4 times the average level o investment over theperiod 2005ndash11 so the lower target may provide a better indication o what will happen

30 All o the estimates are calibrated to match the observed average load actor in 2007ndash11 so that

they should be interpreted as relative rather than absolute values Using the age-perormance

profile or UK onshore wind installations the average normalised load actor is likely to all by

about 19 rom 2007ndash11 to 2015 and by 23ndash28 to 2020 I the age-perormance curves or

Danish onshore installations were to apply the reduction in the average normalised load actor

would be considerably less but there would still be a all o up to 5 by 2020

31 In the medium term the all is greater i turbines operate or 15 years because the average age

o all turbines will be higher and thus the degradation in perormance due to age will have a

larger effect By 2020 this effect reduces the average normalised load actor rom 203 or a 10

year lie to 195 or a 15 year lie here is a complex economic and commercial trade-off here

o meet a target o 10 GW in 2020 it is necessary to increase the level o investment in onshore

capacity by 30 rom 2013 Almost all o this will take the orm o re-powering existing wind

arms11 he costs are airly high because this will usually involve new development consents

decommissioning the original turbines and replacing them with a smaller number o larger and

more powerul turbines

32 Estimates o re-powering costs are generally treated as being commercially sensitive but broadproject costs cited in specialist magazines suggest that the unit costs are likely to be at least

pound1000 per kW or more than pound300 million per year in aggregate On the other hand the increase

in the amount o energy generated would be about 12 Wh or a 10 year turbine lie rather

than a 15 year turbine lie though this does not allow or the loss o production during re-pow-

ering Under the best circumstances the payback period may be quite short but the best option

rom a broader economic and social perspective is not obvious

33 Even under the best scenario 10 GW o onshore wind capacity will only generate 178 Wh o

electricity in 2020 well below the lower end o the target range or onshore wind generation

specified in DECCrsquos road map In act allowing or interactions between the age-perormance

profile and investment the amount o onshore wind capacity required in 2020 to generate 24

Wh would be a minimum o 134 GW At the top end o DECCrsquos target range the capacity

required in 2020 to generate 32 Wh would be a minimum o 178 GW Since these estimates

are based on a 10 year turbine lie the level o investment required rom 2013 onwards would

be 1650 MW per year or the bottom o the range and 2350 MW or the top end o the range

In the last 8 years total investment in onshore wind has only ever exceeded 600 MW in one

11 Re-powering is usually defined as the ull replacement o existing turbines whereas retrofitting covers the replace-

ment o various mechanical or electrical components In the data analysis re-powering involves the creation o a

new unit whereas retrofitting affects the perormance o a continuing unit Hence the potential benefits o retrofit-

ting are taken into account in constructing the age-perormance curves

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2252

20 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

year (883 MW in 2008) so it seems rather unlikely that even the lower end o DECCrsquos range or

onshore wind generation will actually be realised

34 O course this assessment would be different i UK onshore wind operators were able to match

the Danish age-perormance curve here are many actors which differentiate the UK and

Danish wind industries he average scale o wind arms in the UK is larger they are moreremote and there appears to be significantly greater variation between operating and potential

new sites in achieved perormance It would thereore require a very optimistic observer to

conclude that the difference in age-perormance curves between the two countries will disap-

pear in the short or medium term

Figure 7 Projected load factors for UK offshore wind in 2020

20

22

24

26

28

30

32

34

18 GW ndash life 15 yrs 18 GW ndash life 20 yrs 24 GW ndash life 15 yrs 24 GW ndash life 20 yrs

L o a d f a c t o r ( )

2007ndash11 2015 2020

Note Based on Danish age-perormance curves Source Authorrsquos estimates

35 Figure 7 shows the results rom carrying out a similar exercise or offshore wind arms in the

UK he only evidence on the age-perormance curves or offshore installations is based upon

the Danish experience so these have been used to construct the projections Further the profile

o investment assumes a progressive increase in additions to capacity rom 2013 to 2016 and

then a stable level o investment rom 2017 to 2020 he DECC road map involves a very large

commitment to offshore wind investment ndash in the range 25-35 GW o new capacity annually

rom 2016 onwards As a consequence the average age o offshore capacity in 2020 will be quite

low and the ull effects o the age-perormance curves will only be elt during the decade ollow-

ing 2020 Even so the average normalised load actor in 2020 will be about 15 lower than or

2007ndash11 while by 2025 it will be about 30 below the base level

36 On these projections 232 GW o offshore capacity is the minimum necessary to generate 58

Wh o electricity rom offshore wind in 2020 his requirement will increase to 298 GW by

2025 simply to offset the effect o the ageing o offshore generation capacity on total output

37 Combining the estimates or onshore and offshore wind generation the total amount o wind

capacity required in 2020 to produce 90 Wh o electricity rom wind will be in the range 39

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 21

to 41 GW once allowance is made or the observed age-perormance curves or both onshore

and offshore installations hese estimates are roughly a third higher than official predictions o

28ndash31 GW o wind capacity in 2020

38 urning the figures round gives a slightly different picture o the reality o the governmentrsquos

strategy on renewable electricity generation A total o 30 GW o wind capacity in 2020 (12GW onshore 18 GW offshore) will generate about 64 Wh o electricity in a normal year

Since electricity generation rom wind and biomass are expected to account or 50 o renew-

able energy in 2020 the consequence is that biomass is expected to account or 53 Wh In

2011 landfill gas and sewage sludge digestion accounted or 44 o electricity rom bioenergy

(DUKES ables 64) here are limited opportunities to expand production rom these sources

so a continuation o historic rates o growth would mean that renewable generation rom other

sources o bioenergy would have to increase rom 72 Wh in 2011 to 46 Wh in 2020 here

are severe constraints on increasing the amounts o electricity generated rom animal biomass

biodegradable municipal solid waste and anaerobic digestion so it would be optimistic to

assume that these will contribute more than 6-8 Wh in 2020 (up rom 26 Wh in 2011)

39 In practice i the projections or renewable energy are to be believed they rest upon an assump-

tion that there will be a large increase in the amount o electricity that is generated rom plant

biomass ndash either on its own or through co-firing with coal his increase will involve increas-

ing the amount o electricity generated in these ways rom 46 Wh in 2011 to 38-40 Wh in

2020 Since the amounts o straw and UK-grown timber available or this purpose are strictly

limited this will come down to the amount o wood chips that can be imported and the cost o

doing this40 he historic average load actor or biomass plants is just over 60 so about 7500 MW o

biomass capacity will be required to supply 40 Wh o electricity he weight o wood chips per

MWh o electricity depends upon moisture content storage boiler design and other actors but

an indicative range is 085 to 115 tonnes per MWh o wood chips with a moisture content o

30 According to DECC and Forestry Commission estimates the UK imported 265 million

tonnes o steam coal and 26 million tonnes o other wood products (including wood chips)

in 2011 It seems that the net effect o the governmentrsquos strategy will be to replace 15 million

tonnes o coal imported in 2011 by 32 to 44 million tonnes o imported wood chips his is a

somewhat strange way o increasing energy security

Conclusion

41 Wind power is a highly capital-intensive technology or generating electricity Its merits rely

entirely upon a substitution o capital or uel inputs he same is true or hydro or tidal or wave

power In comparison with hydro power wind is a low quality resource because o its variability

and because it cannot be stored So the economic case or wind power must rest on obtaining

the most out o the wind that is available

42 While the decline in the achieved perormance o onshore wind turbines in Denmark is

much less than that or the UK or offshore nonetheless the decline in expected output under

standardised wind conditions over 10 years is 10 unweighted and 13 capacity weighted

hese declines accelerate afer age 10 so that the reductions in perormance are 17 and 20

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2452

22 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

respectively afer 15 years For UK onshore wind arms the reduction in perormance due to

age is much worse at 27 unweighted and 69 capacity weighted by age 10

43 Evidence on the perormance o Danish offshore installations is both restricted and so poor

that there may be concern that the results are affected by a small number o outliers Still the

sample contains a reasonable number o sites with at least 5 years o operating experience andthe decline in perormance by age 5 is 38 unweighted and 26 capacity weighted

44 In addition to these results there is strong evidence that the average normalised load actor or

new onshore wind installations in the UK has allen significantly over the period rom 2000 to

2011 his is consistent with a pattern in which the most avourable sites are developed first

Equally it could mean that wind developers have been unable to keep up with the rate o new

investment while maintaining the quality o development and operations For example the site

design or selection o turbine characteristics may make less effective use o the available wind

resources or the sites available than was the case in the past

45 Whatever the reasons the deterioration in initial perormance means that the expected returns

rom the expansion in wind capacity both or investors and in terms o the reduction in CO983090

emissions have been alling without a concomitant decrease in the private and social costs

that are borne by customers and the general public Clearly this is unsatisactory at best and it

suggests that the benefits claimed or current policies cannot be taken at ace value

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 23

Appendix Data and Methods

A Data for the United Kingdom

he raw data used or this study was extracted by the Renewable Energy Foundation rom the

Renewables and CHP Register database compiled by Ogem12 his inormation is used in the

administration o the market in Renewable Obligation Certificates (ROCs) he Renewables

Obligation is the primary mechanism by which large scale generators o renewable electric-

ity receive subsidies so the operators o wind arms have a very strong incentive to submit

complete and up-to-date inormation to the database he data extracted covered all onshore

wind generators with at least 05 MW o generating capacity which are eligible or ROCs or the

10 year period April 2002 to March 2012 he key variables are the monthly output o electric-

ity (used or the purpose o allocating ROCs) and nominal generating capacity here are 282

reporting units in the final dataset A reporting unit may correspond to an entire wind arm orto separate phases o development Complications arise when an existing wind arm is re-pow-

ered ie when old wind turbines are replaced by newer and usually more powerul turbines

he database reports this as a change in the generating capacity o the recording unit but in

this analysis repowering is treated as the termination o the old recording unit and the creation

o a new recording unit

A considerable amount o data checking and cleaning was required beore the data could

be analysed he first step was to add the age o each reporting unit his was calculated by

reerence to the month in which the turbines were commissioned For wind arms in operation

in or beore June 2002 the commissioning date was obtained rom various online sources ndash

notably the thewindpowernet database o UK wind arms13 which was cross-checked against

the inormation provided on the websites o wind operators and other sources For recording

units whose first data relates to dates afer June 2002 the commissioning date was initially

assumed to be the month or which data is first reported Cross-checks were carried out or

all recording units which commenced reporting in the period July 2002 to December 2003 In

more than 80 o cases the externally-reported commissioning date is within one month o the

first month or which data is reported he exceptions ndash Bu Farm and Mablethorpe ndash appear to

be due to delays by small operators in reporting data to Ogem

In a number o cases it is clear rom the data that plants were not in ull operation at the

time when data was first reported to the database here are two indicators o partial or delayed

commissioning

bull For an initial series o months the calculated load actor is very low ndash usually below 5 ndash

and then increases abruptly In such cases the data or the early months is dropped and the

commissioning date is treated as the month immediately preceding the first month with a

lsquonormalrsquo load actor14

12 httpswwwrenewablesandchpogemgovuk

13 See httpwwwthewindpowernetwindarms_list_enphp

14 he month in which a plant is reported as having been commissioned is not included in the analysis

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2652

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 25

the decline in the plantrsquos load actor as it ages As an example consider a plant whose reported

capacity was increased rom 99 MW to 10 MW at age 3 I this was simply a change in report-

ing practice this would reduce the estimated load actor rom age 3 onwards By increasing

the initial capacity rom 99 MW to 10 MW ndash or by reducing the capacity rom 10 MW to 99

MW rom age 3 onwards ndash this bias is removed It is not important which adjustment is madebecause the difference between the alternative assumptions is captured by the unit fixed effect

he numbers o separate sites and o observations per site mean that it is not possible to

provide a simple visual representation o the data Figure 8 provides a summary by illustrating

the distributions o load actors by the age o the wind arm pooled across all years he ranges

o the load actors or each age are wide while the medians show no clear trend by age hus a

naiumlve analysis might conclude that there is nothing to investigate However as will be explained

below the distributions mask crucial differences between the perormance o wind arms as

they age because they do not control or the differences in location and wind availability over

time

B Data for Denmark

he data or Danish wind arms used in this study comes rom a database compiled by the

Danish Energy Agency covering the characteristics and perormance o all wind turbines rom

2002 up to the end o August 201216 he basic recording unit or the register is the individual

wind turbine but where a number o turbines are connected to a single meter an identical

average output per turbine is recorded or all turbines in the group he presence o what are in

effect repeated observations increases the noise in the data In addition the model that is used

or the analysis assumes that site operator and turbine characteristics are common or all o theunits that make up a single wind arm o address this point separate wind arms have been

identified by grouping all turbines which have identical values or turbine manuacturer and

type local authority date o commissioning and date o decommissioning (where applicable)

he register contains inormation on 7569 turbines commissioned rom 1977 onwards his

includes a large number o small turbines which would not all within the scope o the ROC

scheme in the UK Hence or comparability with the UK data all turbines commissioned

prior to 1992 were dropped as well as all wind arms with a generating capacity o less than 1

MW he resulting sample covered 823 onshore wind arms and 30 offshore wind arms with

monthly load actors or various periods rom January 2002 to August 2012 Given the cut-off

or wind arms commissioned prior to 1992 the maximum age or an onshore wind arm was 20

years and the maximum age or an offshore wind arm was 17 years

he total capacity o the onshore wind arms recorded in the Danish database was 2757 MW

he average size o an onshore wind arm in Denmark is small ndash only 33 MW as compared

with 161 MW or the UK ndash because 84 o the sample consists o installations with between

1 and 4 turbines he largest onshore installation has 39 turbines with a capacity o 234 MW

while the onshore installation with the greatest capacity has 19 turbines with a capacity o 584

MW Overall the sample o Danish onshore wind arms is older and smaller than the sample

o UK onshore wind arms reflecting the much longer history o wind generation in Denmark

16 httpwwwensdkEN-USINFOFACSANDFIGURESENERGY_SAISICS_AND_INDICAORS

OVERVIEWOFHEENERGYSECORREGISEROFWINDURBINESSiderForsideaspx

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2852

26 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

he larger scale o UK wind arms may account or some o the differences in perormance

curves or the two countries

he sample o offshore wind arms in Denmark is quite small with 30 installations in

total Still it is useul to analyse the perormance o these installations since there is very little

comparable data or the UK he sample covers 394 turbines with an average o just over 13turbines per installation and an average capacity o 288 MW he largest installation which

was commissioned in 2003 has 72 turbines with a total capacity o 1656 MW Again offshore

installations are smaller and older in Denmark than is the current pattern in the UK he first

two wind arms in the sample were commissioned in the years 1995 and 2000 but the sample

increases significantly rom 2002 onwards o avoid reliance upon a sample o 1 or 2 wind

arms to determine specific coefficients or ages gt 10 years all observations with an age greater

than 10 were omitted rom the analysis17

A somewhat different problem with the offshore wind arm data arose because the coeffi-

cients estimated using capacity weights (see below) were heavily influenced by the largest windarm his has had an extremely erratic profile o perormance over time with a very low average

load actor he treatment o this wind arm is discussed in more detail in Section F below

C Specication and estimation methods

As Figure 8 illustrates it is necessary to go beyond simple summary statistics in order to assess

whether there is any systematic relationship between age and the average perormance o wind

arms he perormance o wind arms varies over both time and space some months and years

have more or less wind than the long term average while specific characteristics o wind arms

ndash including location type o turbine and operating regime ndash will influence the perormance oeach plant under identical wind conditions

he combination o (a) site-specific characteristics which are constant over time and (b)

period-specific characteristics which are constant across wind arms can be represented by

what statisticians call an error components model with fixed effects or each wind arm and

each time period he dependent variable in the model consists o a sequence o load actors ndash

denoted by or unit i in month t ndash or each wind arm (the panel unit) over time he load

actors are calculated by dividing the total output in MWh by (24 x number o days in month x

reported generating capacity in MW) and are multiplied by 100 to convert to percentages

he additive version o the error components model may be written as

(1)

in which denotes the age o plant i in period t and is some unction that is either a

representation o or an approximation to the normalised age-perormance relationship he

multiplicative version is similar except that is replaced by where ln() denotes the

natural logarithm he terms are wind arm or unit fixed effects which capture actors such

as location type o turbine operating regimes etc he terms are period fixed effects which

capture the influence o wind conditions over the UK or Denmark as a whole as well as seasonal

maintenance and demand Without loss o generality it may be assumed that the fixed effects are

