THE IMPACT OF EXCHANGE RATE & OIL PRICE RETURNS ON WHOLESALE

ELECTRICITY SPOT PRICE RETURNS: AN EMPIRICAL STUDY

OF EUROPEAN MARKETS

Giorgio Castagneto-Gissey, Richard Green

Imperial College Business School

London, United Kingdom

Presented at the

31st USAEE/IAEE Annual North-American Conference on Energy Economics

November 4-7, 2012

Sheraton Hotel, Austin (TX), US

AUTHORS

CONTACT DETAILS

Giorgio Castagneto-Gissey, M.Sc., Ph.D. Candidate (ESRC)

Imperial College Business School

Tanaka Building,

South Kensington Campus

London SW7 2AZ

United Kingdom

E-mail: g.castagneto-gissey10@imperial.ac.uk

Tel.: +44 (0)79 759 111 01

Professor Richard Green

Imperial College Business School

Tanaka Building,

South Kensington Campus

London SW7 2AZ

United Kingdom

E-mail: r.green@imperial.ac.uk

Tel.: +44 (0)207 594 2611

ABSTRACT

This study investigates the relationship between daily electricity spot price returns and both the crude oil

spot price return (in US dollars) and the exchange rate return in six European countries - namely France,

Germany, Italy, The Netherlands, Spain, United Kingdom – during the time periods 2005-2007 and 2008-

2011 – ie, before and after the contagion of the subprime crisis on European markets. The conditional

mean and conditional variance are modelled by AR(1)-GARCH(1,1) and AR(1)-NGARCH(1,1). The

performance superiority of the NGARCH model suggests that the leverage effect is well represented.

In many cases, the level of returns in either the oil price or the exchange rate had little impact on the level

of electricity price returns. Similarly, the volatility of these prices affected the volatility of electricity

prices only in some of the European markets examined, within the time period preceding the crisis

contagion. However, in the succeeding time frame (2008-2011), the volatility of electricity prices in each

of the studied countries was significantly and asymmetrically affected by both exchange rate and oil price

returns volatilities. This volatility, or price risk, has implications for the way in which low-carbon

generators should be supported.

Keywords: Electricity prices, Oil Prices, Exchange Rates, Volatility analysis, Risk, GARCH regressions

1. INTRODUCTION

Electricity is priced in national currencies but produced using fuels (oil, for example) that are often traded

on world markets and priced in US dollars (USD). We might therefore expect that electricity spot market

prices depend on the exchange rate between the local currency and the USD, and on the oil price. This

paper tests that hypothesis for a sample of EU countries between 2005-2007 and 2008-2011 – ie, before

and after the contagion of the US subprime mortgage crisis on European markets. Five of our sample

(France, Germany, Italy, The Netherlands and Spain) use the euro (EUR), while the United Kingdom

utilises the sterling pound (GBP), as exchange currency.

Muñoz and Dickey (2009) have investigated the relationship between Spanish electricity spot prices, oil

prices and the USD/EUR exchange rate during the period 2005-2007. They showed that these variables

were co-integrated, implying the existence of a long-run equilibrium among them. A transmission of

volatility from the USD/EUR exchange rate and oil prices to Spanish electricity prices was also observed,

although the electricity price level remained unaffected.

This field of research was initiated by Sadorsky (2000), who used co-integration procedures to study the

interaction between prices for crude oil, heating oil and gasoline and a trade-weighted index of exchange

rates. Granger causality analysis found a long-run relationship between the energy prices and exchange

rates, thereby providing evidence in favour of a transmission of exogenous exchange rate shocks to

energy prices.

We hypothesise that the volatility and levels of electricity prices in our sample of European countries can

be significantly affected by oil prices and exchange rates towards the USD. Furthermore, we investigate

whether volatility responds symmetrically to positive and negative oil and exchange rate shocks.

Following Thomas and Mitchell (2007) we use discrete daily returns rather than log-based price relatives,

to allow for the presence of negative prices, which Thomas et al. (2006) had previously identified as a

significant feature of electricity time series data1.

The response of electricity prices to oil prices and exchange rates will depend on the fuel mix in each

country, and on the way in which gas prices respond to changes in the oil price. Figure 1 shows these fuel

mixes. Because the price of electricity is linked to the marginal cost of production, some technologies

with high average shares may rarely set the price – the obvious example of this is nuclear power, which

has very low marginal costs. In contrast, oil-fired stations are frequently reserved for peaking use, running

relatively infrequently but almost always setting the price when they do, so that their share of price-

setting is above their share of generation. Natural gas and coal plants can also set electricity prices – in

most of the countries in our sample, the fuel which is more expensive (per MWh of electricity it can

produce) is the one which will set the price more often. Finally, we should note that while the fuel for

hydroelectric generation (important in several of our countries) is free, stations would offer power to the

market based on the opportunity cost of using their water now, rather than saving it for later use, and that

opportunity cost is based on expected power prices (and hence fossil fuels). We should also note that all

of these countries trade power amongst themselves (European Commission Eurostat: Energy Production

and Imports, 2012), and while prices in adjacent markets can differ when the transmission lines between

them are congested, at other times, the price in, say, the Netherlands may actually be set by the marginal

power station in France.

The dependency of EU countries on energy imports from non-member countries is very large and

increasing over time. However, even when a country is self-sufficient in a fuel, its price may still be set

on world markets (as for oil) or sold at prices indexed to the oil price (as was the case for gas for many

years). In the 1990s, British generators had to pay more than the price of imported coal for large volumes

of domestic output, but power prices were based on the cost of the imported coal that would compete for

marginal supplies. In contrast, gas prices in North West Europe (including the UK) have increasingly

1 We do not in fact observe any negative prices in the daily averages we use, although some hourly prices have been

identified as negative. It is possible that a high level of wind power could produce a negative daily average at some

future time, and we have adopted a technique that be used on such data, allowing cross-period comparisons.

been set in national markets, based on national supply and demand, although obviously influenced by the

cost of imports or potential value of exports – where capacity exists.

