Workpapers - Southern California Edison · 2016-09-01 · Workpaper – Southern California Edison...

(U 338-E)

2018 General Rate Case A.16-09-_____

Workpapers

RO-Forecast of Sales, Customers and New Meter Connections SCE-09 Volume 01, Chapter V

September 2016

Workpaper – Southern California Edison / 2018 GRC

Exhibit No. SCE-09 / Vol. 01 / Chapter V Witness: H.Sheng

58

V. 1 FORECASTS OF SALES, CUSTOMERS, AND NEW METER CONNECTIONS 2

This chapter presents the forecasts of: (i) retail electricity sales, (ii) customers, and (iii) new 3

meter connections in SCE’s service area for 2016-2020.66 It comprises a summary of the forecasts and a 4

brief description of the methodology used to produce each forecast. This section also briefly describes 5

the major factors and assumptions that influence each forecast. 6

A. Sales Forecast 7 Total 2015 retail electricity sales in SCE’s service area were 86,856 GWh. We are predicting 8

2016 sales of 84,312 GWh, 84,253 GWh in 2017 and 83,572 GWh in 2018 (0.4% annual decline). The 9

forecast decline in sales between 2015 and 2016 is attributable primarily to two reasons. One is an 10

assumption of normal weather in 2016, compared to the hotter-than-normal weather experienced in 11

much of SCE’s service area during summer 2015. The other main reason is increased behind-the-meter 12

(BTM) solar photovoltaic (PV) generation. The economy has recovered slowly following the 2007-2009 13

Great Recession but is projected to pick up with the anticipated housing recovery over the next few 14

years within SCE’s service territory. However, the rapid increase in customer adoption of BTM solar PV 15

systems has reduced customer need for utility-supplied energy. 16

1. Methodology 17 SCE uses econometric models to forecast monthly retail electricity sales (recorded sales 18

as billed and measured at the customer meter) by customer class. Retail sales include final sales to 19

bundled, direct access, and Community Choice Aggregation customers within the SCE service area. 20

Retail sales do not include sales to public power customers, contractual sales, or inter-changes with 21

other utilities, since these are not considered final sales to SCE’s customers. 22

The retail sales forecast represents sales to seven customer classes: residential, 23

commercial, industrial, other public authority, agriculture, street lighting and inter-department transfers 24

(IDT). Each customer class forecast (except for IDT) is itself the product of two separate forecasts: a 25

forecast of electricity consumption per customer or per building square foot (depending on the type of 26

customer) and a forecast of the number of customers or total building square feet. The IDT sales 27

66 Reference herein to customers refers to customer accounts.

1



59

forecast, which represents a small percentage of total retail sales, is based on the average of recorded 1

monthly sales over the most recent 12-month recorded period. 2

Econometric models employ statistical techniques to quantify the relationships between 3

electricity consumption and the economic, demographic and other factors that influence electricity 4

consumption. Examples of such variables include weather, electricity rates, number of billing days, 5

efficiency index, employment, regional output, and building square footage. Historical data are used to 6

determine these relationships. The typical estimation procedure used to construct these models is 7

ordinary least squares (OLS). 8

Once a satisfactory statistical relationship is established, SCE uses historical average 9

values of weather (specifically, cooling and heating degree days) and the number of billing days to 10

represent typical or normal conditions in future periods. Forecasts of economic drivers such as 11

employment, regional output, and building square footage, with the typical weather and billing day 12

variables, are then added to the models to derive forecast values of electricity consumption per customer 13

and per building square foot. Economic data vendors, such as Moody’s Analytics,67 California 14

Economic Development Department (EDD), and Dodge Data & Analytics are the principal sources of 15

the historical and forecast employment, income and building square footage data that SCE employs in its 16

sales forecast. Model-generated forecasts may be modified as needed based on current trends, judgment, 17

and events not specifically modeled in the econometric equations. 18

A different set of models is used to forecast the number of customers by customer class. 19