17 For the avoidance o doubt this means that no attempt was made to estimate the effect o age on the perormance o

offshore wind arms beyond age 10 while the data or the oldest wind arms was included in the analysis up to and

including age 10

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 27

normalised so that ie all o the error components have a zero mean while the

mean o the random error is zero by definition he terms are random errors which capture

random variations in wind conditions that are not specific to the site turbine breakdowns and

other actors that are uncorrelated with either the site or the time period It is assumed that the

mean o the random error is zero while additional assumptions depend upon the method o

estimation that is adopted

he model specified in equation (1) can be estimated in a variety o ways18 A crucial issue is

whether the are or may be significantly correlated with the age terms or any other independent

regressors I it is assumed that there is no such correlation the model is usually reerred to as

a random effects model In that case the equation can be estimated by a variant o least squares

(LS) generalised least squares (GLS) generalised methods o moments (GMM) or maximum

likelihood (ML) ndash all o which should be yield parameter estimates that will converge in prob-

ability to the true parameter values as the sample size increases An alternative but less restric-

tive specification is known as the fixed effects model in which no assumption is made about

the correlation between the unit fixed effects and other regressors he limitation o the fixed

effects model is that it cannot be used to estimate parameters or regressors that vary across

panel units but are constant over time or a panel unit (time-invariant regressors) ndash eg the

year in which a wind arm was commissioned its location or its rated generating capacity he

influence o such variables is captured by the unit fixed effects

In order to avoid the risk o obtaining biased estimates o the age effects the results reported

in this note are based upon estimates using the fixed effects model he influence o time-invar-

iant regressors is examined in a second stage by regressing the estimates o the unit fixed effectson these variables he simplest method o estimating the fixed effects model is known as the

lsquowithinrsquo estimator in which the mean value or each unit is subtracted rom all observations or

that unit For example i is a general polynomial o order M in A equation (1) becomes

(2)

so that the model can be estimated by least square with the inclusion o dummy variables or

each period t he variance-covariance matrix o the coefficients have been estimated by using

(a) a robust sandwich estimator adjusted or clustering by panel unit and (b) a bootstrap esti-

mator with 400 repetitions he estimates o the standard errors will be consistent even i the

random error component is serially correlated within panel units andor has different variances

across panel units19 Since the number o panel units and the average number o time periods

per panel units are both large (other than or Danish offshore wind arms) the least square esti-

mates o the parameters will be unbiased under these assumptions

he bootstrap method is expensive to carry out but it provides a useul cross-check on

the standard errors generated by other methods Unortunately the bootstrap method o

18 See Chapter 21 o AC Cameron amp PK rivedi ndash Microeconometrics Methods and Applications (Cambridge

Cambridge University Press 2005) or a standard textbook treatment o estimation methods or panel data models

19 he estimation was carried out using the xtreg procedure in Stata Version 12 he methods o constructing the

sandwich and bootstrap estimates o the variance-covariance matrix are described in Chapters 8 amp 13 o AC

Cameron amp PK rivedi ndash Microeconometrics using Stata (Revised Edition College Station exas Stata Press 2010)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 22: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2252

20 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

year (883 MW in 2008) so it seems rather unlikely that even the lower end o DECCrsquos range or

onshore wind generation will actually be realised

34 O course this assessment would be different i UK onshore wind operators were able to match

the Danish age-perormance curve here are many actors which differentiate the UK and

Danish wind industries he average scale o wind arms in the UK is larger they are moreremote and there appears to be significantly greater variation between operating and potential

new sites in achieved perormance It would thereore require a very optimistic observer to

conclude that the difference in age-perormance curves between the two countries will disap-

pear in the short or medium term

Figure 7 Projected load factors for UK offshore wind in 2020

20

22

24

26

28

30

32

34

18 GW ndash life 15 yrs 18 GW ndash life 20 yrs 24 GW ndash life 15 yrs 24 GW ndash life 20 yrs

L o a d f a c t o r ( )

2007ndash11 2015 2020

Note Based on Danish age-perormance curves Source Authorrsquos estimates

35 Figure 7 shows the results rom carrying out a similar exercise or offshore wind arms in the

UK he only evidence on the age-perormance curves or offshore installations is based upon

the Danish experience so these have been used to construct the projections Further the profile

o investment assumes a progressive increase in additions to capacity rom 2013 to 2016 and

then a stable level o investment rom 2017 to 2020 he DECC road map involves a very large

commitment to offshore wind investment ndash in the range 25-35 GW o new capacity annually

rom 2016 onwards As a consequence the average age o offshore capacity in 2020 will be quite

low and the ull effects o the age-perormance curves will only be elt during the decade ollow-

ing 2020 Even so the average normalised load actor in 2020 will be about 15 lower than or

2007ndash11 while by 2025 it will be about 30 below the base level

36 On these projections 232 GW o offshore capacity is the minimum necessary to generate 58

Wh o electricity rom offshore wind in 2020 his requirement will increase to 298 GW by

2025 simply to offset the effect o the ageing o offshore generation capacity on total output

37 Combining the estimates or onshore and offshore wind generation the total amount o wind

capacity required in 2020 to produce 90 Wh o electricity rom wind will be in the range 39

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 21

to 41 GW once allowance is made or the observed age-perormance curves or both onshore

and offshore installations hese estimates are roughly a third higher than official predictions o

28ndash31 GW o wind capacity in 2020

38 urning the figures round gives a slightly different picture o the reality o the governmentrsquos

strategy on renewable electricity generation A total o 30 GW o wind capacity in 2020 (12GW onshore 18 GW offshore) will generate about 64 Wh o electricity in a normal year

Since electricity generation rom wind and biomass are expected to account or 50 o renew-

able energy in 2020 the consequence is that biomass is expected to account or 53 Wh In

2011 landfill gas and sewage sludge digestion accounted or 44 o electricity rom bioenergy

(DUKES ables 64) here are limited opportunities to expand production rom these sources

so a continuation o historic rates o growth would mean that renewable generation rom other

sources o bioenergy would have to increase rom 72 Wh in 2011 to 46 Wh in 2020 here

are severe constraints on increasing the amounts o electricity generated rom animal biomass

biodegradable municipal solid waste and anaerobic digestion so it would be optimistic to

assume that these will contribute more than 6-8 Wh in 2020 (up rom 26 Wh in 2011)

39 In practice i the projections or renewable energy are to be believed they rest upon an assump-

tion that there will be a large increase in the amount o electricity that is generated rom plant

biomass ndash either on its own or through co-firing with coal his increase will involve increas-

ing the amount o electricity generated in these ways rom 46 Wh in 2011 to 38-40 Wh in

2020 Since the amounts o straw and UK-grown timber available or this purpose are strictly

limited this will come down to the amount o wood chips that can be imported and the cost o

doing this40 he historic average load actor or biomass plants is just over 60 so about 7500 MW o

biomass capacity will be required to supply 40 Wh o electricity he weight o wood chips per

MWh o electricity depends upon moisture content storage boiler design and other actors but

an indicative range is 085 to 115 tonnes per MWh o wood chips with a moisture content o

30 According to DECC and Forestry Commission estimates the UK imported 265 million

tonnes o steam coal and 26 million tonnes o other wood products (including wood chips)

in 2011 It seems that the net effect o the governmentrsquos strategy will be to replace 15 million

tonnes o coal imported in 2011 by 32 to 44 million tonnes o imported wood chips his is a

somewhat strange way o increasing energy security

Conclusion

41 Wind power is a highly capital-intensive technology or generating electricity Its merits rely

entirely upon a substitution o capital or uel inputs he same is true or hydro or tidal or wave

power In comparison with hydro power wind is a low quality resource because o its variability

and because it cannot be stored So the economic case or wind power must rest on obtaining

the most out o the wind that is available

42 While the decline in the achieved perormance o onshore wind turbines in Denmark is

much less than that or the UK or offshore nonetheless the decline in expected output under

standardised wind conditions over 10 years is 10 unweighted and 13 capacity weighted

hese declines accelerate afer age 10 so that the reductions in perormance are 17 and 20

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2452

22 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

respectively afer 15 years For UK onshore wind arms the reduction in perormance due to

age is much worse at 27 unweighted and 69 capacity weighted by age 10

43 Evidence on the perormance o Danish offshore installations is both restricted and so poor

that there may be concern that the results are affected by a small number o outliers Still the

sample contains a reasonable number o sites with at least 5 years o operating experience andthe decline in perormance by age 5 is 38 unweighted and 26 capacity weighted

44 In addition to these results there is strong evidence that the average normalised load actor or

new onshore wind installations in the UK has allen significantly over the period rom 2000 to

2011 his is consistent with a pattern in which the most avourable sites are developed first

Equally it could mean that wind developers have been unable to keep up with the rate o new

investment while maintaining the quality o development and operations For example the site

design or selection o turbine characteristics may make less effective use o the available wind

resources or the sites available than was the case in the past

45 Whatever the reasons the deterioration in initial perormance means that the expected returns

rom the expansion in wind capacity both or investors and in terms o the reduction in CO983090

emissions have been alling without a concomitant decrease in the private and social costs

that are borne by customers and the general public Clearly this is unsatisactory at best and it

suggests that the benefits claimed or current policies cannot be taken at ace value

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 23

Appendix Data and Methods

A Data for the United Kingdom

he raw data used or this study was extracted by the Renewable Energy Foundation rom the

Renewables and CHP Register database compiled by Ogem12 his inormation is used in the

administration o the market in Renewable Obligation Certificates (ROCs) he Renewables

Obligation is the primary mechanism by which large scale generators o renewable electric-

ity receive subsidies so the operators o wind arms have a very strong incentive to submit

complete and up-to-date inormation to the database he data extracted covered all onshore

wind generators with at least 05 MW o generating capacity which are eligible or ROCs or the

10 year period April 2002 to March 2012 he key variables are the monthly output o electric-

ity (used or the purpose o allocating ROCs) and nominal generating capacity here are 282

reporting units in the final dataset A reporting unit may correspond to an entire wind arm orto separate phases o development Complications arise when an existing wind arm is re-pow-

ered ie when old wind turbines are replaced by newer and usually more powerul turbines

he database reports this as a change in the generating capacity o the recording unit but in

this analysis repowering is treated as the termination o the old recording unit and the creation

o a new recording unit

A considerable amount o data checking and cleaning was required beore the data could

be analysed he first step was to add the age o each reporting unit his was calculated by

reerence to the month in which the turbines were commissioned For wind arms in operation

in or beore June 2002 the commissioning date was obtained rom various online sources ndash

notably the thewindpowernet database o UK wind arms13 which was cross-checked against

the inormation provided on the websites o wind operators and other sources For recording

units whose first data relates to dates afer June 2002 the commissioning date was initially

assumed to be the month or which data is first reported Cross-checks were carried out or

all recording units which commenced reporting in the period July 2002 to December 2003 In

more than 80 o cases the externally-reported commissioning date is within one month o the

first month or which data is reported he exceptions ndash Bu Farm and Mablethorpe ndash appear to

be due to delays by small operators in reporting data to Ogem

In a number o cases it is clear rom the data that plants were not in ull operation at the

time when data was first reported to the database here are two indicators o partial or delayed

commissioning

bull For an initial series o months the calculated load actor is very low ndash usually below 5 ndash

and then increases abruptly In such cases the data or the early months is dropped and the

commissioning date is treated as the month immediately preceding the first month with a

lsquonormalrsquo load actor14

12 httpswwwrenewablesandchpogemgovuk

13 See httpwwwthewindpowernetwindarms_list_enphp

14 he month in which a plant is reported as having been commissioned is not included in the analysis

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2652

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 25

the decline in the plantrsquos load actor as it ages As an example consider a plant whose reported

capacity was increased rom 99 MW to 10 MW at age 3 I this was simply a change in report-

ing practice this would reduce the estimated load actor rom age 3 onwards By increasing

the initial capacity rom 99 MW to 10 MW ndash or by reducing the capacity rom 10 MW to 99

MW rom age 3 onwards ndash this bias is removed It is not important which adjustment is madebecause the difference between the alternative assumptions is captured by the unit fixed effect

he numbers o separate sites and o observations per site mean that it is not possible to

provide a simple visual representation o the data Figure 8 provides a summary by illustrating

the distributions o load actors by the age o the wind arm pooled across all years he ranges

o the load actors or each age are wide while the medians show no clear trend by age hus a

naiumlve analysis might conclude that there is nothing to investigate However as will be explained

below the distributions mask crucial differences between the perormance o wind arms as

they age because they do not control or the differences in location and wind availability over

time

B Data for Denmark

he data or Danish wind arms used in this study comes rom a database compiled by the

Danish Energy Agency covering the characteristics and perormance o all wind turbines rom

2002 up to the end o August 201216 he basic recording unit or the register is the individual

wind turbine but where a number o turbines are connected to a single meter an identical

average output per turbine is recorded or all turbines in the group he presence o what are in

effect repeated observations increases the noise in the data In addition the model that is used

or the analysis assumes that site operator and turbine characteristics are common or all o theunits that make up a single wind arm o address this point separate wind arms have been

identified by grouping all turbines which have identical values or turbine manuacturer and

type local authority date o commissioning and date o decommissioning (where applicable)

he register contains inormation on 7569 turbines commissioned rom 1977 onwards his

includes a large number o small turbines which would not all within the scope o the ROC

scheme in the UK Hence or comparability with the UK data all turbines commissioned

prior to 1992 were dropped as well as all wind arms with a generating capacity o less than 1

MW he resulting sample covered 823 onshore wind arms and 30 offshore wind arms with

monthly load actors or various periods rom January 2002 to August 2012 Given the cut-off

or wind arms commissioned prior to 1992 the maximum age or an onshore wind arm was 20

years and the maximum age or an offshore wind arm was 17 years

he total capacity o the onshore wind arms recorded in the Danish database was 2757 MW

he average size o an onshore wind arm in Denmark is small ndash only 33 MW as compared

with 161 MW or the UK ndash because 84 o the sample consists o installations with between

1 and 4 turbines he largest onshore installation has 39 turbines with a capacity o 234 MW

while the onshore installation with the greatest capacity has 19 turbines with a capacity o 584

MW Overall the sample o Danish onshore wind arms is older and smaller than the sample

o UK onshore wind arms reflecting the much longer history o wind generation in Denmark

16 httpwwwensdkEN-USINFOFACSANDFIGURESENERGY_SAISICS_AND_INDICAORS

OVERVIEWOFHEENERGYSECORREGISEROFWINDURBINESSiderForsideaspx

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2852

26 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

he larger scale o UK wind arms may account or some o the differences in perormance

curves or the two countries

he sample o offshore wind arms in Denmark is quite small with 30 installations in

total Still it is useul to analyse the perormance o these installations since there is very little

comparable data or the UK he sample covers 394 turbines with an average o just over 13turbines per installation and an average capacity o 288 MW he largest installation which

was commissioned in 2003 has 72 turbines with a total capacity o 1656 MW Again offshore

installations are smaller and older in Denmark than is the current pattern in the UK he first

two wind arms in the sample were commissioned in the years 1995 and 2000 but the sample

increases significantly rom 2002 onwards o avoid reliance upon a sample o 1 or 2 wind

arms to determine specific coefficients or ages gt 10 years all observations with an age greater

than 10 were omitted rom the analysis17

A somewhat different problem with the offshore wind arm data arose because the coeffi-

cients estimated using capacity weights (see below) were heavily influenced by the largest windarm his has had an extremely erratic profile o perormance over time with a very low average

load actor he treatment o this wind arm is discussed in more detail in Section F below