In the present study we chose three European countries – which utilise greater amounts of nuclear, hydro

and alternative sources for electricity production (France, Spain and Germany) – and may be less

sensitive to changes in crude oil prices and to the EUR/USD exchange rate, and compared them with

three EU-27 countries which are more dependent on fossil fuels (the Netherlands, Italy and United

Kingdom), see Figure 1. Crude oil prices and the exchange rate will feed through to the cost of oil-fired

power generation in national currencies. The exchange rate will also affect the cost of coal in national

currencies, when this is traded in dollars. Gas prices can depend on the oil price and the exchange rate in

countries where they are still oil-indexed, such as Italy, which still imports much of its gas through

pipelines with long-term oil indexed contracts. In countries with so-called “gas-on-gas” competition,

these factors may have less impact – for example, the price of liquefied natural gas (LNG) imported into

Spain is linked to the price that LNG cargoes could obtain in other markets.

We studied the volatility of electricity prices during two different time frames, the first corresponding to

that studied by Muñoz and Dickey (2005) (January 2005 - September 2007), and the second following

that period - ie, from January 2008 - December 2011. Our first period covers the end of the noughties

boom; the second coincides with the subprime crisis and its aftermath.

Our study examines the volatility of daily electricity price returns using two alternative classes of

volatility models: the Generalised Autoregressive Conditional Heteroscedasticity (GARCH) and the Non-

linear Asymmetric GARCH (NGARCH) models. GARCH models assume that a data series is normally

distributed and that the volatility response to innovations in the market is symmetric. However, empirical

evidence also applies to the present case, suggesting that positive and negative returns of equal magnitude

may not generate the same response in volatility (Black, 1976; Nelson, 1991). The leverage effect, or

negative correlations between returns and volatility, is often observed in financial time series. Some of

these effects can be captured only by non-linear models.

The main contribution of our study is to show that there is a transmission of volatility between both the

exchange rate and oil prices, and electricity prices. Furthermore, we have shown that the electricity price

level is affected by changes in the exchange rate and the price of oil.

The manuscript is organised as follows: data and preliminary statistical analyses are reported in section 2.

Models and econometric methodology are provided in section 3. Section 4 summarises the main findings.

Section 5 discusses the findings in relation to previous studies reported in the literature and provides

suggestions for further related research, before a brief conclusion (section 6).

2. DATA

Daily data relative to the price of the OPEC basket of crude oil, for weekdays covering the two time

frames, January 3, 2005 - September 28, 2007 and January 2, 2008 - December 30, 2011, were obtained

from the Energy Information Administration (EIA) (sourcekey: OILOPEC). For the same periods, time

series relative to spot reference market prices (EUR/MWh or GBP/MWh) for wholesale electricity - for

the following countries: France (Powernext), Germany (EEX), Italy (GME), The Netherlands (APX),

Spain (OMEL), and the United Kingdom (APX UK) - and for the EUR/USD or GBP/USD exchange rates

were obtained from Thomson Reuters.

The choice of the two time frames depends on the intention to study two specific periods: the first being

one previous to the subprime mortgage crisis, whilst the second time frame depicts the period during the

crisis, which hit European countries after its origination in the United States. In fact, Kazi et al. (2011)

estimated the break date of the contagion effect between US stock markets and those of sixteen OECD

countries due to the 2007-2009 global financial crisis, by means of a single break (dynamic conditional

correlation GARCH) model. They found it to be exactly the day of October 1st, 2007, which conveniently

coincides with the end of Munoz and Dickey’s (2009) sample period.

Descriptive statistics for the daily return series are shown in Table 1. The mean, standard deviation,

skewness, kurtosis, Jarque-Bera, Ljung–Box Q tests and Augmented Dickey-Fuller (ADF) statistics are

reported.

An examination of sample autocorrelations, partial autocorrelations, as well as formal unit root tests

revealed that the data were non-stationary in levels. Given the stationarity requirements of the analysis,

the returns of each single variable were computed as

, where where and are

current and one-period lagged prices (applicable to electricity prices, oil prices and exchange rates),

respectively. A log transformation of the series was avoided because it would have reduced the magnitude

of their volatility, possibly masking the statistical relationship investigated (see Karakatsani and Bunn,

2008, and Zhang et al., 2003).

Mean returns are quite small, but the corresponding standard deviations are larger, by an order of several

magnitudes. The distributions of the electricity price returns demonstrate high positive skewness and high

positive kurtosis, as shown by the highly significant Jarque-Bera test. Positive skewness suggests

significant asymmetric response to positive shocks, while the negative skewness values for the EUR/USD

exchange rate suggest a greater probability of large decreases during the sample period. The high value of

kurtosis statistics suggest that extreme price changes occur very frequently. The ADF test clearly rejects

the hypothesis of a unit root in the first differenced series without exception at the 1‰ significance level,

meaning that the electricity price returns are stationary, consistently with a large number of studies in the

literature. The Box-Pierce Q-statistics do not reject autocorrelations up to 20 orders in returns. The returns

are thus serially autocorrelated and subject to time-varying volatility. The evolution of returns of the

electricity prices are graphed in Figure 2A and Figure 2B for the time frames 2005-2007 and 2008-2011

respectively, with their shape suggesting that the time series display volatility and volatility clustering.

3. EMPIRICAL MODELS

The distribution of electricity price returns is asymmetric, as shown by the highly positive skewness and

leptokurtic values for all the examined countries. Spanish and Italian 2008-2011 electricity price data

were corrected by removing outliers - identified as values exceeding by five standard deviations the

autoregressive mean - and were replaced by polynomial interpolation. This follows Trück et al. (2007),

who observed that the robustness of the findings can be improved by removing outliers from the data

before applying a test. In particular, three points were removed and substituted within the Spanish time

series and five points within the Italian one.

Modelling electricity price returns and their volatility consists of two essential steps: the first involves the

specification of the ARMA(p,q) model for mean returns, comprising specific diagnostic tests on the

residuals, whilst the second relates to the specification of the GARCH (p,q) models for conditional

volatility (followed by relative diagnostic tests).

3.1 Conditional mean

We assume that returns follow an AR(1) process with stochastic variance. In fact, the AR(1) model

provides a better fit than the MA(1) model, given that it exhibits the lower standard deviation of the

residual series. Furthermore, the residuals correlogram indicates that an MA(1) model does not provide a

satisfactory fit, as the residual series is clearly not a realistic realisation of white noise. The model for the

conditional mean level of electricity prices is reported below:

After testing the residual series of the model, a significant high order ARCH effect with respect to the

time series studied was observed and, therefore, a generalised autoregressive conditional heteroscedastic

(GARCH) model was applied.

3.2 Conditional variance

In the Bollerslev (1986) GARCH(1,1) model, the equation describing the conditional variance is specified

as follows:

where the real valued parameters ω, α and β satisfy the conditions ω > 0, α ≥ 0 and β ≥ 0 and (α + β) < 1.

However, in many cases it was shown that the sum of the α and β parameter estimates is relatively close

to unity and, therefore, the stationarity condition is violated. The estimate of β allows for an evaluation of

the persistence of the shocks, with an absolute value of β<1 ensuring stationarity and ergodicity for the

model. Often, the β parameter estimate is too large and mistakenly shows an exaggerated volatility

persistence.

The GARCH model assumes the conditional variance is a linear function of the lagged squared returns.

Therefore, one potential short-coming of the GARCH model is the assumption that symmetrically

responds to news about volatility from the previous period.

A special case and alternative of a non-linear GARCH (Higgins and Bera, 1992) is given by:

where ω, α and β are as in Eq. 2, and is an increasing function, similar to the cumulative distribution

function of a positive continuous random variable. The function can be used to allow for a smooth

shift in the parameter ω, which determines the level of the conditional variance, . When

takes a

small value the NGARCH model approaches the GARCH(1,1) model and the relation is symmetric. Since

the G1 function is taken as continuous, the change in the level parameter is smooth, in contrast with the

abrupt change which is observed in the threshold models. The introduction of this function can remove

the non-stationarity behaviour of the conventional GARCH(1,1) model.

In order to ensure a well-defined process, all the parameters of the infinite-order ARCH representation

must be positive. When the γ-parameter, which reflects the leverage effect, is estimated to be positive, it

implies that negative shocks have a greater impact on the conditional volatility compared to positive

shocks.

Both models were estimated by maximum likelihood. Akaike (AIC) and Bayesian (BIC) information

criteria were also used to compare the two models. Since information criteria penalise models with

additional parameters, the AIC and BIC criteria for model order selection are based on parsimony as well

as goodness of fit.

To capture the spillover of exchange rate and oil price return volatility into domestic electricity price

return volatility, in addition to ARCH and GARCH terms, we embed multiplicative heteroscedasticity

components into the standard GARCH model. Following Judge et al. (1985), the functional form of the

multiplicative heteroscedasticity employed in this paper is exponential. Thus, the term

was added to the conditional-volatility equation (Eq.3) to specify the functional form of the multiplicative

heteroscedasticity included within the model.

In Eq. 4, denotes the volatility of external factors, namely exchange rate and oil price returns, the ’s

denote coefficients of the multiplicative heteroscedasticity variables. One of the advantages of using the

exponential function for conditional volatility is that it rules out the possibility of negative variance.

The estimates of and capture the spillover of the external factor volatility into the conditional

variance and therefore, in our case, the effect of the exchange rate and oil price returns on the volatility of

the domestic electricity price returns.

3.3 Sign and Value Expectations of Model Coefficients

For the oil price, for example, a 10% increase would lead to a 2% direct increase in power prices, if oil-

fired power stations set the market price in 20% of the hours – assuming power prices are directly

proportional to the cost of the marginal fuel. If gas-fired stations set the price in another 40% of hours,

and the price of gas is directly indexed to that of oil, we would get an overall effect of a 6% increase, and

a coefficient of 0.6. The coefficient would be lower if gas prices were only partially linked to those of oil.

We measure the exchange rate in terms of EUR (or GBP) per USD. This implies that an increase in our

variable (which implies a depreciation of the EUR) will lead to an increase in the price of electricity (as

imported fuel becomes more expensive) and so the sign should also be positive. A 10% depreciation of

the exchange rate would lead to a 10% rise in the domestic oil price, all else being equal, which would

feed through to a 2% increase in power prices, as in the first example above; however, the exchange rate

would also affect the domestic cost of coal priced in dollars and so the coefficient might be greater than

that for the oil price.

The combined effect of fuel price and exchange rate should be additional in terms of ARCH-in-mean. For

example, if the oil price increases by 20% from $60/barrel to $72/barrel and meanwhile there is a 10%

depreciation of the EUR against the USD (from EUR 1 per USD to EUR 1.1 per USD), then the local cost

of the oil per barrel would be €79.20 instead of €60, which corresponds to an overall price increase of

32%.

4. RESULTS

Tables 2 and 3 present the estimated coefficients, standard errors and p-values for the conditional mean

and variance equations of AR-GARCH and AR-NGARCH models, along with the log likelihood, AIC

and BIC and Q-test, for the two periods of time under inspection. The upper section of Tables 2 and 3

describes the mean equation structure for each country, while in the lower section the results from fitting

the two GARCH specifications are reported. In all models, the lags of the dependent variables are

included as exogenous variables for the underlying equations.