C Specication and estimation methods

As Figure 8 illustrates it is necessary to go beyond simple summary statistics in order to assess

whether there is any systematic relationship between age and the average perormance o wind

arms he perormance o wind arms varies over both time and space some months and years

have more or less wind than the long term average while specific characteristics o wind arms

ndash including location type o turbine and operating regime ndash will influence the perormance oeach plant under identical wind conditions

he combination o (a) site-specific characteristics which are constant over time and (b)

period-specific characteristics which are constant across wind arms can be represented by

what statisticians call an error components model with fixed effects or each wind arm and

each time period he dependent variable in the model consists o a sequence o load actors ndash

denoted by or unit i in month t ndash or each wind arm (the panel unit) over time he load

actors are calculated by dividing the total output in MWh by (24 x number o days in month x

reported generating capacity in MW) and are multiplied by 100 to convert to percentages

he additive version o the error components model may be written as

(1)

in which denotes the age o plant i in period t and is some unction that is either a

representation o or an approximation to the normalised age-perormance relationship he

multiplicative version is similar except that is replaced by where ln() denotes the

natural logarithm he terms are wind arm or unit fixed effects which capture actors such

as location type o turbine operating regimes etc he terms are period fixed effects which

capture the influence o wind conditions over the UK or Denmark as a whole as well as seasonal

maintenance and demand Without loss o generality it may be assumed that the fixed effects are

17 For the avoidance o doubt this means that no attempt was made to estimate the effect o age on the perormance o

offshore wind arms beyond age 10 while the data or the oldest wind arms was included in the analysis up to and

including age 10

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 27

normalised so that ie all o the error components have a zero mean while the

mean o the random error is zero by definition he terms are random errors which capture

random variations in wind conditions that are not specific to the site turbine breakdowns and

other actors that are uncorrelated with either the site or the time period It is assumed that the

mean o the random error is zero while additional assumptions depend upon the method o

estimation that is adopted

he model specified in equation (1) can be estimated in a variety o ways18 A crucial issue is

whether the are or may be significantly correlated with the age terms or any other independent

regressors I it is assumed that there is no such correlation the model is usually reerred to as

a random effects model In that case the equation can be estimated by a variant o least squares

(LS) generalised least squares (GLS) generalised methods o moments (GMM) or maximum

likelihood (ML) ndash all o which should be yield parameter estimates that will converge in prob-

ability to the true parameter values as the sample size increases An alternative but less restric-

tive specification is known as the fixed effects model in which no assumption is made about

the correlation between the unit fixed effects and other regressors he limitation o the fixed

effects model is that it cannot be used to estimate parameters or regressors that vary across

panel units but are constant over time or a panel unit (time-invariant regressors) ndash eg the

year in which a wind arm was commissioned its location or its rated generating capacity he

influence o such variables is captured by the unit fixed effects

In order to avoid the risk o obtaining biased estimates o the age effects the results reported

in this note are based upon estimates using the fixed effects model he influence o time-invar-

iant regressors is examined in a second stage by regressing the estimates o the unit fixed effectson these variables he simplest method o estimating the fixed effects model is known as the

lsquowithinrsquo estimator in which the mean value or each unit is subtracted rom all observations or

that unit For example i is a general polynomial o order M in A equation (1) becomes

(2)

so that the model can be estimated by least square with the inclusion o dummy variables or

each period t he variance-covariance matrix o the coefficients have been estimated by using

(a) a robust sandwich estimator adjusted or clustering by panel unit and (b) a bootstrap esti-

mator with 400 repetitions he estimates o the standard errors will be consistent even i the

random error component is serially correlated within panel units andor has different variances

across panel units19 Since the number o panel units and the average number o time periods

per panel units are both large (other than or Danish offshore wind arms) the least square esti-

mates o the parameters will be unbiased under these assumptions

he bootstrap method is expensive to carry out but it provides a useul cross-check on

the standard errors generated by other methods Unortunately the bootstrap method o

18 See Chapter 21 o AC Cameron amp PK rivedi ndash Microeconometrics Methods and Applications (Cambridge

Cambridge University Press 2005) or a standard textbook treatment o estimation methods or panel data models

19 he estimation was carried out using the xtreg procedure in Stata Version 12 he methods o constructing the

sandwich and bootstrap estimates o the variance-covariance matrix are described in Chapters 8 amp 13 o AC

Cameron amp PK rivedi ndash Microeconometrics using Stata (Revised Edition College Station exas Stata Press 2010)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 23: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 21

to 41 GW once allowance is made or the observed age-perormance curves or both onshore

and offshore installations hese estimates are roughly a third higher than official predictions o

28ndash31 GW o wind capacity in 2020

38 urning the figures round gives a slightly different picture o the reality o the governmentrsquos

strategy on renewable electricity generation A total o 30 GW o wind capacity in 2020 (12GW onshore 18 GW offshore) will generate about 64 Wh o electricity in a normal year

Since electricity generation rom wind and biomass are expected to account or 50 o renew-

able energy in 2020 the consequence is that biomass is expected to account or 53 Wh In

2011 landfill gas and sewage sludge digestion accounted or 44 o electricity rom bioenergy

(DUKES ables 64) here are limited opportunities to expand production rom these sources

so a continuation o historic rates o growth would mean that renewable generation rom other

sources o bioenergy would have to increase rom 72 Wh in 2011 to 46 Wh in 2020 here

are severe constraints on increasing the amounts o electricity generated rom animal biomass

biodegradable municipal solid waste and anaerobic digestion so it would be optimistic to

assume that these will contribute more than 6-8 Wh in 2020 (up rom 26 Wh in 2011)

39 In practice i the projections or renewable energy are to be believed they rest upon an assump-

tion that there will be a large increase in the amount o electricity that is generated rom plant

biomass ndash either on its own or through co-firing with coal his increase will involve increas-

ing the amount o electricity generated in these ways rom 46 Wh in 2011 to 38-40 Wh in

2020 Since the amounts o straw and UK-grown timber available or this purpose are strictly

limited this will come down to the amount o wood chips that can be imported and the cost o

doing this40 he historic average load actor or biomass plants is just over 60 so about 7500 MW o

biomass capacity will be required to supply 40 Wh o electricity he weight o wood chips per

MWh o electricity depends upon moisture content storage boiler design and other actors but

an indicative range is 085 to 115 tonnes per MWh o wood chips with a moisture content o

30 According to DECC and Forestry Commission estimates the UK imported 265 million

tonnes o steam coal and 26 million tonnes o other wood products (including wood chips)

in 2011 It seems that the net effect o the governmentrsquos strategy will be to replace 15 million

tonnes o coal imported in 2011 by 32 to 44 million tonnes o imported wood chips his is a

somewhat strange way o increasing energy security

Conclusion

41 Wind power is a highly capital-intensive technology or generating electricity Its merits rely

entirely upon a substitution o capital or uel inputs he same is true or hydro or tidal or wave

power In comparison with hydro power wind is a low quality resource because o its variability

and because it cannot be stored So the economic case or wind power must rest on obtaining

the most out o the wind that is available

42 While the decline in the achieved perormance o onshore wind turbines in Denmark is

much less than that or the UK or offshore nonetheless the decline in expected output under

standardised wind conditions over 10 years is 10 unweighted and 13 capacity weighted

hese declines accelerate afer age 10 so that the reductions in perormance are 17 and 20

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2452

22 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

respectively afer 15 years For UK onshore wind arms the reduction in perormance due to

age is much worse at 27 unweighted and 69 capacity weighted by age 10

43 Evidence on the perormance o Danish offshore installations is both restricted and so poor

that there may be concern that the results are affected by a small number o outliers Still the

sample contains a reasonable number o sites with at least 5 years o operating experience andthe decline in perormance by age 5 is 38 unweighted and 26 capacity weighted

44 In addition to these results there is strong evidence that the average normalised load actor or

new onshore wind installations in the UK has allen significantly over the period rom 2000 to

2011 his is consistent with a pattern in which the most avourable sites are developed first

Equally it could mean that wind developers have been unable to keep up with the rate o new

investment while maintaining the quality o development and operations For example the site

design or selection o turbine characteristics may make less effective use o the available wind

resources or the sites available than was the case in the past

45 Whatever the reasons the deterioration in initial perormance means that the expected returns

rom the expansion in wind capacity both or investors and in terms o the reduction in CO983090

emissions have been alling without a concomitant decrease in the private and social costs

that are borne by customers and the general public Clearly this is unsatisactory at best and it

suggests that the benefits claimed or current policies cannot be taken at ace value

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 23

Appendix Data and Methods

A Data for the United Kingdom

he raw data used or this study was extracted by the Renewable Energy Foundation rom the

Renewables and CHP Register database compiled by Ogem12 his inormation is used in the

administration o the market in Renewable Obligation Certificates (ROCs) he Renewables

Obligation is the primary mechanism by which large scale generators o renewable electric-

ity receive subsidies so the operators o wind arms have a very strong incentive to submit

complete and up-to-date inormation to the database he data extracted covered all onshore

wind generators with at least 05 MW o generating capacity which are eligible or ROCs or the

10 year period April 2002 to March 2012 he key variables are the monthly output o electric-

ity (used or the purpose o allocating ROCs) and nominal generating capacity here are 282

reporting units in the final dataset A reporting unit may correspond to an entire wind arm orto separate phases o development Complications arise when an existing wind arm is re-pow-

ered ie when old wind turbines are replaced by newer and usually more powerul turbines

he database reports this as a change in the generating capacity o the recording unit but in

this analysis repowering is treated as the termination o the old recording unit and the creation

o a new recording unit

A considerable amount o data checking and cleaning was required beore the data could

be analysed he first step was to add the age o each reporting unit his was calculated by

reerence to the month in which the turbines were commissioned For wind arms in operation

in or beore June 2002 the commissioning date was obtained rom various online sources ndash

notably the thewindpowernet database o UK wind arms13 which was cross-checked against

the inormation provided on the websites o wind operators and other sources For recording

units whose first data relates to dates afer June 2002 the commissioning date was initially

assumed to be the month or which data is first reported Cross-checks were carried out or

all recording units which commenced reporting in the period July 2002 to December 2003 In

more than 80 o cases the externally-reported commissioning date is within one month o the

first month or which data is reported he exceptions ndash Bu Farm and Mablethorpe ndash appear to

be due to delays by small operators in reporting data to Ogem

In a number o cases it is clear rom the data that plants were not in ull operation at the

time when data was first reported to the database here are two indicators o partial or delayed

commissioning

bull For an initial series o months the calculated load actor is very low ndash usually below 5 ndash

and then increases abruptly In such cases the data or the early months is dropped and the

commissioning date is treated as the month immediately preceding the first month with a

lsquonormalrsquo load actor14

12 httpswwwrenewablesandchpogemgovuk

13 See httpwwwthewindpowernetwindarms_list_enphp

14 he month in which a plant is reported as having been commissioned is not included in the analysis

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2652

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 25

the decline in the plantrsquos load actor as it ages As an example consider a plant whose reported

capacity was increased rom 99 MW to 10 MW at age 3 I this was simply a change in report-

ing practice this would reduce the estimated load actor rom age 3 onwards By increasing

the initial capacity rom 99 MW to 10 MW ndash or by reducing the capacity rom 10 MW to 99

MW rom age 3 onwards ndash this bias is removed It is not important which adjustment is madebecause the difference between the alternative assumptions is captured by the unit fixed effect

he numbers o separate sites and o observations per site mean that it is not possible to

provide a simple visual representation o the data Figure 8 provides a summary by illustrating

the distributions o load actors by the age o the wind arm pooled across all years he ranges

o the load actors or each age are wide while the medians show no clear trend by age hus a

naiumlve analysis might conclude that there is nothing to investigate However as will be explained

below the distributions mask crucial differences between the perormance o wind arms as

they age because they do not control or the differences in location and wind availability over

time

B Data for Denmark

he data or Danish wind arms used in this study comes rom a database compiled by the

Danish Energy Agency covering the characteristics and perormance o all wind turbines rom

2002 up to the end o August 201216 he basic recording unit or the register is the individual

wind turbine but where a number o turbines are connected to a single meter an identical

average output per turbine is recorded or all turbines in the group he presence o what are in

effect repeated observations increases the noise in the data In addition the model that is used

or the analysis assumes that site operator and turbine characteristics are common or all o theunits that make up a single wind arm o address this point separate wind arms have been

identified by grouping all turbines which have identical values or turbine manuacturer and

type local authority date o commissioning and date o decommissioning (where applicable)

he register contains inormation on 7569 turbines commissioned rom 1977 onwards his

includes a large number o small turbines which would not all within the scope o the ROC

scheme in the UK Hence or comparability with the UK data all turbines commissioned

prior to 1992 were dropped as well as all wind arms with a generating capacity o less than 1

MW he resulting sample covered 823 onshore wind arms and 30 offshore wind arms with

monthly load actors or various periods rom January 2002 to August 2012 Given the cut-off

or wind arms commissioned prior to 1992 the maximum age or an onshore wind arm was 20

years and the maximum age or an offshore wind arm was 17 years

he total capacity o the onshore wind arms recorded in the Danish database was 2757 MW

he average size o an onshore wind arm in Denmark is small ndash only 33 MW as compared

with 161 MW or the UK ndash because 84 o the sample consists o installations with between

1 and 4 turbines he largest onshore installation has 39 turbines with a capacity o 234 MW

while the onshore installation with the greatest capacity has 19 turbines with a capacity o 584

MW Overall the sample o Danish onshore wind arms is older and smaller than the sample

o UK onshore wind arms reflecting the much longer history o wind generation in Denmark

16 httpwwwensdkEN-USINFOFACSANDFIGURESENERGY_SAISICS_AND_INDICAORS

OVERVIEWOFHEENERGYSECORREGISEROFWINDURBINESSiderForsideaspx

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2852

26 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

he larger scale o UK wind arms may account or some o the differences in perormance

curves or the two countries

he sample o offshore wind arms in Denmark is quite small with 30 installations in

total Still it is useul to analyse the perormance o these installations since there is very little

comparable data or the UK he sample covers 394 turbines with an average o just over 13turbines per installation and an average capacity o 288 MW he largest installation which

was commissioned in 2003 has 72 turbines with a total capacity o 1656 MW Again offshore

installations are smaller and older in Denmark than is the current pattern in the UK he first

two wind arms in the sample were commissioned in the years 1995 and 2000 but the sample

increases significantly rom 2002 onwards o avoid reliance upon a sample o 1 or 2 wind

arms to determine specific coefficients or ages gt 10 years all observations with an age greater

than 10 were omitted rom the analysis17

A somewhat different problem with the offshore wind arm data arose because the coeffi-

cients estimated using capacity weights (see below) were heavily influenced by the largest windarm his has had an extremely erratic profile o perormance over time with a very low average

load actor he treatment o this wind arm is discussed in more detail in Section F below

C Specication and estimation methods

As Figure 8 illustrates it is necessary to go beyond simple summary statistics in order to assess

whether there is any systematic relationship between age and the average perormance o wind

arms he perormance o wind arms varies over both time and space some months and years

have more or less wind than the long term average while specific characteristics o wind arms

ndash including location type o turbine and operating regime ndash will influence the perormance oeach plant under identical wind conditions

he combination o (a) site-specific characteristics which are constant over time and (b)

period-specific characteristics which are constant across wind arms can be represented by

what statisticians call an error components model with fixed effects or each wind arm and

each time period he dependent variable in the model consists o a sequence o load actors ndash

denoted by or unit i in month t ndash or each wind arm (the panel unit) over time he load

actors are calculated by dividing the total output in MWh by (24 x number o days in month x

reported generating capacity in MW) and are multiplied by 100 to convert to percentages

he additive version o the error components model may be written as

(1)

in which denotes the age o plant i in period t and is some unction that is either a

representation o or an approximation to the normalised age-perormance relationship he

multiplicative version is similar except that is replaced by where ln() denotes the

natural logarithm he terms are wind arm or unit fixed effects which capture actors such

as location type o turbine operating regimes etc he terms are period fixed effects which

capture the influence o wind conditions over the UK or Denmark as a whole as well as seasonal

maintenance and demand Without loss o generality it may be assumed that the fixed effects are