As shown in Tables 2 and 3, the volatility of the exchange rates and oil price returns significantly affects

domestic electricity price return volatility and, in some cases, also its mean level.

In the first period (Table 2), the positive sign and significant t-statistic for the coefficient on the exchange

rate, within the mean equation for the Italian case, indicates that the appreciation of the EUR against the

USD leads to a lower electricity price in that country, as expected; a similar behaviour, but relative to the

increments in oil price returns, is observed for the French electricity price return. In contrast, there is a

statistically significant, and anomalous, negative correlation between the level of the oil price and

electricity prices in The Netherlands. The ARCH-M test, on the other hand, shows that the volatility of

electricity prices affected their level only in Italy in the first period.

In the same period, the conditional variance is significantly influenced by the EUR/USD exchange rate

returns for Spain, - in agreement with the findings of Muñoz and Dickey (2005), in spite of the different

methodological approach used - by the oil price returns for both France and Germany and by both

exchange rate and oil price returns for Italy and The Netherlands, although the magnitude of the effect (ie,

the spillover) is different among the countries. Therefore, the only country with no statistical evidence

that electricity prices are influenced by either the exchange rate (USD/GBP) or oil price returns in the

time frame 2005-2007 is the United Kingdom. We note that electricity prices in Great Britain are strongly

linked to the price of gas there, and that this in turn is set by “gas-on-gas” competition in a spot market,

rather than following the earlier tradition of indexation to oil prices.

As far as the second time frame is concerned – ie, the period 2008-2011, shown in Table 3 - the level of

electricity prices in France is positively affected by the level of the exchange rate, whereas there is a

statistically significant negative relationship in the case of The Netherlands. The level of the oil price has

a statistically significant effect for France. For France and The Netherlands, the t-statistics resulting from

the one-period lagged ARCH-M test are significant for both GARCH models, indicating that there is

statistical evidence implying the conditional variance affects the conditional mean in those countries

(alone). Instead, as far as the conditional variance is concerned, the ARCH (α) and GARCH (β)

parameters are positive and significant in both models, indicating the presence of ARCH and GARCH

effects in the conditional variance equations. The highest own-innovation or ARCH spillovers are

observed for Germany and Spain, indicating the presence of strong ARCH effects. The lagged volatility

or GARCH spillovers are also significant for all countries, but larger in magnitude for Italy, The

Netherlands and the UK. In particular, the GARCH parameter for the UK electricity market is very high

and the sum of α and β is close to unity, suggesting a long time persistence of price volatility. The

GARCH model assumes that the conditional variance is governed by a linear autoregressive process of

past squared returns and variances. Although this model is able to capture heteroscedasticity and volatility

clustering, excess kurtosis with fat tails and the leverage effect of conditional distribution cannot be

described by the classical GARCH model. Accordingly, in the present study the log likelihood, AIC and

BIC criteria provide evidence that augmenting the GARCH model with leverage terms enhances the

model’s performance. In fact, while the sum of the α and β parameters in the GARCH model is in two

cases larger than one, suggesting an explosive ARCH process, the sum of the NGARCH parameters α, β

and γ is always lower than one, indicating model goodness.

The parameter for the asymmetric volatility response (γ) is negative and significant for all the examined

countries, representative of an asymmetric response to positive innovations in the conditional variance

equation. This result is generally consistent with the skewness values reported in Table 1 and reflects the

conditions that electricity price volatility tends to rise in response to positive spikes and falls in response

to negative spikes.

5. DISCUSSION

In the time period 2005-2007, the effect of exchange rate returns volatility on electricity price return

volatility was significant only for Italy, The Netherlands and Spain, but became significant for all the

examined countries within the time frame 2008-2011. This increased volatility transmission may reflect

the greater correlation between a wide variety of markets in the aftermath of the 2008 financial crisis

contagion. There was also more volatility to transmit: the coefficient of variation of the monthly

EUR/USD exchange rate rose by a third between the two periods, while that for the GBP/USD rate

doubled. Exchange rate returns only had a statistically significant effect on the Italian electricity price

level in the first period. In the second period, we find significant coefficients for France and for The

Netherlands (although the latter has the wrong sign). This divergence is surprising, given the strong

process of price convergence which has been observed among the electricity markets of The Netherlands,

Germany and France (see Bijkgraaf and Jansen, 2007).

Except for the UK, at least within the period 2005-2007, oil price return volatility significantly affected

the volatility of electricity prices in the other countries under study. This is not surprising given the

importance of fossil fuels in European electricity production, as shown in Figure 1.

The better performance of the NGARCH- over the GARCH-model indicates that the effect of exchange

rates and oil prices on electricity price volatility is asymmetric. Asymmetric effects imply that exchange

rate and oil price increases have a clear negative impact on electricity price volatility while oil price

decreases do not significantly affect the electricity price. A volatile environment weakens the effect on

price level changes since it reduces the surprise. While there is no literature regarding the asymmetric

effects of the EUR exchange rate against the USD, the presence of asymmetric effects of oil prices on

electricity price volatility was previously described by Hadsell et al. (2004) and Higgs and Worthington

(2005).

It is interesting to note that, despite the different methods used to measure electricity price volatility, our

study exhibits a similar outcome for the Spanish electricity price volatility to that previously shown by

Muñoz and Dickey (2005) – ie, the volatility of Spanish electricity spot prices was affected by the

EUR/USD exchange rate. In relation to the different approaches, while we used GARCH models to assess

the conditional variance and the conditional mean of electricity price returns, Muñoz and Dickey (2005)

defined the current time volatility as the squared difference of present and one-period lagged electricity

prices, to then apply a vector error correction model.