17 For the avoidance o doubt this means that no attempt was made to estimate the effect o age on the perormance o

offshore wind arms beyond age 10 while the data or the oldest wind arms was included in the analysis up to and

including age 10

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 27

normalised so that ie all o the error components have a zero mean while the

mean o the random error is zero by definition he terms are random errors which capture

random variations in wind conditions that are not specific to the site turbine breakdowns and

other actors that are uncorrelated with either the site or the time period It is assumed that the

mean o the random error is zero while additional assumptions depend upon the method o

estimation that is adopted

he model specified in equation (1) can be estimated in a variety o ways18 A crucial issue is

whether the are or may be significantly correlated with the age terms or any other independent

regressors I it is assumed that there is no such correlation the model is usually reerred to as

a random effects model In that case the equation can be estimated by a variant o least squares

(LS) generalised least squares (GLS) generalised methods o moments (GMM) or maximum

likelihood (ML) ndash all o which should be yield parameter estimates that will converge in prob-

ability to the true parameter values as the sample size increases An alternative but less restric-

tive specification is known as the fixed effects model in which no assumption is made about

the correlation between the unit fixed effects and other regressors he limitation o the fixed

effects model is that it cannot be used to estimate parameters or regressors that vary across

panel units but are constant over time or a panel unit (time-invariant regressors) ndash eg the

year in which a wind arm was commissioned its location or its rated generating capacity he

influence o such variables is captured by the unit fixed effects

In order to avoid the risk o obtaining biased estimates o the age effects the results reported

in this note are based upon estimates using the fixed effects model he influence o time-invar-

iant regressors is examined in a second stage by regressing the estimates o the unit fixed effectson these variables he simplest method o estimating the fixed effects model is known as the

lsquowithinrsquo estimator in which the mean value or each unit is subtracted rom all observations or

that unit For example i is a general polynomial o order M in A equation (1) becomes

(2)

so that the model can be estimated by least square with the inclusion o dummy variables or

each period t he variance-covariance matrix o the coefficients have been estimated by using

(a) a robust sandwich estimator adjusted or clustering by panel unit and (b) a bootstrap esti-

mator with 400 repetitions he estimates o the standard errors will be consistent even i the

random error component is serially correlated within panel units andor has different variances

across panel units19 Since the number o panel units and the average number o time periods

per panel units are both large (other than or Danish offshore wind arms) the least square esti-

mates o the parameters will be unbiased under these assumptions

he bootstrap method is expensive to carry out but it provides a useul cross-check on

the standard errors generated by other methods Unortunately the bootstrap method o

18 See Chapter 21 o AC Cameron amp PK rivedi ndash Microeconometrics Methods and Applications (Cambridge

Cambridge University Press 2005) or a standard textbook treatment o estimation methods or panel data models

19 he estimation was carried out using the xtreg procedure in Stata Version 12 he methods o constructing the

sandwich and bootstrap estimates o the variance-covariance matrix are described in Chapters 8 amp 13 o AC

Cameron amp PK rivedi ndash Microeconometrics using Stata (Revised Edition College Station exas Stata Press 2010)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 24: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2452

22 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

respectively afer 15 years For UK onshore wind arms the reduction in perormance due to

age is much worse at 27 unweighted and 69 capacity weighted by age 10

43 Evidence on the perormance o Danish offshore installations is both restricted and so poor

that there may be concern that the results are affected by a small number o outliers Still the

sample contains a reasonable number o sites with at least 5 years o operating experience andthe decline in perormance by age 5 is 38 unweighted and 26 capacity weighted

44 In addition to these results there is strong evidence that the average normalised load actor or

new onshore wind installations in the UK has allen significantly over the period rom 2000 to

2011 his is consistent with a pattern in which the most avourable sites are developed first

Equally it could mean that wind developers have been unable to keep up with the rate o new

investment while maintaining the quality o development and operations For example the site

design or selection o turbine characteristics may make less effective use o the available wind

resources or the sites available than was the case in the past

45 Whatever the reasons the deterioration in initial perormance means that the expected returns

rom the expansion in wind capacity both or investors and in terms o the reduction in CO983090

emissions have been alling without a concomitant decrease in the private and social costs

that are borne by customers and the general public Clearly this is unsatisactory at best and it

suggests that the benefits claimed or current policies cannot be taken at ace value

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 23

Appendix Data and Methods

A Data for the United Kingdom

he raw data used or this study was extracted by the Renewable Energy Foundation rom the

Renewables and CHP Register database compiled by Ogem12 his inormation is used in the

administration o the market in Renewable Obligation Certificates (ROCs) he Renewables

Obligation is the primary mechanism by which large scale generators o renewable electric-

ity receive subsidies so the operators o wind arms have a very strong incentive to submit

complete and up-to-date inormation to the database he data extracted covered all onshore

wind generators with at least 05 MW o generating capacity which are eligible or ROCs or the

10 year period April 2002 to March 2012 he key variables are the monthly output o electric-

ity (used or the purpose o allocating ROCs) and nominal generating capacity here are 282

reporting units in the final dataset A reporting unit may correspond to an entire wind arm orto separate phases o development Complications arise when an existing wind arm is re-pow-

ered ie when old wind turbines are replaced by newer and usually more powerul turbines

he database reports this as a change in the generating capacity o the recording unit but in

this analysis repowering is treated as the termination o the old recording unit and the creation

o a new recording unit

A considerable amount o data checking and cleaning was required beore the data could

be analysed he first step was to add the age o each reporting unit his was calculated by

reerence to the month in which the turbines were commissioned For wind arms in operation

in or beore June 2002 the commissioning date was obtained rom various online sources ndash

notably the thewindpowernet database o UK wind arms13 which was cross-checked against

the inormation provided on the websites o wind operators and other sources For recording

units whose first data relates to dates afer June 2002 the commissioning date was initially

assumed to be the month or which data is first reported Cross-checks were carried out or

all recording units which commenced reporting in the period July 2002 to December 2003 In

more than 80 o cases the externally-reported commissioning date is within one month o the

first month or which data is reported he exceptions ndash Bu Farm and Mablethorpe ndash appear to

be due to delays by small operators in reporting data to Ogem

In a number o cases it is clear rom the data that plants were not in ull operation at the

time when data was first reported to the database here are two indicators o partial or delayed

commissioning

bull For an initial series o months the calculated load actor is very low ndash usually below 5 ndash

and then increases abruptly In such cases the data or the early months is dropped and the

commissioning date is treated as the month immediately preceding the first month with a

lsquonormalrsquo load actor14

12 httpswwwrenewablesandchpogemgovuk

13 See httpwwwthewindpowernetwindarms_list_enphp

14 he month in which a plant is reported as having been commissioned is not included in the analysis

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2652

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 25

the decline in the plantrsquos load actor as it ages As an example consider a plant whose reported

capacity was increased rom 99 MW to 10 MW at age 3 I this was simply a change in report-

ing practice this would reduce the estimated load actor rom age 3 onwards By increasing

the initial capacity rom 99 MW to 10 MW ndash or by reducing the capacity rom 10 MW to 99

MW rom age 3 onwards ndash this bias is removed It is not important which adjustment is madebecause the difference between the alternative assumptions is captured by the unit fixed effect

he numbers o separate sites and o observations per site mean that it is not possible to

provide a simple visual representation o the data Figure 8 provides a summary by illustrating

the distributions o load actors by the age o the wind arm pooled across all years he ranges

o the load actors or each age are wide while the medians show no clear trend by age hus a

naiumlve analysis might conclude that there is nothing to investigate However as will be explained

below the distributions mask crucial differences between the perormance o wind arms as

they age because they do not control or the differences in location and wind availability over

time

B Data for Denmark

he data or Danish wind arms used in this study comes rom a database compiled by the

Danish Energy Agency covering the characteristics and perormance o all wind turbines rom

2002 up to the end o August 201216 he basic recording unit or the register is the individual

wind turbine but where a number o turbines are connected to a single meter an identical

average output per turbine is recorded or all turbines in the group he presence o what are in

effect repeated observations increases the noise in the data In addition the model that is used

or the analysis assumes that site operator and turbine characteristics are common or all o theunits that make up a single wind arm o address this point separate wind arms have been

identified by grouping all turbines which have identical values or turbine manuacturer and

type local authority date o commissioning and date o decommissioning (where applicable)

he register contains inormation on 7569 turbines commissioned rom 1977 onwards his

includes a large number o small turbines which would not all within the scope o the ROC

scheme in the UK Hence or comparability with the UK data all turbines commissioned

prior to 1992 were dropped as well as all wind arms with a generating capacity o less than 1

MW he resulting sample covered 823 onshore wind arms and 30 offshore wind arms with

monthly load actors or various periods rom January 2002 to August 2012 Given the cut-off

or wind arms commissioned prior to 1992 the maximum age or an onshore wind arm was 20

years and the maximum age or an offshore wind arm was 17 years

he total capacity o the onshore wind arms recorded in the Danish database was 2757 MW

he average size o an onshore wind arm in Denmark is small ndash only 33 MW as compared

with 161 MW or the UK ndash because 84 o the sample consists o installations with between

1 and 4 turbines he largest onshore installation has 39 turbines with a capacity o 234 MW

while the onshore installation with the greatest capacity has 19 turbines with a capacity o 584

MW Overall the sample o Danish onshore wind arms is older and smaller than the sample

o UK onshore wind arms reflecting the much longer history o wind generation in Denmark

16 httpwwwensdkEN-USINFOFACSANDFIGURESENERGY_SAISICS_AND_INDICAORS

OVERVIEWOFHEENERGYSECORREGISEROFWINDURBINESSiderForsideaspx

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2852

26 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

he larger scale o UK wind arms may account or some o the differences in perormance

curves or the two countries

he sample o offshore wind arms in Denmark is quite small with 30 installations in

total Still it is useul to analyse the perormance o these installations since there is very little

comparable data or the UK he sample covers 394 turbines with an average o just over 13turbines per installation and an average capacity o 288 MW he largest installation which

was commissioned in 2003 has 72 turbines with a total capacity o 1656 MW Again offshore

installations are smaller and older in Denmark than is the current pattern in the UK he first

two wind arms in the sample were commissioned in the years 1995 and 2000 but the sample

increases significantly rom 2002 onwards o avoid reliance upon a sample o 1 or 2 wind

arms to determine specific coefficients or ages gt 10 years all observations with an age greater

than 10 were omitted rom the analysis17

A somewhat different problem with the offshore wind arm data arose because the coeffi-

cients estimated using capacity weights (see below) were heavily influenced by the largest windarm his has had an extremely erratic profile o perormance over time with a very low average

load actor he treatment o this wind arm is discussed in more detail in Section F below

C Specication and estimation methods

As Figure 8 illustrates it is necessary to go beyond simple summary statistics in order to assess

whether there is any systematic relationship between age and the average perormance o wind

arms he perormance o wind arms varies over both time and space some months and years

have more or less wind than the long term average while specific characteristics o wind arms

ndash including location type o turbine and operating regime ndash will influence the perormance oeach plant under identical wind conditions

he combination o (a) site-specific characteristics which are constant over time and (b)

period-specific characteristics which are constant across wind arms can be represented by

what statisticians call an error components model with fixed effects or each wind arm and

each time period he dependent variable in the model consists o a sequence o load actors ndash

denoted by or unit i in month t ndash or each wind arm (the panel unit) over time he load

actors are calculated by dividing the total output in MWh by (24 x number o days in month x

reported generating capacity in MW) and are multiplied by 100 to convert to percentages

he additive version o the error components model may be written as

(1)

in which denotes the age o plant i in period t and is some unction that is either a

representation o or an approximation to the normalised age-perormance relationship he

multiplicative version is similar except that is replaced by where ln() denotes the

natural logarithm he terms are wind arm or unit fixed effects which capture actors such

as location type o turbine operating regimes etc he terms are period fixed effects which

capture the influence o wind conditions over the UK or Denmark as a whole as well as seasonal

maintenance and demand Without loss o generality it may be assumed that the fixed effects are

17 For the avoidance o doubt this means that no attempt was made to estimate the effect o age on the perormance o

offshore wind arms beyond age 10 while the data or the oldest wind arms was included in the analysis up to and

including age 10

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 27

normalised so that ie all o the error components have a zero mean while the

mean o the random error is zero by definition he terms are random errors which capture

random variations in wind conditions that are not specific to the site turbine breakdowns and

other actors that are uncorrelated with either the site or the time period It is assumed that the

mean o the random error is zero while additional assumptions depend upon the method o

estimation that is adopted

he model specified in equation (1) can be estimated in a variety o ways18 A crucial issue is

whether the are or may be significantly correlated with the age terms or any other independent

regressors I it is assumed that there is no such correlation the model is usually reerred to as

a random effects model In that case the equation can be estimated by a variant o least squares

(LS) generalised least squares (GLS) generalised methods o moments (GMM) or maximum

likelihood (ML) ndash all o which should be yield parameter estimates that will converge in prob-

ability to the true parameter values as the sample size increases An alternative but less restric-

tive specification is known as the fixed effects model in which no assumption is made about

the correlation between the unit fixed effects and other regressors he limitation o the fixed

effects model is that it cannot be used to estimate parameters or regressors that vary across

panel units but are constant over time or a panel unit (time-invariant regressors) ndash eg the

year in which a wind arm was commissioned its location or its rated generating capacity he

influence o such variables is captured by the unit fixed effects

In order to avoid the risk o obtaining biased estimates o the age effects the results reported

in this note are based upon estimates using the fixed effects model he influence o time-invar-

iant regressors is examined in a second stage by regressing the estimates o the unit fixed effectson these variables he simplest method o estimating the fixed effects model is known as the

lsquowithinrsquo estimator in which the mean value or each unit is subtracted rom all observations or

that unit For example i is a general polynomial o order M in A equation (1) becomes

(2)

so that the model can be estimated by least square with the inclusion o dummy variables or

each period t he variance-covariance matrix o the coefficients have been estimated by using

(a) a robust sandwich estimator adjusted or clustering by panel unit and (b) a bootstrap esti-

mator with 400 repetitions he estimates o the standard errors will be consistent even i the

random error component is serially correlated within panel units andor has different variances

across panel units19 Since the number o panel units and the average number o time periods

per panel units are both large (other than or Danish offshore wind arms) the least square esti-

mates o the parameters will be unbiased under these assumptions

he bootstrap method is expensive to carry out but it provides a useul cross-check on

the standard errors generated by other methods Unortunately the bootstrap method o

18 See Chapter 21 o AC Cameron amp PK rivedi ndash Microeconometrics Methods and Applications (Cambridge

Cambridge University Press 2005) or a standard textbook treatment o estimation methods or panel data models

19 he estimation was carried out using the xtreg procedure in Stata Version 12 he methods o constructing the

sandwich and bootstrap estimates o the variance-covariance matrix are described in Chapters 8 amp 13 o AC

Cameron amp PK rivedi ndash Microeconometrics using Stata (Revised Edition College Station exas Stata Press 2010)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 25: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 23

Appendix Data and Methods

A Data for the United Kingdom

he raw data used or this study was extracted by the Renewable Energy Foundation rom the

Renewables and CHP Register database compiled by Ogem12 his inormation is used in the

administration o the market in Renewable Obligation Certificates (ROCs) he Renewables

Obligation is the primary mechanism by which large scale generators o renewable electric-

ity receive subsidies so the operators o wind arms have a very strong incentive to submit

complete and up-to-date inormation to the database he data extracted covered all onshore

wind generators with at least 05 MW o generating capacity which are eligible or ROCs or the

10 year period April 2002 to March 2012 he key variables are the monthly output o electric-

ity (used or the purpose o allocating ROCs) and nominal generating capacity here are 282

reporting units in the final dataset A reporting unit may correspond to an entire wind arm orto separate phases o development Complications arise when an existing wind arm is re-pow-

ered ie when old wind turbines are replaced by newer and usually more powerul turbines

he database reports this as a change in the generating capacity o the recording unit but in

this analysis repowering is treated as the termination o the old recording unit and the creation

o a new recording unit

A considerable amount o data checking and cleaning was required beore the data could

be analysed he first step was to add the age o each reporting unit his was calculated by

reerence to the month in which the turbines were commissioned For wind arms in operation

in or beore June 2002 the commissioning date was obtained rom various online sources ndash

notably the thewindpowernet database o UK wind arms13 which was cross-checked against

the inormation provided on the websites o wind operators and other sources For recording

units whose first data relates to dates afer June 2002 the commissioning date was initially

assumed to be the month or which data is first reported Cross-checks were carried out or

all recording units which commenced reporting in the period July 2002 to December 2003 In

more than 80 o cases the externally-reported commissioning date is within one month o the

first month or which data is reported he exceptions ndash Bu Farm and Mablethorpe ndash appear to

be due to delays by small operators in reporting data to Ogem

In a number o cases it is clear rom the data that plants were not in ull operation at the

time when data was first reported to the database here are two indicators o partial or delayed

commissioning

bull For an initial series o months the calculated load actor is very low ndash usually below 5 ndash

and then increases abruptly In such cases the data or the early months is dropped and the

commissioning date is treated as the month immediately preceding the first month with a

lsquonormalrsquo load actor14

12 httpswwwrenewablesandchpogemgovuk

13 See httpwwwthewindpowernetwindarms_list_enphp

14 he month in which a plant is reported as having been commissioned is not included in the analysis

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2652

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 25

the decline in the plantrsquos load actor as it ages As an example consider a plant whose reported

capacity was increased rom 99 MW to 10 MW at age 3 I this was simply a change in report-

ing practice this would reduce the estimated load actor rom age 3 onwards By increasing

the initial capacity rom 99 MW to 10 MW ndash or by reducing the capacity rom 10 MW to 99

MW rom age 3 onwards ndash this bias is removed It is not important which adjustment is madebecause the difference between the alternative assumptions is captured by the unit fixed effect

he numbers o separate sites and o observations per site mean that it is not possible to

provide a simple visual representation o the data Figure 8 provides a summary by illustrating

the distributions o load actors by the age o the wind arm pooled across all years he ranges

o the load actors or each age are wide while the medians show no clear trend by age hus a

naiumlve analysis might conclude that there is nothing to investigate However as will be explained

below the distributions mask crucial differences between the perormance o wind arms as

they age because they do not control or the differences in location and wind availability over

time

B Data for Denmark

he data or Danish wind arms used in this study comes rom a database compiled by the

Danish Energy Agency covering the characteristics and perormance o all wind turbines rom

2002 up to the end o August 201216 he basic recording unit or the register is the individual

wind turbine but where a number o turbines are connected to a single meter an identical

average output per turbine is recorded or all turbines in the group he presence o what are in

effect repeated observations increases the noise in the data In addition the model that is used

or the analysis assumes that site operator and turbine characteristics are common or all o theunits that make up a single wind arm o address this point separate wind arms have been

identified by grouping all turbines which have identical values or turbine manuacturer and

type local authority date o commissioning and date o decommissioning (where applicable)

he register contains inormation on 7569 turbines commissioned rom 1977 onwards his

includes a large number o small turbines which would not all within the scope o the ROC

scheme in the UK Hence or comparability with the UK data all turbines commissioned

prior to 1992 were dropped as well as all wind arms with a generating capacity o less than 1

MW he resulting sample covered 823 onshore wind arms and 30 offshore wind arms with

monthly load actors or various periods rom January 2002 to August 2012 Given the cut-off

or wind arms commissioned prior to 1992 the maximum age or an onshore wind arm was 20

years and the maximum age or an offshore wind arm was 17 years

he total capacity o the onshore wind arms recorded in the Danish database was 2757 MW

he average size o an onshore wind arm in Denmark is small ndash only 33 MW as compared

with 161 MW or the UK ndash because 84 o the sample consists o installations with between

1 and 4 turbines he largest onshore installation has 39 turbines with a capacity o 234 MW

while the onshore installation with the greatest capacity has 19 turbines with a capacity o 584

MW Overall the sample o Danish onshore wind arms is older and smaller than the sample

o UK onshore wind arms reflecting the much longer history o wind generation in Denmark

16 httpwwwensdkEN-USINFOFACSANDFIGURESENERGY_SAISICS_AND_INDICAORS

OVERVIEWOFHEENERGYSECORREGISEROFWINDURBINESSiderForsideaspx

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2852

26 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

he larger scale o UK wind arms may account or some o the differences in perormance

curves or the two countries

he sample o offshore wind arms in Denmark is quite small with 30 installations in

total Still it is useul to analyse the perormance o these installations since there is very little

comparable data or the UK he sample covers 394 turbines with an average o just over 13turbines per installation and an average capacity o 288 MW he largest installation which

was commissioned in 2003 has 72 turbines with a total capacity o 1656 MW Again offshore

installations are smaller and older in Denmark than is the current pattern in the UK he first

two wind arms in the sample were commissioned in the years 1995 and 2000 but the sample

increases significantly rom 2002 onwards o avoid reliance upon a sample o 1 or 2 wind

arms to determine specific coefficients or ages gt 10 years all observations with an age greater

than 10 were omitted rom the analysis17

A somewhat different problem with the offshore wind arm data arose because the coeffi-

cients estimated using capacity weights (see below) were heavily influenced by the largest windarm his has had an extremely erratic profile o perormance over time with a very low average

load actor he treatment o this wind arm is discussed in more detail in Section F below

C Specication and estimation methods

As Figure 8 illustrates it is necessary to go beyond simple summary statistics in order to assess

whether there is any systematic relationship between age and the average perormance o wind

arms he perormance o wind arms varies over both time and space some months and years

have more or less wind than the long term average while specific characteristics o wind arms

ndash including location type o turbine and operating regime ndash will influence the perormance oeach plant under identical wind conditions

he combination o (a) site-specific characteristics which are constant over time and (b)

period-specific characteristics which are constant across wind arms can be represented by

what statisticians call an error components model with fixed effects or each wind arm and

each time period he dependent variable in the model consists o a sequence o load actors ndash

denoted by or unit i in month t ndash or each wind arm (the panel unit) over time he load

actors are calculated by dividing the total output in MWh by (24 x number o days in month x

reported generating capacity in MW) and are multiplied by 100 to convert to percentages

he additive version o the error components model may be written as

(1)

in which denotes the age o plant i in period t and is some unction that is either a

representation o or an approximation to the normalised age-perormance relationship he

multiplicative version is similar except that is replaced by where ln() denotes the

natural logarithm he terms are wind arm or unit fixed effects which capture actors such

as location type o turbine operating regimes etc he terms are period fixed effects which

capture the influence o wind conditions over the UK or Denmark as a whole as well as seasonal

maintenance and demand Without loss o generality it may be assumed that the fixed effects are

17 For the avoidance o doubt this means that no attempt was made to estimate the effect o age on the perormance o

offshore wind arms beyond age 10 while the data or the oldest wind arms was included in the analysis up to and

including age 10

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 27

normalised so that ie all o the error components have a zero mean while the

mean o the random error is zero by definition he terms are random errors which capture

random variations in wind conditions that are not specific to the site turbine breakdowns and

other actors that are uncorrelated with either the site or the time period It is assumed that the

mean o the random error is zero while additional assumptions depend upon the method o

estimation that is adopted

he model specified in equation (1) can be estimated in a variety o ways18 A crucial issue is

whether the are or may be significantly correlated with the age terms or any other independent

regressors I it is assumed that there is no such correlation the model is usually reerred to as

a random effects model In that case the equation can be estimated by a variant o least squares

(LS) generalised least squares (GLS) generalised methods o moments (GMM) or maximum

likelihood (ML) ndash all o which should be yield parameter estimates that will converge in prob-

ability to the true parameter values as the sample size increases An alternative but less restric-

tive specification is known as the fixed effects model in which no assumption is made about

the correlation between the unit fixed effects and other regressors he limitation o the fixed

effects model is that it cannot be used to estimate parameters or regressors that vary across

panel units but are constant over time or a panel unit (time-invariant regressors) ndash eg the

year in which a wind arm was commissioned its location or its rated generating capacity he

influence o such variables is captured by the unit fixed effects

In order to avoid the risk o obtaining biased estimates o the age effects the results reported

in this note are based upon estimates using the fixed effects model he influence o time-invar-

iant regressors is examined in a second stage by regressing the estimates o the unit fixed effectson these variables he simplest method o estimating the fixed effects model is known as the

lsquowithinrsquo estimator in which the mean value or each unit is subtracted rom all observations or

that unit For example i is a general polynomial o order M in A equation (1) becomes

(2)

so that the model can be estimated by least square with the inclusion o dummy variables or

each period t he variance-covariance matrix o the coefficients have been estimated by using

(a) a robust sandwich estimator adjusted or clustering by panel unit and (b) a bootstrap esti-

mator with 400 repetitions he estimates o the standard errors will be consistent even i the

random error component is serially correlated within panel units andor has different variances

across panel units19 Since the number o panel units and the average number o time periods

per panel units are both large (other than or Danish offshore wind arms) the least square esti-

mates o the parameters will be unbiased under these assumptions

he bootstrap method is expensive to carry out but it provides a useul cross-check on

the standard errors generated by other methods Unortunately the bootstrap method o

18 See Chapter 21 o AC Cameron amp PK rivedi ndash Microeconometrics Methods and Applications (Cambridge

Cambridge University Press 2005) or a standard textbook treatment o estimation methods or panel data models

19 he estimation was carried out using the xtreg procedure in Stata Version 12 he methods o constructing the

sandwich and bootstrap estimates o the variance-covariance matrix are described in Chapters 8 amp 13 o AC

Cameron amp PK rivedi ndash Microeconometrics using Stata (Revised Edition College Station exas Stata Press 2010)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 26: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2652

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 25

the decline in the plantrsquos load actor as it ages As an example consider a plant whose reported

capacity was increased rom 99 MW to 10 MW at age 3 I this was simply a change in report-

ing practice this would reduce the estimated load actor rom age 3 onwards By increasing

the initial capacity rom 99 MW to 10 MW ndash or by reducing the capacity rom 10 MW to 99

MW rom age 3 onwards ndash this bias is removed It is not important which adjustment is madebecause the difference between the alternative assumptions is captured by the unit fixed effect

he numbers o separate sites and o observations per site mean that it is not possible to

provide a simple visual representation o the data Figure 8 provides a summary by illustrating

the distributions o load actors by the age o the wind arm pooled across all years he ranges

o the load actors or each age are wide while the medians show no clear trend by age hus a

naiumlve analysis might conclude that there is nothing to investigate However as will be explained

below the distributions mask crucial differences between the perormance o wind arms as

they age because they do not control or the differences in location and wind availability over

time

B Data for Denmark

he data or Danish wind arms used in this study comes rom a database compiled by the

Danish Energy Agency covering the characteristics and perormance o all wind turbines rom

2002 up to the end o August 201216 he basic recording unit or the register is the individual

wind turbine but where a number o turbines are connected to a single meter an identical

average output per turbine is recorded or all turbines in the group he presence o what are in

effect repeated observations increases the noise in the data In addition the model that is used

or the analysis assumes that site operator and turbine characteristics are common or all o theunits that make up a single wind arm o address this point separate wind arms have been

identified by grouping all turbines which have identical values or turbine manuacturer and

type local authority date o commissioning and date o decommissioning (where applicable)

he register contains inormation on 7569 turbines commissioned rom 1977 onwards his

includes a large number o small turbines which would not all within the scope o the ROC

scheme in the UK Hence or comparability with the UK data all turbines commissioned

prior to 1992 were dropped as well as all wind arms with a generating capacity o less than 1

MW he resulting sample covered 823 onshore wind arms and 30 offshore wind arms with

monthly load actors or various periods rom January 2002 to August 2012 Given the cut-off

or wind arms commissioned prior to 1992 the maximum age or an onshore wind arm was 20

years and the maximum age or an offshore wind arm was 17 years

he total capacity o the onshore wind arms recorded in the Danish database was 2757 MW

he average size o an onshore wind arm in Denmark is small ndash only 33 MW as compared

with 161 MW or the UK ndash because 84 o the sample consists o installations with between

1 and 4 turbines he largest onshore installation has 39 turbines with a capacity o 234 MW

while the onshore installation with the greatest capacity has 19 turbines with a capacity o 584

MW Overall the sample o Danish onshore wind arms is older and smaller than the sample

o UK onshore wind arms reflecting the much longer history o wind generation in Denmark

16 httpwwwensdkEN-USINFOFACSANDFIGURESENERGY_SAISICS_AND_INDICAORS

OVERVIEWOFHEENERGYSECORREGISEROFWINDURBINESSiderForsideaspx

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2852

26 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

he larger scale o UK wind arms may account or some o the differences in perormance

curves or the two countries

he sample o offshore wind arms in Denmark is quite small with 30 installations in

total Still it is useul to analyse the perormance o these installations since there is very little

comparable data or the UK he sample covers 394 turbines with an average o just over 13turbines per installation and an average capacity o 288 MW he largest installation which

was commissioned in 2003 has 72 turbines with a total capacity o 1656 MW Again offshore

installations are smaller and older in Denmark than is the current pattern in the UK he first

two wind arms in the sample were commissioned in the years 1995 and 2000 but the sample

increases significantly rom 2002 onwards o avoid reliance upon a sample o 1 or 2 wind

arms to determine specific coefficients or ages gt 10 years all observations with an age greater

than 10 were omitted rom the analysis17

A somewhat different problem with the offshore wind arm data arose because the coeffi-

cients estimated using capacity weights (see below) were heavily influenced by the largest windarm his has had an extremely erratic profile o perormance over time with a very low average

load actor he treatment o this wind arm is discussed in more detail in Section F below

C Specication and estimation methods

As Figure 8 illustrates it is necessary to go beyond simple summary statistics in order to assess

whether there is any systematic relationship between age and the average perormance o wind

arms he perormance o wind arms varies over both time and space some months and years

have more or less wind than the long term average while specific characteristics o wind arms

ndash including location type o turbine and operating regime ndash will influence the perormance oeach plant under identical wind conditions

he combination o (a) site-specific characteristics which are constant over time and (b)

period-specific characteristics which are constant across wind arms can be represented by

what statisticians call an error components model with fixed effects or each wind arm and

each time period he dependent variable in the model consists o a sequence o load actors ndash

denoted by or unit i in month t ndash or each wind arm (the panel unit) over time he load

actors are calculated by dividing the total output in MWh by (24 x number o days in month x

reported generating capacity in MW) and are multiplied by 100 to convert to percentages

he additive version o the error components model may be written as

(1)

in which denotes the age o plant i in period t and is some unction that is either a

representation o or an approximation to the normalised age-perormance relationship he

multiplicative version is similar except that is replaced by where ln() denotes the

natural logarithm he terms are wind arm or unit fixed effects which capture actors such

as location type o turbine operating regimes etc he terms are period fixed effects which

capture the influence o wind conditions over the UK or Denmark as a whole as well as seasonal

maintenance and demand Without loss o generality it may be assumed that the fixed effects are

17 For the avoidance o doubt this means that no attempt was made to estimate the effect o age on the perormance o

offshore wind arms beyond age 10 while the data or the oldest wind arms was included in the analysis up to and

including age 10

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 27

normalised so that ie all o the error components have a zero mean while the

mean o the random error is zero by definition he terms are random errors which capture

random variations in wind conditions that are not specific to the site turbine breakdowns and

other actors that are uncorrelated with either the site or the time period It is assumed that the

mean o the random error is zero while additional assumptions depend upon the method o

estimation that is adopted

he model specified in equation (1) can be estimated in a variety o ways18 A crucial issue is

whether the are or may be significantly correlated with the age terms or any other independent

regressors I it is assumed that there is no such correlation the model is usually reerred to as

a random effects model In that case the equation can be estimated by a variant o least squares