The few other studies reported in the literature regarding the effect of oil prices and/or exchange rates on

electricity price volatility reach differing results. Mohammadi (2009) did not find a long-term relationship

between oil and US electricity prices. In contrast, Narayan et al. (2008) observed that previous values of

oil price and USD/EUR exchange rates affect the evolution of future values of the exchange rate, both in

terms of levels and volatility.

Future studies including a larger number of countries where there is a working electricity spot market to

set the price, such as those in Latin America, Australia and New Zealand, might help to clarify the role of

these or other countries’ currencies exchange rate against the USD and that of the oil price in the

conditional variance and the conditional mean of the electricity price.

6. CONCLUSION

The present study investigates the effect of daily exchange rate returns and crude oil spot price returns on

the electricity spot price return for a sample of European countries. The EU countries investigated were

selected on the basis of their national currency (EUR/USD or GBP/USD) and their dependency on fossil

fuels for electricity production, with France, Spain and Germany using more nuclear, hydro and

alternative sources than UK, the Netherlands and Italy. Furthermore, Spain was also chosen in order to

provide terms of comparison with the results reported in Muñoz and Dickey (2005).

We show that there is in many cases a transmission of volatility between both the exchange rate and the

oil price returns towards the price return of wholesale electricity and that, for some countries, the level of

the electricity price return is also affected by changes in the exchange rate and oil price returns. The

effects become greater for all the countries studied in the time frame 2008-2011, likely as a consequence

of the economic crisis that struck the Eurozone after contagion derived from the US subprime mortgage

crisis.

A volatile wholesale price that will be passed through to electricity consumers implies significant risks to

their bills. It is also unattractive to most low-carbon generators whose costs are not linked to fossil fuel

prices. Ensuring that payments to these generators are largely de-linked from the overall level of power

prices (as with Feed-in-Tariffs or the UK government’s proposed Electricity Market Reform) would

reduce risks for generators and consumers alike.

REFERENCES

1. Black, F. (1976) “Studies Of Stock Price Volatility Changes” Proceedings of the 1976 Meetings of the

Business and Economics Statistics Section, American Statistical Association, pp. 177-181.

2. Bollerslev, T. (1986) Generalized autoregressive conditional heteroskedasticity. Journal of Economics

31, 307-327.

3. Dijkgraaf, E., and Janssen, M.C.W. (2007) Price convergence in the European electricity market.

Erasmus Competition and Regulation, Institute Erasmus University Rotterdam.

4. European Commission. Eurostat: Energy Production and Imports. (2011).

<http://epp.eurostat.ec.europa.eu/statistics_explained/index.php/Energy_production_and_imports>

5. Hadsell, L., Marathe, A., Shawky, H.A. (2004) Estimating the Volatility of Wholesale Electricity Spot

Prices in the US. The Energy Journal 25, 4; ABI/INFORM Global, 23-40

6. Higgins, M.L., and Bera, A.K. (1992) A class of nonlinear ARCH models. International Economic

Review 33:137-158.

7. Higgs, H., and Worthington, A.C. (2005) Systematic Features of High-Frequency Volatility in

Australian Electricity Markets: Intraday Patterns, Information arrival and Calendar Effects, The Energy

Journal 26:1-20.

8. Judge, G.G., Griffiths, W.E., Hill, R.C., Lütkepohl, H., Lee, T.C. (1985) The Theory and Practice of

Econometrics. John Wiley & Sons. New York.

9. Karakatsani, N., and Bunn, D. (2008) Forecasting electricity prices: The impact of fundamentals and

time-varying coefficients. International Journal of Forecasting 24:764-785.

10. Kazi, I., Guesmi, K., and Kaabia, O. (2011) Contagion effect of financial crisis on OECD stock

markets. Discussion paper n. 2011-15. Economics June 6.

11. Muñoz, M.P., and Dickey, D.A. (2009) Are electricity prices affected by the US dollar to Euro

exchange rate? The Spanish case. Energy Economics 31:857-866.

12. Nelson, D.B. (1991) Conditional Heteroskedasticity in Asset Returns: A New Approach.

Econometrica 59: 347-370.

13. Thomas, S., Ramiah, V., Mitchell, H., Heaney, R. (2006) Seasonal Factors and Outlier Effects in

Returns on Electricity Spot Prices in Australia’s National Electricity Market, Working Paper, Melbourne

Centre for Financial Studies.

14. Thomas, S., and Mitchell, H. (2007) GARCH Modelling of High-Frequency Volatility in Australia’s

National Electricity Market. Working Paper, Melbourne Centre for Financial Studies.

15. Trück, S., Weron, R., Wolff, R. (2007) Outlier Treatment and Robust Approaches for Modelling

Electricity Spot Prices. MPRA Working Paper No. 4771.

16. Sadorsky, P. (2000) The empirical relationship between energy futures prices and exchange rate.

Energy Economics 22:253–266.

17. Zhang, L., Luh, P.B., Kasiviswanathan, K. (2003) Energy clearing price prediction and confidence

interval estimation with cascaded neural networks, IEEE Trans. Power Systems 18: 99-105.

ACKNOWLEDGEMENTS

Support from the Economic & Social Research Council (ESRC), via grant no. ES /I029281/1, and from

the Alan Howard Charitable Trust, are gratefully acknowledged.

We would also like to thank Professor Valentina Corradi who provided helpful training on the techniques

used in this paper, and Professor Marcus Miller, for useful discussions (Warwick Economics Department,

2009-2010).

In addition, a worthy mention goes to colleagues of the UK Energy Research Centre (UKERC, Annual

Assembly 2012), and those of the International Association for Energy Economics (IAEE), for helpful

talks.