(LS) generalised least squares (GLS) generalised methods o moments (GMM) or maximum

likelihood (ML) ndash all o which should be yield parameter estimates that will converge in prob-

ability to the true parameter values as the sample size increases An alternative but less restric-

tive specification is known as the fixed effects model in which no assumption is made about

the correlation between the unit fixed effects and other regressors he limitation o the fixed

effects model is that it cannot be used to estimate parameters or regressors that vary across

panel units but are constant over time or a panel unit (time-invariant regressors) ndash eg the

year in which a wind arm was commissioned its location or its rated generating capacity he

influence o such variables is captured by the unit fixed effects

In order to avoid the risk o obtaining biased estimates o the age effects the results reported

in this note are based upon estimates using the fixed effects model he influence o time-invar-

iant regressors is examined in a second stage by regressing the estimates o the unit fixed effectson these variables he simplest method o estimating the fixed effects model is known as the

lsquowithinrsquo estimator in which the mean value or each unit is subtracted rom all observations or

that unit For example i is a general polynomial o order M in A equation (1) becomes

(2)

so that the model can be estimated by least square with the inclusion o dummy variables or

each period t he variance-covariance matrix o the coefficients have been estimated by using

(a) a robust sandwich estimator adjusted or clustering by panel unit and (b) a bootstrap esti-

mator with 400 repetitions he estimates o the standard errors will be consistent even i the

random error component is serially correlated within panel units andor has different variances

across panel units19 Since the number o panel units and the average number o time periods

per panel units are both large (other than or Danish offshore wind arms) the least square esti-

mates o the parameters will be unbiased under these assumptions

he bootstrap method is expensive to carry out but it provides a useul cross-check on

the standard errors generated by other methods Unortunately the bootstrap method o

18 See Chapter 21 o AC Cameron amp PK rivedi ndash Microeconometrics Methods and Applications (Cambridge

Cambridge University Press 2005) or a standard textbook treatment o estimation methods or panel data models

19 he estimation was carried out using the xtreg procedure in Stata Version 12 he methods o constructing the

sandwich and bootstrap estimates o the variance-covariance matrix are described in Chapters 8 amp 13 o AC

Cameron amp PK rivedi ndash Microeconometrics using Stata (Revised Edition College Station exas Stata Press 2010)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 27: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 25

the decline in the plantrsquos load actor as it ages As an example consider a plant whose reported

capacity was increased rom 99 MW to 10 MW at age 3 I this was simply a change in report-

ing practice this would reduce the estimated load actor rom age 3 onwards By increasing

the initial capacity rom 99 MW to 10 MW ndash or by reducing the capacity rom 10 MW to 99

MW rom age 3 onwards ndash this bias is removed It is not important which adjustment is madebecause the difference between the alternative assumptions is captured by the unit fixed effect

he numbers o separate sites and o observations per site mean that it is not possible to

provide a simple visual representation o the data Figure 8 provides a summary by illustrating

the distributions o load actors by the age o the wind arm pooled across all years he ranges

o the load actors or each age are wide while the medians show no clear trend by age hus a

naiumlve analysis might conclude that there is nothing to investigate However as will be explained

below the distributions mask crucial differences between the perormance o wind arms as

they age because they do not control or the differences in location and wind availability over

time

B Data for Denmark

he data or Danish wind arms used in this study comes rom a database compiled by the

Danish Energy Agency covering the characteristics and perormance o all wind turbines rom

2002 up to the end o August 201216 he basic recording unit or the register is the individual

wind turbine but where a number o turbines are connected to a single meter an identical

average output per turbine is recorded or all turbines in the group he presence o what are in

effect repeated observations increases the noise in the data In addition the model that is used

or the analysis assumes that site operator and turbine characteristics are common or all o theunits that make up a single wind arm o address this point separate wind arms have been

identified by grouping all turbines which have identical values or turbine manuacturer and

type local authority date o commissioning and date o decommissioning (where applicable)

he register contains inormation on 7569 turbines commissioned rom 1977 onwards his

includes a large number o small turbines which would not all within the scope o the ROC

scheme in the UK Hence or comparability with the UK data all turbines commissioned

prior to 1992 were dropped as well as all wind arms with a generating capacity o less than 1

MW he resulting sample covered 823 onshore wind arms and 30 offshore wind arms with

monthly load actors or various periods rom January 2002 to August 2012 Given the cut-off

or wind arms commissioned prior to 1992 the maximum age or an onshore wind arm was 20

years and the maximum age or an offshore wind arm was 17 years

he total capacity o the onshore wind arms recorded in the Danish database was 2757 MW

he average size o an onshore wind arm in Denmark is small ndash only 33 MW as compared

with 161 MW or the UK ndash because 84 o the sample consists o installations with between

1 and 4 turbines he largest onshore installation has 39 turbines with a capacity o 234 MW

while the onshore installation with the greatest capacity has 19 turbines with a capacity o 584

MW Overall the sample o Danish onshore wind arms is older and smaller than the sample

o UK onshore wind arms reflecting the much longer history o wind generation in Denmark

16 httpwwwensdkEN-USINFOFACSANDFIGURESENERGY_SAISICS_AND_INDICAORS

OVERVIEWOFHEENERGYSECORREGISEROFWINDURBINESSiderForsideaspx

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2852

26 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

he larger scale o UK wind arms may account or some o the differences in perormance

curves or the two countries

he sample o offshore wind arms in Denmark is quite small with 30 installations in

total Still it is useul to analyse the perormance o these installations since there is very little

comparable data or the UK he sample covers 394 turbines with an average o just over 13turbines per installation and an average capacity o 288 MW he largest installation which

was commissioned in 2003 has 72 turbines with a total capacity o 1656 MW Again offshore

installations are smaller and older in Denmark than is the current pattern in the UK he first

two wind arms in the sample were commissioned in the years 1995 and 2000 but the sample

increases significantly rom 2002 onwards o avoid reliance upon a sample o 1 or 2 wind

arms to determine specific coefficients or ages gt 10 years all observations with an age greater

than 10 were omitted rom the analysis17

A somewhat different problem with the offshore wind arm data arose because the coeffi-

cients estimated using capacity weights (see below) were heavily influenced by the largest windarm his has had an extremely erratic profile o perormance over time with a very low average

load actor he treatment o this wind arm is discussed in more detail in Section F below

C Specication and estimation methods

As Figure 8 illustrates it is necessary to go beyond simple summary statistics in order to assess

whether there is any systematic relationship between age and the average perormance o wind

arms he perormance o wind arms varies over both time and space some months and years

have more or less wind than the long term average while specific characteristics o wind arms

ndash including location type o turbine and operating regime ndash will influence the perormance oeach plant under identical wind conditions

he combination o (a) site-specific characteristics which are constant over time and (b)

period-specific characteristics which are constant across wind arms can be represented by

what statisticians call an error components model with fixed effects or each wind arm and

each time period he dependent variable in the model consists o a sequence o load actors ndash

denoted by or unit i in month t ndash or each wind arm (the panel unit) over time he load

actors are calculated by dividing the total output in MWh by (24 x number o days in month x

reported generating capacity in MW) and are multiplied by 100 to convert to percentages

he additive version o the error components model may be written as

(1)

in which denotes the age o plant i in period t and is some unction that is either a

representation o or an approximation to the normalised age-perormance relationship he

multiplicative version is similar except that is replaced by where ln() denotes the

natural logarithm he terms are wind arm or unit fixed effects which capture actors such

as location type o turbine operating regimes etc he terms are period fixed effects which

capture the influence o wind conditions over the UK or Denmark as a whole as well as seasonal

maintenance and demand Without loss o generality it may be assumed that the fixed effects are

17 For the avoidance o doubt this means that no attempt was made to estimate the effect o age on the perormance o

offshore wind arms beyond age 10 while the data or the oldest wind arms was included in the analysis up to and

including age 10

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 27

normalised so that ie all o the error components have a zero mean while the

mean o the random error is zero by definition he terms are random errors which capture

random variations in wind conditions that are not specific to the site turbine breakdowns and

other actors that are uncorrelated with either the site or the time period It is assumed that the

mean o the random error is zero while additional assumptions depend upon the method o

estimation that is adopted

he model specified in equation (1) can be estimated in a variety o ways18 A crucial issue is

whether the are or may be significantly correlated with the age terms or any other independent

regressors I it is assumed that there is no such correlation the model is usually reerred to as

a random effects model In that case the equation can be estimated by a variant o least squares

(LS) generalised least squares (GLS) generalised methods o moments (GMM) or maximum

likelihood (ML) ndash all o which should be yield parameter estimates that will converge in prob-

ability to the true parameter values as the sample size increases An alternative but less restric-

tive specification is known as the fixed effects model in which no assumption is made about

the correlation between the unit fixed effects and other regressors he limitation o the fixed

effects model is that it cannot be used to estimate parameters or regressors that vary across

panel units but are constant over time or a panel unit (time-invariant regressors) ndash eg the

year in which a wind arm was commissioned its location or its rated generating capacity he

influence o such variables is captured by the unit fixed effects

In order to avoid the risk o obtaining biased estimates o the age effects the results reported

in this note are based upon estimates using the fixed effects model he influence o time-invar-

iant regressors is examined in a second stage by regressing the estimates o the unit fixed effectson these variables he simplest method o estimating the fixed effects model is known as the

lsquowithinrsquo estimator in which the mean value or each unit is subtracted rom all observations or

that unit For example i is a general polynomial o order M in A equation (1) becomes

(2)

so that the model can be estimated by least square with the inclusion o dummy variables or

each period t he variance-covariance matrix o the coefficients have been estimated by using

(a) a robust sandwich estimator adjusted or clustering by panel unit and (b) a bootstrap esti-

mator with 400 repetitions he estimates o the standard errors will be consistent even i the

random error component is serially correlated within panel units andor has different variances

across panel units19 Since the number o panel units and the average number o time periods

per panel units are both large (other than or Danish offshore wind arms) the least square esti-

mates o the parameters will be unbiased under these assumptions

he bootstrap method is expensive to carry out but it provides a useul cross-check on

the standard errors generated by other methods Unortunately the bootstrap method o

18 See Chapter 21 o AC Cameron amp PK rivedi ndash Microeconometrics Methods and Applications (Cambridge

Cambridge University Press 2005) or a standard textbook treatment o estimation methods or panel data models

19 he estimation was carried out using the xtreg procedure in Stata Version 12 he methods o constructing the

sandwich and bootstrap estimates o the variance-covariance matrix are described in Chapters 8 amp 13 o AC

Cameron amp PK rivedi ndash Microeconometrics using Stata (Revised Edition College Station exas Stata Press 2010)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 28: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2852

26 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

he larger scale o UK wind arms may account or some o the differences in perormance

curves or the two countries

he sample o offshore wind arms in Denmark is quite small with 30 installations in

total Still it is useul to analyse the perormance o these installations since there is very little

comparable data or the UK he sample covers 394 turbines with an average o just over 13turbines per installation and an average capacity o 288 MW he largest installation which

was commissioned in 2003 has 72 turbines with a total capacity o 1656 MW Again offshore

installations are smaller and older in Denmark than is the current pattern in the UK he first

two wind arms in the sample were commissioned in the years 1995 and 2000 but the sample

increases significantly rom 2002 onwards o avoid reliance upon a sample o 1 or 2 wind

arms to determine specific coefficients or ages gt 10 years all observations with an age greater

than 10 were omitted rom the analysis17

A somewhat different problem with the offshore wind arm data arose because the coeffi-

cients estimated using capacity weights (see below) were heavily influenced by the largest windarm his has had an extremely erratic profile o perormance over time with a very low average

load actor he treatment o this wind arm is discussed in more detail in Section F below

C Specication and estimation methods

As Figure 8 illustrates it is necessary to go beyond simple summary statistics in order to assess

whether there is any systematic relationship between age and the average perormance o wind

arms he perormance o wind arms varies over both time and space some months and years

have more or less wind than the long term average while specific characteristics o wind arms

ndash including location type o turbine and operating regime ndash will influence the perormance oeach plant under identical wind conditions

he combination o (a) site-specific characteristics which are constant over time and (b)

period-specific characteristics which are constant across wind arms can be represented by

what statisticians call an error components model with fixed effects or each wind arm and

each time period he dependent variable in the model consists o a sequence o load actors ndash

denoted by or unit i in month t ndash or each wind arm (the panel unit) over time he load

actors are calculated by dividing the total output in MWh by (24 x number o days in month x

reported generating capacity in MW) and are multiplied by 100 to convert to percentages

he additive version o the error components model may be written as

(1)

in which denotes the age o plant i in period t and is some unction that is either a

representation o or an approximation to the normalised age-perormance relationship he

multiplicative version is similar except that is replaced by where ln() denotes the

natural logarithm he terms are wind arm or unit fixed effects which capture actors such

as location type o turbine operating regimes etc he terms are period fixed effects which

capture the influence o wind conditions over the UK or Denmark as a whole as well as seasonal

maintenance and demand Without loss o generality it may be assumed that the fixed effects are

17 For the avoidance o doubt this means that no attempt was made to estimate the effect o age on the perormance o

offshore wind arms beyond age 10 while the data or the oldest wind arms was included in the analysis up to and

including age 10

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 27

normalised so that ie all o the error components have a zero mean while the

mean o the random error is zero by definition he terms are random errors which capture

random variations in wind conditions that are not specific to the site turbine breakdowns and

other actors that are uncorrelated with either the site or the time period It is assumed that the

mean o the random error is zero while additional assumptions depend upon the method o

estimation that is adopted

he model specified in equation (1) can be estimated in a variety o ways18 A crucial issue is

whether the are or may be significantly correlated with the age terms or any other independent

regressors I it is assumed that there is no such correlation the model is usually reerred to as

a random effects model In that case the equation can be estimated by a variant o least squares

(LS) generalised least squares (GLS) generalised methods o moments (GMM) or maximum

likelihood (ML) ndash all o which should be yield parameter estimates that will converge in prob-

ability to the true parameter values as the sample size increases An alternative but less restric-

tive specification is known as the fixed effects model in which no assumption is made about

the correlation between the unit fixed effects and other regressors he limitation o the fixed

effects model is that it cannot be used to estimate parameters or regressors that vary across

panel units but are constant over time or a panel unit (time-invariant regressors) ndash eg the

year in which a wind arm was commissioned its location or its rated generating capacity he

influence o such variables is captured by the unit fixed effects

In order to avoid the risk o obtaining biased estimates o the age effects the results reported

in this note are based upon estimates using the fixed effects model he influence o time-invar-

iant regressors is examined in a second stage by regressing the estimates o the unit fixed effectson these variables he simplest method o estimating the fixed effects model is known as the

lsquowithinrsquo estimator in which the mean value or each unit is subtracted rom all observations or

that unit For example i is a general polynomial o order M in A equation (1) becomes

(2)

so that the model can be estimated by least square with the inclusion o dummy variables or

each period t he variance-covariance matrix o the coefficients have been estimated by using

(a) a robust sandwich estimator adjusted or clustering by panel unit and (b) a bootstrap esti-

mator with 400 repetitions he estimates o the standard errors will be consistent even i the

random error component is serially correlated within panel units andor has different variances

across panel units19 Since the number o panel units and the average number o time periods

per panel units are both large (other than or Danish offshore wind arms) the least square esti-

mates o the parameters will be unbiased under these assumptions

he bootstrap method is expensive to carry out but it provides a useul cross-check on

the standard errors generated by other methods Unortunately the bootstrap method o

18 See Chapter 21 o AC Cameron amp PK rivedi ndash Microeconometrics Methods and Applications (Cambridge

Cambridge University Press 2005) or a standard textbook treatment o estimation methods or panel data models

19 he estimation was carried out using the xtreg procedure in Stata Version 12 he methods o constructing the

sandwich and bootstrap estimates o the variance-covariance matrix are described in Chapters 8 amp 13 o AC

Cameron amp PK rivedi ndash Microeconometrics using Stata (Revised Edition College Station exas Stata Press 2010)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 29: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 2952

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 27

normalised so that ie all o the error components have a zero mean while the

mean o the random error is zero by definition he terms are random errors which capture

random variations in wind conditions that are not specific to the site turbine breakdowns and

other actors that are uncorrelated with either the site or the time period It is assumed that the

mean o the random error is zero while additional assumptions depend upon the method o

estimation that is adopted

he model specified in equation (1) can be estimated in a variety o ways18 A crucial issue is

whether the are or may be significantly correlated with the age terms or any other independent

regressors I it is assumed that there is no such correlation the model is usually reerred to as

a random effects model In that case the equation can be estimated by a variant o least squares

(LS) generalised least squares (GLS) generalised methods o moments (GMM) or maximum

likelihood (ML) ndash all o which should be yield parameter estimates that will converge in prob-

ability to the true parameter values as the sample size increases An alternative but less restric-

tive specification is known as the fixed effects model in which no assumption is made about

the correlation between the unit fixed effects and other regressors he limitation o the fixed

effects model is that it cannot be used to estimate parameters or regressors that vary across

panel units but are constant over time or a panel unit (time-invariant regressors) ndash eg the

year in which a wind arm was commissioned its location or its rated generating capacity he

influence o such variables is captured by the unit fixed effects

In order to avoid the risk o obtaining biased estimates o the age effects the results reported

in this note are based upon estimates using the fixed effects model he influence o time-invar-

iant regressors is examined in a second stage by regressing the estimates o the unit fixed effectson these variables he simplest method o estimating the fixed effects model is known as the

lsquowithinrsquo estimator in which the mean value or each unit is subtracted rom all observations or

that unit For example i is a general polynomial o order M in A equation (1) becomes

(2)

so that the model can be estimated by least square with the inclusion o dummy variables or

each period t he variance-covariance matrix o the coefficients have been estimated by using

(a) a robust sandwich estimator adjusted or clustering by panel unit and (b) a bootstrap esti-

mator with 400 repetitions he estimates o the standard errors will be consistent even i the

random error component is serially correlated within panel units andor has different variances

across panel units19 Since the number o panel units and the average number o time periods

per panel units are both large (other than or Danish offshore wind arms) the least square esti-

mates o the parameters will be unbiased under these assumptions

he bootstrap method is expensive to carry out but it provides a useul cross-check on

the standard errors generated by other methods Unortunately the bootstrap method o

18 See Chapter 21 o AC Cameron amp PK rivedi ndash Microeconometrics Methods and Applications (Cambridge

Cambridge University Press 2005) or a standard textbook treatment o estimation methods or panel data models

19 he estimation was carried out using the xtreg procedure in Stata Version 12 he methods o constructing the

sandwich and bootstrap estimates o the variance-covariance matrix are described in Chapters 8 amp 13 o AC

Cameron amp PK rivedi ndash Microeconometrics using Stata (Revised Edition College Station exas Stata Press 2010)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 30: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3052

28 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

resampling panel units with replacement can lead to degeneracy in the estimation or some o

the bootstrap samples When the proportion o degenerate samples is significant the statisti-

cal properties o the bootstrap standard errors are not clear his was a particular problem or

the sample o Danish offshore installations and the robust standard errors are reported in this

case In practice the robust and bootstrap standard errors are very similar so that the methodo estimating the standard errors does not alter any general conclusions about changes in the

perormance o wind arms as they age

I standard methods o estimation are applied to the data or the UK and Denmark the

results will generate perormance curves that reflect the perormance o the typical wind arm

Since the distribution o wind arms by capacity is heavily skewed the typical wind arm has

a much smaller capacity than the average o all wind arms I there is any kind o relationship

between scale and perormance the perormance curves may not provide a good guide to the

aggregate perormance o all wind arms Hence as an alternative the models have also been

estimated using weights or each wind arm that are proportional to the capacity o the windarm normalised so that the sum o the weights is equal to the to number o observations hese

are reerred to as capacity-weighted estimates and the perormance curves derived rom the

estimates reflect the perormance o the typical MW o wind capacity

D Period xed effects vs normalisation by wind speeds

In its quarterly publication Energy Trends the UK Department o Energy and Climate Change

publishes a monthly index o average wind speeds based upon data collected at 14 sites spread

over the UK he descriptive material that accompanies the table (including an article published

in the September 2008 issue o Energy Trends) does not provide details o how the data or theseparate sites is combined but it seems reasonable to iner that the index is simply an average o

the wind speeds or each site

An obvious question is whether this wind index offers a satisactory or perhaps better way

o normalising or wind availability than the inclusion o period fixed effects as specified

in equation (1) above hough it may be surprising to some there are strong mathematical

reasons or preerring the period fixed effects he reason is based on the act that the relation-

ship between wind speed and electricity output or a wind turbines is highly non-linear he

direct relationship between wind speed and energy ollows a cube power law In addition wind

turbines have cut-in speeds and are designed so that they will not produce more than their

rated capacity even i wind speed increases above the level at which capacity output is achieved

Finally or saety reasons turbines have a cut-out wind speed so that output is zero above that

level though this does not have much effect on the analysis that ollows

Hence the amount o electricity generated per MW o capacity in monitoring location j

in period t o month m with a steady wind speed o may be written as

(3)

where is a non-linear (roughly S-shaped) unction Wind speeds vary almost continuously

but it is airly standard to use an averaging period o 5 minutes so that t reers to time measuredin 5 minute intervals he total output or month m will be

(4)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 31: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 29

where is the average wind speed at location j or month m he inequality states that we

cannot rely on being able to calculate total electricity output during the month as a unction

φ() o average wind speed Clearly the inequality remains i is replaced by the wind index

Now consider the unobserved profile o wind speeds over periods t = 1hellip at unit i ormonth m he monthly equivalent wind speed is defined by

(5)

his is the wind speed which i maintained steadily at unit i throughout month m would have

generated the same level o output as generated by the actual profile o wind speeds Averaging

the equivalent wind speeds over all units gives and in which the

are deviations rom average equivalent wind speed or unit i and month m By construction

the monthly averages o these deviations are all zero Using a standard first order expansion the

electricity output per MW or unit i in month m may be expressed as

(6)

where is the first derivative o the output unction with respect to wind speed at the average

equivalent wind speed

he point o this derivation is that the first term in equation (6) is a period fixed effect which

is common to all units in month m while the second term is a random error that is specific

to the unit and month It ollows that the specification in equation (1) may be interpreted as a

generalisation o (6) A corollary is that a statistical model in which the period fixed effects are

replaced by a unction o the wind index ndash denoted by ndash will only perorm as well as (1)

i or all m which requires that the wind index is a perect measure o equiva-

lent wind speed or all wind arms in the UK In practice this is impossible with a fixed index

because the population o wind arms changes over time

he relevance o this exercise is that it establishes as a matter o principle not just empir-

ical observation that the specification in terms o period fixed effects is a more efficient way

o normalising perormance or variations in wind availability over time than using any wind

index o the kind constructed rom (weighted) averages o wind speeds at a fixed set o loca-

tions I the output unction was linear in W then it might be possible to construct a windindex that would provide a close approximation to the average equivalent wind speed but that

is certainly not the case with the data analysed or this study However in the general case it is

simple to show that estimating a model with some unction o the wind index may yield biased

estimates o the coefficients i the period fixed effects are not included as well 20

20 Wind operators tend to treat detailed data on output and local wind conditions as being commercially confidential

It would be instructive to extend the analysis in this study by using examining daily output as a unction o age local

wind speeds and other variables i such data were available

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 32: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3252

30 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

E Estimation results for the UK

Alternative specifications or the period effects have been examined he most general spec-

ification is to estimate 120 separate coefficients one or each month in the 10 year period A

more restrictive specification is to assume that i period t corresponds to year s and month m

then so that the period effect is a composite o a year effect and a month effectSince weather patterns are clearly seasonal the assumption o a monthly fixed effect seems

reasonable he addition o an annual fixed effect is less obvious but there are clear differences

over a run o years in average wind speeds and the inclusion o an annual fixed effect eliminates

potential correlations between age and year he results are very similar or the two specifica-

tions so the tables report the more general variant with a ull set o period effects

For each o the main models which have been estimated a comparison has been carried out

between three alternative specifications o wind availability (a) a ull set o period fixed effects

(b) the log o average wind speed and (c) the combination o period fixed effects and the log o

average wind speed he results are quite consistent and correspond to the implications o the

analysis in the previous section hey are

bull he between-units values o R-square which is the key measure o goodness o fit or the

model are substantially higher or the specification with period fixed effects than or the

one with the wind variable

bull he between-units values o R-square and estimates o the coefficients on age and other

independent regressors are identical or the specifications with ull period effects with or

without the wind variable In effect the only consequence o adding the wind variable is to

redistribute the explanatory power o the equation between the period fixed effects and thewind variable

It ollows that the wind variable is redundant when the period fixed effects are included in

the model but it perorms much less well than the period fixed effects when a comparison is

made between the alternative specifications with each variable or set o variables included on

their own Since this outcome conorms with what would have been expected the results are

not reported in detail and the wind variable is not examined in the discussion o the estimation

results

Detailed results are reported or two versions o the age-perormance relationship he

first is a quadratic approximation ndash ndash while the second includes a ull set

o age effects ndash where Ds = 1 i A=s and Ds = 0 otherwise he quadratic specifica-

tion was chosen as a smooth approximation to the lsquotruersquo age-perormance relationship because

statistical tests showed that higher order polynomials added little to the explanatory power o

the estimated equations while the coefficient on the quadratic term was significantly different

rom zero with p lt 001 in almost all o the models examined he constant in the quadratic

approximation gives the normalised load actor or age = 0

he reason or estimating the specification with a ull set o age effects is that this does not

orce year-to-year changes in perormance to ollow some smooth pattern For example there

are strong a priori reasons to expect that the average load actor in the first year o operation

(age = 0) will be compromised by the time required to establish a satisactory operating regime

and to sort out any initial mechanical problems Hence it should be expected that the second

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 33: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 31

year o operation will be the year in which the design perormance o a wind arm is achieved

As a consequence age = 1 is used as the baseline category in presenting the results estimated

with a ull set o age effects as this means that the coefficients capture either out-perormance

(positive values) or the shortall in perormance (negative values) relative to expected design

perormancehe additive model can yield extreme or ineasible projections such as negative load actors

i it is extrapolated too ar whereas the projections rom the multiplicative model must be

greater than zero Hence unless stated otherwise the results o estimating the multiplicative

(log-linear) model provide the basis or the analysis in the remainder o the paper here is no

straightorward way o testing statistically whether a linear or log-linear specification fits the

data better as they do not provide nested hypotheses and the different transormations mean

that the variances o the errors cannot be compared so that their R 2 and other summary statis-

tics do not measure the same thing

he results o estimating the alternative specifications with a ull set o period effects areshown in able 2 (page 41) he period effects have been normalised by adopting August

2007 as the deault period ie with = 0 his normalisation was selected in order to ensure

that As a consequence the constant terms in the equations are equal to the means o

the load actors afer normalising or unit characteristics and age hese means are 240 or

the multiplicative specification with quadratic age effects and 243 or the ull set o age effects

With a very large number o degrees o reedom any coefficient with a t-ratio whose absolute

value is greater than 258 is significantly different rom zero at the 1 level or better Both o the

coefficients in the quadratic specification and all o the age effects or age gt 1 meet this crite-

rion so that there can be no doubt that there is a statistically significant deterioration in plant

perormance as wind arms get older Both variants o the relationship between load actor and

age perorm relatively well in capturing within-unit variance in load actors but the variance

between units (measured by the standard deviation o ) due to site and other characteristics is

as large as the variance o the pure error term (measured by the standard deviation o )

Figures 9A and 9B illustrates the results o using the estimated coefficients or the additive

and multiplicative models to generate the age-perormance curves standardised or wind

conditions and site characteristics he error bars illustrate the 95 confidence intervals or

the models with a ull set o age effects he quadratic representation o the age-perormance

curves yields a close approximation to the more detailed specification with individual age

effects Comparing the additive and multiplicative (log-linear) versions o the age-perormance

curves the latter has narrower confidence intervals or ages o 10 years or greater which is a

urther reason or preerring this specification

he normalised age-perormance curves in Figures 9A and 9B were estimated by giving

an equal weight in the estimation to each wind arm irrespective o the amount o installed

capacity hese results are representative o the typical wind arm However to obtain results

that are representative o the average MW o wind generating capacity it is necessary to estimate

the model using a weight or each wind arm that is proportional to the installed generatingcapacity o the plant ndash reerred to here as capacity weights Figure 10 compares the normalised

age-perormance curves (including the 95 confidence intervals) or the multiplicative specifi-

cation with ull age effects estimated using equal and capacity weights

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 34: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3452

32 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 9A Additive age-performance curves for UK onshore wind farms

-5

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r (

)

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

Figure 9B Multiplicative age-performance curves for UK onshore wind farms

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d

f a c t o r ( )

Wind farm age (years)

Quadratic Age effects

Source Authorrsquos estimates

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 35: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 33

Figure 10 UK onshore age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

L o a d f a c t o r ( )

Wind farm age (years)

Equal weights Capacity weights

Source Authorrsquos estimates

he decline in perormance with age is considerably greater when capacity weights are used

his implies that the perormance o large wind arms declines more rapidly than that o smaller

ones From age 3 onwards the confidence intervals or the two age-perormance curves do not

overlap so that it is unlikely that the difference between the two curves arises merely by chance

he normalised load actor per MW o capacity alls to about 7 at age 10 and 35 at age 15

With such low levels o perormance it seems very unlikely that large wind arms will continue

in operation beyond 10 years o age with a strong likelihood o re-powering at that point he

consequence is that large scale reliance upon wind power seems likely to involve a regular ndash and

costly ndash commitment to upgrading major components o the wind turbines

As a cross-check on the estimated models Figures 11A and 11B show box plots o the distri-

butions o residuals plotted against plant age or the specifications with ull age effects using

equal and capacity weights respectively he inter-quartile and ukey adjacent values show

little variation across plant age he numbers o observations are much greater or plant ages

at the bottom o the range (N gt 2000 or ages = 0 1 or 2 but N lt 400 or ages gt 13) he larger

numbers o outliers or the lowest age groups reflect differences in sample size he standard

deviations o the residuals by age group in Figure 11A all in a range rom 022 to 042 withthe highest values or ages 0 2 10 11 and 15 While the residuals are heteroskedastic there is

no systematic relationship between plant age and the residuals he use o robust or bootstrap

standard errors means that there should be no reason to be concerned about statistical iner-

ence based on the results in able 2 (page 41)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 36: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3652

34 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Figure 11A Residuals by age for performance curves using equal weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g

e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

Figure 11B Residuals by age for performance curves using capacity weights

- 4

- 2

0

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R a n d o m c

o m p o n e n t w i t h f u l l a g e e f f e c t s

Wind farm age (years)

Source Authorrsquos estimates

he unit fixed effects or each unit i can be used as an indicator o the relative effectiveness o

different wind plants Since wind turbines use no uel the crucial determinant o their effective-

ness is the number o hours per month or year that they operate assuming that their output isnot constrained by demand or transmission considerations Periods o constrained production

were minimal over the period covered by the data so this is not a significant consideration

Hence a unit with a value o at the top end o the distribution will operate or more hours

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 37: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 35

in any year than the average afer adjusting or wind conditions while a unit with a value

o at the bottom end o the distribution will operate or ewer hours per years he actors

which influence average efficiency or particular units will include site location the type o

wind turbines installed and operating practices Site location is likely to be the most important

actor since this will determine how ar the turbines can take advantage o exogenous windconditions

Figure 12 plots the unit fixed effects ndash ie the efficiencies ndash o wind plants in England and

Scotland commissioned in or afer 2000 together with trend lines over time21 A small number

o extreme outliers all with very low load actors have been excluded here is a very clear

downward trend over time in the unit fixed effects or Scotland ndash marked with red triangles and

a red dashed trend line he variance o perormance o wind plants in Scotland commissioned

in any one year is also large he perormance o wind plants in England has also allen over

time but more gradually

Figure 12 Unit fixed effects by year of commissioning for England and Scotland

-10

-08

-06

-04

-02

00

02

04

06

08

10

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

U n i t f i x e d

e f f e c t

Year of commissioning

England Scotland Linear (England) Linear (Scotland)