TABLES

Variables France

EP*

Germany

EP

Italy

EP

Netherlands

EP

Spain

EP

UK

EP

Oil Price

Returns

EUR/USD

Returns

GBP/USD

Returns

Time frame 2005-2007

N 714 714 714 714 714 714 714 714 714

Mean 0.02085 0.03098 0.00683 0.0296 0.00661 0.02107 0.00114 -0.0000657 -0.0000714

SD 0.215 0.299 0.118 0.293 0.117 0.226 0.0187 0.00474 0.00489

Skewness 1.615 6.295 1.756 6.186 0.513 4.559 -0.045 -0.142 -0.025

Kurtosis 6.346 85.697 11.752 79.057 3.283 54.543 0.338 0.796 0.829

Jarque-Bera

P<0.0001

P<0.0001

P<0.0001

P<0.0001

P<0.0001

P<0.0001

P=0.2077

P=0.0022

P=0.0043

Q(20) 116.653

P<0.0001

59.1012

P<0.0001

119.629

P<0.0001

64.454

P<0.0001

83.987 46.531

P=0.0007

12.6095

P=0.8935

11.512

P=0.9319

14.486

P=0.8050

*ADF(20) -6.822

P<0.0001

-5.571

P<0.0001

-7.071

P<0.0001

-5.085

P<0.0001

-6.5901

P<0.0001

-6.3021

P<0.0001

-5.636

P<0.0001

-5.2802

P<0.0001

-6.3203

P<0.0001

Time frame 2008-2011

N 1025 1025 1025 1025 1025 1025 1025 1025 1025

Mean 0.00991 0.00912 0.00654 0.00690 0.00256 0.0109 0.000422 0.000155 0.000268

SD 0.150 0.143 0.120 0.125 0.122 0.158 0.0254 0.00775 0.00778

Skewness 3.716 1.751 1.439 1.359 2.523 2.172 0.422 -0.251 0.116

Kurtosis 37.598 10.777 7.108 7.868 35.574 17.859 6.871 3.269 3.455

Jarque-Bera

P<0.0001

P<0.0001

P<0.0001

P<0.0001

P<0.0001

P<0.0001

P<0.0001

P<0.0001

P<0.0001

Q(20) 97.367

P<0.0001

114.840

P<0.0001

144.422

P<0.0001

121.087

P<0.0001

98.264

P<0.0001

163.085

P<0.0001

36.240

P=0.014

17.924

P=592

54.949

P<0.0001

*ADF(20) -8.382

P<0.0001

-8.424

P<0.0001

-8.142

P<0.0001

-8.251

P<0.0001

-7.151

P<0.0001

-7.038

P<0.0001

-6.153

P<0.0001

-6.218

P<0.0001

-4.906

P<0.0001

Table 1 Summary statistics for returns to Electricity and crude oil prices, and to EUR/USD and

GBP/USD exchange rates. At least three decimal non-zero figures are provided. *EP indicates daily

wholesale electricity spot price return. **

All Augmented Dickey-Fuller test statistics reject the

hypothesis of a unit root at the 1% confidence level.

GARCH model 2005-2007 NGARCH model 2005-2007

France Germany Italy Neth Spain UK France Germany Italy Neth Spain UK

Conditional Mean

Exchange

(β1) Rate

0.227±

1.056

P=0.830

-1.197±

1.159

P=0.302

0.596±

0.181

P=0.015

0.081±

0.958

P=0.933

-0.954±

0.932

P=0.251

1.571±

1.202

P=0.261

-0.462±

1.283

P=0.719

-1.169±

1.159

P=0.313

0.259±

0.067

P<0.0001

0.054±

0.956

P=0.955

-0.923±

0.877

P=0.293

1.840±

1.011

P=0.095

Oil Price

(β2)

-0.194±

0.265

P=0.463

-0.134±

0.292

P=0.646

-0.183±

0.165

P=0.262

-0.0994±

0.0339

P=0.003

0.035±

0.201

P=0.861

0.149±

0.336

P=0.358

0.623±

0.285

P=0.029

-0.197±

0.291

P=0.499

0.396±

0.235

P=0.092

-0.994±

0.249

P=0.004

0.052±

0.201

P=0.795

0.352±

0.219

P=0.091

Cons 0.00918±

0.00598

P=0.126

0.0117±

0.00522

P=0.025

-0.00014±

0.00006

P<0.0001

0.025±

0.007

P<0.0001

-0.0049±

0.0007

P<0.0001

0.0281±

0.0075

P<0.0001

-0.0156±

0.0079

P=0.049

0.0123±

0.00512

P=0.031

1.085±

0.0296

P<0.0001

0.025±

0.008

P=0.002

-0.0089±

0.0003

P<0.0001

0.0246±

0.0081

P=0.002

ARCHM

(ψ)

0.282±

0.805

P=0.783

-0.238±

0.194

P=0.219

-0.377±

0.188

P=0.046

-0.062±

0.054

P=0.245

-1.98±

1.75

P=0.257

-0.233±

0.235

P=0.321

0.399±

0.197

P=0.051

-0.2001±

0.1891

P=0.290

0.391±

0.011

P<0.0001

-0.027±

0.090

P=0.760

-0.897±

2.294

P=0.696

-0.247±

0.232

P=0.287

AR(1) -0.359±

0.0464

P<0.0001

-0.344±

0.0453

P<0.0001

-0.263±

0.021

P<0.0001

-0.283±

0.049

P<0.0001

-0.247±

0.0387

P<0.0001

-0.337±

0.0488

P<0.0001

-0.283±

0.040

P<0.0001

-0.332±

0.042

P<0.0001

-0.224±

0.053

P<0.0001

-0.271±

0.053

P<0.0001

-0.221±

0.043

P<0.0001

-0.342±

0.047

P<0.0001

Conditional Variance

Exchange

Rate

returns

( )