Note Te unit fixed effects are based on the log-linear specification with ull age effects

Source Authorrsquos estimates

A more comprehensive analysis can be constructed by treating the 282 values o the unit fixed

effects as data observations and estimating a regression equation in which the perormance o

a plant is affected by the year in which it was commissioned and its generating capacity able 3

(page 43) shows the results o estimating regression equations to identiy how the unit fixed

effects vary with the date o commissioning and capacity o wind arms he results indicate

an annual reduction o 38 in the normalised load actor by date o commissioning or the

equally weighted fixed effects and o 113 or the capacity-weighted fixed effects hese trends

21 A all in the unit fixed effect rom 02 to 01 translates to multiplying the normalised load actor by exp(01)exp(02)

= 0905

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 38: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3852

36 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

have been reinorced by diseconomies o scale so that recently commissioned and larger wind

arms have much lower normalised load actors than older and smaller wind arms In addition

wind arms in Northern Ireland and Scotland perorm better than those in England and Wales

he average load actor or a plant with a generating capacity o 50 MW is estimated to be

about 8 lower than that or a plant with a generating capacity o 10 MW commissioned inthe same year he combination o less avourable sites and larger units as the number o wind

arms has grown has had a major impact on average load actors

One hypothesis which merits investigation is that any perormance degradation is deter-

mined by usage rather than the passage o time his would imply that wind plants with

relatively low load actors experience less rapid degradation in perormance than those with

relatively high load actors his may be tested by including cumulative actual output normal-

ised by generating capacity as an explanatory variable along with age effects Since output is

not recorded or periods prior to April 2002 or wind plants that were commissioned beore

that date the analysis is restricted to the subset o plants commissioned afer April 2002 heanalysis (available on request rom the author) indicates that cumulative output does not have

any statistically significant influence on perormance degradation which depends upon age

alone

he most plausible explanation is that perormance degradation is linked to the cumulative

number o starts and stops or the wind turbines his is certainly the case or thermal gener-

ating plants or which maintenance requirements and perormance are strongly influenced by

the thermal stresses o start and stop cycles As a consequence thermal plants operate most

efficiently on base load when the number o starts and stops is minimised he inescapable

variability o wind speeds means that the stresses on mechanical and other components due to

start and stop cycles cannot be minimised by similar strategies

F Estimation results for Denmark

he results o estimating the model or Danish onshore and offshore wind arms are shown

in ables 4 (page 44) and 5 (page 46) Because the sample o offshore wind arms is airly

small the process o bootstrapping the standard errors generates a relatively high proportion

o degenerate results especially or the specification with a ull set o age effects so cluster-

robust standard errors are reported or all o the models estimated or offshore wind arms For

both onshore and offshore wind arms the specification with ull age effects yields no statistical

improvement on the quadratic specification

In Section B it was noted that the largest offshore wind arm appears to be an outlier and has

a very large influence on offshore age-perormance curve estimated using capacity weights22

o highlight the impact o including this observation the normalised load actor at age = 0 is

768 but it alls to 06 at age = 10 i this observation is included In contrast the equivalent

estimates are 328 at age = 0 alling to 99 at age = 10 when the observation is excluded

he time profile o normalised load actors when this observation is retained in the sample is

so extreme and implausible on technical grounds that the wind arm was excluded rom the

sample used to estimate the capacity-weighted model or offshore wind arms An alternative

22 his wind arm suffered rom some kind o major equipment ailure in 2007 with the consequence that the total

output rom 72 turbines was almost zero or more than 4 months

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 39: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 3952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 40: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4052

38 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

reject the random effects assumption or offshore installations In practice using the random

effects model or onshore wind arms generates results that are broadly similar to those reported

based on the fixed effects model

he normalised age-perormance curves or Danish wind arms using equal weights are

shown in Figures 13A and 13B while the comparison between the age-perormance curvesestimated using equal and capacity weights is shown in Figure 14 he confidence intervals

or the age-perormance curves or offshore wind arms are large because o the limited sample

o sites but the hypothesis that the load actor remains constant as plants age ndash ie that the

coefficients on age in columns (5) to (8) o able 4 (page 44) are all zero ndash is rejected or each

specification (p le 002) he comparison o the age-perormance curves derived using equal

and capacity weights in Figure 13 is based upon the quadratic specifications as these provide

simpler approximations which are not statistically different rom the estimates based on the

models with the ull set o age effects

Figure 14 Danish age-performance curves using equal and capacity weights

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10 12 14 16 18 20

L o a d f a c t o r ( )

Age (years)

Onshore Offshore Onshore ndash capacity Offshore ndash capacity

Source Authorrsquos estimates

he decline in the perormance o onshore wind arms in Denmark is less marked than or

the UK but there is a significant decline at an average o 1 o the previous yearrsquos load actor

or each year he decline is much more rapid or offshore wind arms with highly significant

negative coefficient on the quadratic term In both cases the rates o decline are larger when

wind arms are weighted by capacity in the analysis he estimates using a ull set o age effects

show that the normalised load actor in the first year o operation (age = 0) is lower than in the

baseline year with age = 1 From age = 3 to age = 16 the coefficients are increasingly negative

he results or ages gt 16 are erratic but it is likely that this reflects sample selection bias ie

older wind arms with poor perormance are more likely to be decommissioned early and thus

not appear in the sample

he age-related decline in the perormance o offshore wind arms is very rapid he

normalised perormance o an offshore wind arm alls rom a load actor o 45 at age = 0 to

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 41: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4152

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 39

12 at age = 10 he decline in the capacity-weighted load actors has a slightly different profile

with a stronger quadratic term that reflects a slower initial rate o decline in perormance which

accelerates afer age = 8 Overall the average load actors or ages up to and including age =

10 are very similar ndash about 28 ndash or the estimates derived using equal and capacity weights

his average is well below the load actor immediately afer commissioning Since typical loadactors afer age = 10 are likely to be well below this average the steady state load actor or a

large sample o offshore wind arms will be well below the figure o 35 that is ofen used as the

basis or policy assessments in the UK his will have important consequences or the cost o

offshore wind generation and its potential contribution to meeting the demand or electricity

Analysis o the unit fixed effects or the perormance equations reveals another systematic

pattern that must raise concerns about the uture o the offshore wind industry able 5 (page

46) gives the median values o capacity and the unit fixed effects or onshore and offshore

wind arms by year o commissioning he unit fixed effects used in constructing the table are

based upon the quadratic perormance equations but similar results are obtained i the unitfixed effects or the perormance equations with a ull set o age effects had been used instead

In the case o onshore wind arms both the size o new installations and the associated unit

fixed effects have tended to increase with time Hence the typical onshore wind arm commis-

sioned in 2010 was larger and had a better perormance than the typical installation commis-

sioned in 2000 though onshore wind arms remain very small by UK standards he year 2007

appears to be an anomaly with respect to this general trend but it should be noted that the

median capacity o new plants commissioned in 2007 was only 15 MW well below the medians

or other years in the period 2005-10 he trend in the typical load actor trend over time is

confirmed by estimating regressions or the unit fixed effects with capacity and year o commis-

sioning as regressors In Denmark larger wind arms tend to have higher load actors than small

wind arms Even afer allowing or that trend the normalised load actor or new wind arms

has been increasing at about 1 per year over time his is consistent with the normal pattern

o technical progress and learning which one would expect to observe or a (relatively) mature

industry

he pattern is very different or offshore wind arms In this case the results o the regres-

sions show no significant relationship between capacity and the unit fixed effects combined

with a very substantial deterioration (at rates o 8ndash10 per year) in the unit fixed effects Even

allowing or the small sample o offshore wind arms the trends are so strong that the proba-

bilities o obtaining the results by chance are well below 01 hus the median values o the

unit fixed effects or offshore wind arms shown in able 5 (page 46) illustrate a sharp and

systematic decline in the perormance o new offshore wind arms in Denmark I this were to

continue andor to be reproduced in other countries where offshore wind is being developed

then there can be no prospect that offshore wind arms will ever be financially viable at reason-

able prices or electricity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 42: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4252

40 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 1 Average load factors by year and country ()

Onshore Offshore

EnglandNorthern

IrelandScotland Wales Denmark Denmark

2002 222 264 215 218 261

2003 241 286 249 201 301

2004 250 277 258 228 333

2005 250 308 271 248 221 394

2006 239 294 245 265 202 371

2007 242 263 269 258 247 371

2008 244 294 239 299 231 412

2009 241 300 272 255 213 380

2010 208 235 216 189 210 3982011 266 307 279 270 253 449

Source Authorrsquos estimates See text or source o data

Note he average load actors are the sums o total electricity output by country and year divided by

the average total nameplate capacity in the country multiplied by number o hours in the year For new

installations the first ull month afer the date o commissioning is excluded otal nameplate capacity is

calculated monthly and averaged to give a monthly average o total nameplate capacity

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 43: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4352

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 41

Table 2 Estimation results for UK onshore wind farms

Unweighted Weighted by capacity

Linear Log-Linear Linear Log-Linear

(1) (2) (3) (4) (5) (6) (7) (8)

Age -074 -0036 -131 -0092

(021) (0010) (031) (0020)

Age 2 -0017 -00010 -0043 -00026

(0005) (00003) (0011) (00007)

Age effects

0 -089 -0050 -185 -0103

(034) (0015) (090) (0055)

2 -104 -0066 -187 -0127

(032) (0018) (036) (0025)

3 -189 -0093 -401 -0269

(051) (0024) (074) (0062)

4 -286 -0140 -652 -0446

(075) (0036) (121) (0102)

5 -423 -0200 -900 -0582

(096) (0047) (168) (0135)

6 -452 -0231 -998 -0676

(119) (0057) (203) (0158)

7 -645 -0308 -1302 -0850

(141) (0069) (239) (0186)

8 -761 -0394 -1550 -1033

(160) (0082) (272) (0218)

9 -889 -0465 -1758 -1161

(188) (0097) (311) (0244)

10 -1045 -0574 -1975 -1310

(212) (0117) (339) (0265)

11 -1114 -0591 -2167 -1422

(240) (0120) (369) (0289)

12 -1254 -0632 -2455 -1598

(266) (0131) (416) (0330)

13 -1284 -0666 -2542 -1701

(283) (0140) (457) (0356)

14 -1337 -0708 -2787 -1855

(304) (0148) (478) (0373)15 -1445 -0752 -2878 -1910

(326) (0159) (514) (0395)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 44: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4452

42 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

16 -1627 -0825 -3287 -2155

(342) (0171) (563) (0447)

17 -1695 -0911 -3537 -2372

(371) (0191) (598) (0473)

18 -1899 -0982 -3800 -2482

(396) (0195) (630) (0493)

19 -2240 -1156 -4605 -3011

(463) (0276) (740) (0582)

Constant 2490 2493 3180 3190 2394 2388 3206 3192

(099) (088) (0045) (0038) (074) (0696) (0052) (0052)

Observations 18224 18224 18224 18224 18224 18224 18224 18224

No o units 282 282 282 282 282 282 282 282

R-squareWithin

0653 0657 0549 0555 0687 0700 0585 0607

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)ndash(4) cluster- robust standard errors or models (5)-(8)

(c) he equations are estimated with a ull set o period effects in addition to the variables reported he

deault (missing categories) are age = 1 and period = August 2007 he constant term corresponds to the

estimate o the load actor or ln(load actor) or a unit o age=1 in August 2007

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 45: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4552

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 43

Table 3 Equations for trends in unit fixed effects

Quadratic Age effectsQuadratic ndashcapacity

Age effects ndashcapacity

(1) (2) (3) (4)

Year o commissioning ndash 2000 -0038 -0039 -0117 -0121

(0003) (0003) (0003) (0003)

Capacity (MW) -00021 -00021 -00020 -00020

(00005) (00005) (00005) (00005)

Northern Ireland 0198 0192 0207 0199

(0040) (0041) (0041) (0042)

Scotland 0207 0206 0202 0202

(0036) (0035) (0036) (0040)

Wales 0081 00824 00844 00882(0042) (0042) (0040) (0042)

Constant 0022 00273 0475 0504

(0029) (0032) (0030) (0032)

Observations 282 282 282 282

R-square 0413 0423 0848 0857

Source Authorrsquos estimates

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) England is the baseline country in the model

(c) Bootstrap standard errors

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 46: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4652

44 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 4 Estimation results for Danish wind farms

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

Age -0010 -0013 -0062 -0022

(0004) (0005) (0077) (0020)

Age 2 -00002 -00002 -00067 -00075

(00001) (00002) (00019) (00022)

Age in years

0 -0041 -0036 0036 00155

(0011) (0012) (0090) (0025)

2 -0014 -0020 -0197 -00600

(0009) (0010) (0091) (0032)

3 -0031 -0043 -0127 00784

(0012) (0013) (0200) (0061)

4 -0040 -0053 -0312 -0224

(0015) (0016) (0307) (0112)

5 -0052 -0068 -0374 -0359

(0019) (0021) (0308) (0118)

6 -0069 -0088 -0432 -0432

(0022) (0026) (0380) (0182)

7 -0076 -0097 -0700 -0681

(0027) (0031) (0451) (0192)

8 -0087 -0115 -0838 -0734

(0031) (0035) (0511) (0208)

9 -0100 -0131 -0983 -0828

(0035) (0041) (0590) (0254)

10 -0115 -0151 -1313 -1183

(0040) (0046) (0676) (0270)

11 -0131 -0170

(0044) (0051)

12 -0144 -0184

(0048) (0057)

13 -0154 -0201

(0053) (0062)

14 -0160 -0210

(0057) (0066)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 47: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4752

983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147 45

Regressors Dependent variable ln(load factor)

Onshore wind farms Offshore wind farms

Unweighted Weighted by capacity Unweighted Weighted by capacity

(1) (2) (3) (4) (5) (6) (7) (8)

15 -0184 -0228

(0064) (0073)

16 -0217 -0231

(0083) (0090)

17 -0157 -0179

(0073) (0097)

18 -0188 -0256

(0083) (0095)

19 -0083 -0143

(0082) (0095)

20 -0161 -0246

(0084) (0101)

Constant 3073 3074 3154 3160 3798 3701 3668 3475

(0048) (0045) (0053) (0050) (0179) (0158) -0032 -0056

Observations 93929 93929 93929 93929 2201 2201 2091 2091

No o units 823 823 823 823 30 30 29 29

R-squareWithin

0766 0766 0762 0762 0345 0350 0738 0751

Source Authorrsquos estimates See text or source o data

Notes (a) Standard errors in parentheses Probabilities are measured in stars

p lt 005 p lt 001 p lt 0001

(b) Bootstrap standard errors or models (1)-(2) cluster- robust standard errors or models (3)ndash(8)

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 48: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4852

46 983124983144983141 983120983141983154983142983151983154983149983137983150983139983141 983151983142 983127983145983150983140 983110983137983154983149983155 983145983150 983156983144983141 983125983150983145983156983141983140 983115983145983150983143983140983151983149 983137983150983140 983108983141983150983149983137983154983147

Table 5 Performance by year of commissioning for Danish wind farms

Year Onshore Offshore

Capacity

(MW)Unit fixed effects

Capacity

(MW)Unit fixed effects

Unweighted Capacityweighted

Unweighted Capacityweighted

2000 23 -0055 -0076 400 0001 -0035

2001 26 -0151 -0177

2002 27 0036 0008 220 0196 0215

2003 23 0185 0153 40 -0059 0082

2004 30 0047 0012

2005 36 0278 0241

2006 56 0256 0216

2007 15 -0451 -0494

2008 40 0263 0216

2009 46 0326 0274 232 -0546 -0305

2010 72 0406 0351 391 -0727 -0466

2011 65 0050 -0008 36 -0979 -0673

Note Median values o capacity and unit fixed effects Source Authorrsquos estimates See text or source o data

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 49: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 4952

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 50: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5052

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 51: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5152

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252

Page 52: REF report on Performance of Wind Farms 2013

892019 REF report on Performance of Wind Farms 2013

httpslidepdfcomreaderfullref-report-on-performance-of-wind-farms-2013 5252