104.91±

52.97

P=0.049

-34.89±

34.39

P=0.310

17.186±

3.721

P<0.0001

13.05±

0.915

P<0.0001

16.74±

6.390

P=0.009

21.858±

28.173

P=0.438

20.502±

15.124

P=0.175

-23.84±

32.666

P=0.465

22.56±

38.20

P=0.451

12.87±

91.28

P=0.764

139.45±

90.947

P=0.222

26.54±

31.236

P=0.475

Oil Price

returns

( )

3.17±

14.95

P=0.832

19.17±

6.41

P=0.003

11.724±

1.677

P<0.0001

-30.38±

3.67

P<0.0001

-4.047±

14.51

P=0.780

42.11±

0.336

P<0.0001

10.005±

13.299

P=0.326

20.971±

6.837

P=0.002

0.316±

0.101

P=0.002

-30.02±

3.685

P<0.0001

-0.858±

13.543

P=0.949

19.91±

14.125

P=0.134

Ω -6.28±

0.269

P<0.0001

-4.92±

0.137

P<0.0001

-5.269±

0.0145

P<0.0001

-4.054±

0.099

P<0.0001

-7.396±

0.315

P<0.0001

-4.643±

0.0180

P<0.0001

-4.071±

0.089

P<0.0001

-4.877±

0.146

P<0.0001

-6.129±

0.0009

P<0.0001

-4.034±

0.0981

P<0.0001

-7.346±

0.293

P<0.0001

-4.678±

0.0240

P<0.0001

Α 0.249±

0.0351

P<0.0001

0.767±

0.0597

P<0.0001

0.611±

0.062

P<0.0001

0.841±

0.051

P<0.0001

0.187±

0.033

P<0.0001

0.324±

0.070

P<0.0001

0.446±

0.064

P<0.0001

0.771±

0.061

P<0.0001

0.731±

0.080

P<0.0001

0.850±

0.057

P<0.0001

0.150±

0.028

P<0.0001

0.335±

0.0704

P<0.0001

Β 0.728±

0.0255

P<0.0001

0.359±

0.0246

P<0.0001

-0.053±

0.0062

P<0.0001

0.126±

0.027

P<0.0001

0.756±

0.031

P<0.0001

0.416±

0.0354

P<0.0001

0.133±

0.061

P=0.027

0.253±

0.025

P<0.0001

0.141±

0.021

P<0.0001

0.117±

0.026

P<0.0001

0.789±

0.0299

P<0.0001

0.411±

0.0382

P<0.0001

Γ -0.081±

0.012

P<0.0001

-0.052±

0.0151

P=0.001

-0.060±

0.009

P<0.0001

-0.0072±

0.018

P=0.691

-0.0123±

0.0035

P<0.0001

-0.0285±

0.0024

P<0.0001

Model Diagnostics

α + β 0.977 1.126 0.558 0.967 0.943 0.740 0.498 0.972 0.812 0.960 0.877 0.718

Q(20) 44.469

P<0.0001

43.346

P=0.0018

58.176

P<0.0001

66.492

P<0.0001

43.346

P=0.0018

65.477

P<0.0001

69.016

P<0.0001

50.745

P<0.0001

36.248

P<0.0001

58.176

P<0.0001

69.016

P<0.0001

69.919

P<0.0001

AIC -4.32.147 -193.355 -153.531 -156.634 -1211.942 -292.518 -430.057 -197.813 -1302.21 -153.530 -1219.45 -289.761

BIC -372.726 -133.934 -89.539 -101.783 -1157.092 -242.238 -366.065 -133.821 -1233.65 -89.538 -1155.45 -231.541

Log(L) 229.074

P<0.0001

109.678

P<0.0001

779.321

P<0.0001

90.317

P<0.0001

617.971

P<0.0001

157.259

P<0.0001

229.029

P<0.0001

112.907

P<0.0001

666.107

P<0.0001

90.766

P<0.0001

623.723

P<0.0001

157.881

P<0.0001

Table 2 AR(1)-GARCH(1,1) and AR(1)-NGARCH(1,1) models of 2005-2007 daily wholesale electricity

spot price returns. The table presents parameter estimates, standard errors and p-values.

GARCH model 2008-2011 NGARCH model 2008-2011

France Germany Italy Neth Spain UK France Germany Italy Neth Spain UK

Conditional mean

Exchange

(β1) Rate

0.733±

0.455

P=0.108

0.315±

0.412

P=0.444

-0.321±

0.365

P=0.378

-0.673±

0.417

P=0.106

-0.019±

0.233

P=0.935

0.543±

0.514

P=0.291

0.2353±

0.0715

P=0.001

0.631±

0.407

P=0.121

-0.0066±

0.437

P=0.988

-0.112±

0.0418

P=0.007

0.154±

0.208

P=0.458

-0.198±

0.500

P=0.692

Oil Price

(β2)

0.404±

0.136

P=0.003

-0.0932±

0.132

P=0.480

0.231±

0.110

P=0.036

0.115±

0.132

P=0.384

-0.023±

0.071

P=0.748

0.304±

0.160

P=0.068

0.840±

0.241

P=0.001

0.245±

0.140

P=0.081

0.132±

0.127

P=0.295

0.0567±

0.142

P=0.690

0.0165±

0.062

P=0.789

0.269±

0.153

P=0.064

Cons -0.0097±

0.0045

P=0.034

0.00333±

0.00394

P=0.398

-0.016±

0.0049

P=0.001

0.0020±

0.0014

P=0.162

0.0054±

0.0040

P=0.181

2.734±

0.889

P=0.002

-0.0675±

0.0017

P<0.0001

-0.0186±

0.0054

P=0.001

-0.0444±

0.0084

P<0.0001

-0.0068±

0.0017

P<0.0001

-0.0130±

0.0097

P=0.112

ARCHM

(ψ)

-0.525±

0.309

P=0.039

-0.893±

0.458

P=0.051

-3.282±

1.847

P=0.074

-0.210±

0.108

P=0.039

0.209±

0.336

P=0.533

-3.090±

1.959

P=0.115

-1.148±

14.100

P=0.935

0.264±

0.828

P=0.750

-0.325±

0.057

P<0.0001

-0.241±

0.107

P=0.017

0.719±

0.368

P=0.051

-2.819±

2.349

P=0.230

AR(1) -0.342±

0.040

P<0.0001

-0.285±

0.037

P<0.0001

-0.398±

0.035

P<0.0001

-0.297±

0.037

P<0.0001

-0.260±

0.041

P<0.0001

-0.356±

0.033

P<0.0001

-0.272±

0.045

P<0.0001

-0.211±

0.046

P<0.0001

-0.304±

0.051

P<0.0001

-0.296±

0.038

P<0.0001

-0.285±

0.040

P<0.0001

-0.362±

0.036

P<0.0001

Conditional Variance

Exchange

Rate

returns

( )

-21.017±

8.539

P=0.014

-2.950±

7.723

P=0.702

-50.702±

28.457

P=0.075

70.58±

13.76

P<0.0001

11.498±

20.291

P=0.571

36.791±

5.774

P<0.0001

8.267±

0.018

P<0.0001

36.989±

7.405

P<0.0001

-26.837±

13.013

P=0.039

45.849±

8.591

P<0.0001

24.329±

9.973

P=0.015

81.770±

8.836

P<0.0001

Oil Price

returns

( )

-0.416±

4.085

P=0.919

7.735±

2.341

P=0.001

-12.498±

8.639

P=0.148

8.867±

3.759

P=0.018

12.074±

1.66

P=0.029

14.493±

4.177

P=0.001

0.167±

0.008

P<0.0001

15.106±

2.229

P<0.0001

9.873±

2.560

P<0.0001

7.911±

2.567

P=0.002

10.423±

2.806

P<0.0001

-21.069±

7.431

P=0.005

Ω -5.624±

0.0908

P<0.0001

-4.923±

0.0094

P<0.0001

-7.208±

0.199

P<0.0001

-5.722±

0.159

P<0.0001

-7.224±

0.166

P<0.0001

-8.396±

0.341

P<0.0001

-4.815±

0.0002

P<0.0001

-5.407±

0.104

P<0.0001

-6.696±

0.183

P<0.0001

-5.886±

0.148

P<0.0001

-6.645±

0.110

P<0.0001

-11.349±

2.026

P<0.0001

Table 3 AR(1)-GARCH(1,1) and AR(1)-NGARCH(1,1) models of 2008-2011 daily wholesale electricity

spot price returns. The table presents parameter estimates, standard errors and p-values.

Α 0.336±

0.0318

P<0.0001

0.438±

0.0489

P<0.0001

0.201±

0.023

P<0.0001

0.279±

0.031

P<0.0001

0.622±

0.039

P<0.0001

0.078±

0.016

P<0.0001

0.0134±

0.0004

P<0.0001

0.179±

0.026

P<0.0001

0.0705±

0.011

P<0.0001

0.055±

0.009

P<0.0001

0.467±

0.036

P<0.0001

0.0052±

0.0106

P<0.0001

Β 0.503±

0.0346

P<0.0001

0.173±

0.055

P=0.002

0.760±

0.0223

P<0.0001

0.476±

0.0504

P<0.0001

0.432±

0.033

P<0.0001

0.907±

0.016

P<0.0001

0.569±

0.0099

P<0.0001

0.496±

0.048

P<0.0001

0.810±

0.026

P<0.0001

0.653±

0.042

P<0.0001

0.444±

0.034

P<0.0001

0.889±

0.014

P<0.0001

Γ 0.00059±

0.00024

P=0.013

-0.00468±

0.00389

P<0.0001

-0.0215±

0.0026

P<0.0001

-0.0333±

0.0035

P<0.0001

-0.044±

0.003

P<0.0001

0.091±

0.014

P<0.0001

Model Diagnostics

α + β 0.839 0.611 0.961 0.755 1.054 0.985 0.583 0.675 0.859 0.675 0.867 0.985

Q(20) 52.667

P<0.0001

37.437

P<0.0001

152.930

P<0.0001

36.851

P=0.012

537.180

P<0.0001

70.526

P<0.0001

52.667

P<0.0001

158.653

P<0.0001

82.352

P<0.0001

365.619

P<0.0001

551.700

P<0.0001

79.274

P<0.0001

AIC -1421.963 -1426.091 -1728.723

-1598.389

-2398.409 -1300.022

-1135.485 -1485.668

-1777.587

-1723.740

-2496.658

-1329.969

BIC -1362.774 -1361.969

-1664.602

-1534.217

-2334.287 -1240.832 -1066.43 -1421.546

-1708.533 -1654.686 -2427.604 -1265.847

Log(L) 722.982

P<0.0001

726.046

P<0.0001

877.362

P<0.0001

812.170

P<0.0001

1212.204

P<0.0001

662.011

P<0.0001

581.742

P<0.0001

755.834

P<0.0001

902.738

P<0.0001

875.870

P<0.0001

1262.329

P<0.0001

677.984

P<0.0001

FIGURES

Figure 1 Fuels used for electricity generation in Europe in 2010 (percentage shares).

Figure 2A Daily data of electricity price returns covering January 2, 2005 - September 20, 2007.

Figure 2B Daily data of electricity price returns covering January 2, 2008 - December 30, 2011.