Forecasts of residential and non-residential customers are based on econometric models that relate 20

changes in the number of customers to the changes in economic activities. With residential customers, 21

housing starts are the leading economic variable to forecast the number of customers. For small 22

commercial customers, changes in the number of small commercial customers are assumed to be 23

influenced by changes in the number of residential customers. Manufacturing and agricultural 24

employment data are the leading economic variables to forecast the number of industrial and agricultural 25

customers, respectively. Building square footage data from Dodge Data & Analytics is used to forecast 26

the total building square footage for other public authority and industrial customers. 27

67 SCE decided to use only forecasts from Moody’s Analytics for its economic driver outlooks rather than using

the average forecasts from both Moody’s and IHS Global Insight (as it has done in past GRCs). In general, Global Insight has an overly optimistic housing recovery outlook for Southern California.

2



60

2. Historical Trends 1 On a recorded basis, SCE’s total electricity sales decreased at an average annual rate of 2

0.2 percent per year during the period of 2007 to 2015. However, sales growth has not been consistently 3

negative during this period. Annual sales decreased meaningfully in 2009 and 2010, about a 3.8 percent 4

average annual drop for those two years due to the Great Recession. Annual retail sales grew as the 5

economy slowly recovered. Annual sales grew at a relatively slow pace of 0.8 percent in 2011 but grew 6

at a faster clip (an average of 1.4 percent) between 2011 to 2014, mainly due to above-average 7

temperatures. 8

Customer growth, which is measured by the number of customer accounts, has remained 9

positive during the post-recession period of 2009 to 2015 averaging 0.5% annual growth. In 2014 and 10

2015, annual customer growth reached 0.6%, the highest level since 2007 but well below the levels 11

reached during the regional housing boom period of 2005 to 2006. 12

The year-over-year percent change in total non-farm employment in the counties served 13

by SCE during the years 2006 to 2015 is shown in Figure V-3. Non-farm employment dropped sharply 14

between 2007-2010 as a results of the Great Recession. However, starting in 2011, counties in the SCE 15

service area have experienced positive non-farm employment growth, although not at pre-recession 16

levels. Employment growth peaked in late 2012 and early 2013, at the same time as housing starts lifted 17

from recession-level lows. Year-over-year job growth has slowed since 2013 but remains positive. 18

As shown in Figure V-4 below, the number of housing construction starts according to 19

Moody’s Analytics experienced steep declines prior to and during the Great Recession driven by the 20

housing crisis. Since 2011, the housing sector has shown healthy signs of improvement. Housing starts 21

have been increasing steadily from 2010 to 2014, with an average annual growth rate of 35 percent. 22

Multifamily construction growth was a primary source of the housing start increases during the 2010 to 23

2014 period. The acceleration in the housing recovery anticipated by economic forecast vendors 24

(including both Moody’s Analytics and IHS Global Insights) didn’t materialize in 2015 but SCE expects 25

it to materialize in 2016. With pent-up demand in Southern California metro areas for new housing, 26

there are still expectations that the market for new houses will pick up again during the 2016-to-2018 27

period as job markets and overall economic conditions improve further in Southern California. 28

3



61

Figure V-3 Total Non-Farm Employment Growth in the Counties

Served by SCE

Figure V-4 Housing Starts in the SCE Service Area

4



62

3. Economic Outlook 1 Economic recovery in Southern California since the end of the Great Recession has been 2

smaller and slower than anticipated in earlier economic forecasts. The main contributing factor is that 3

the housing recovery in the region has been less robust and later than anticipated as evidenced by actual 4

economic data including housing starts. Despite the slump in homebuilding activities in 2015, housing 5

starts are expected to ramp up quickly in 2016 and continue to increase in 2017. Housing starts are 6

projected to reach their peak in 2017 with 54,650 new units forecast for all counties served by SCE, 7

which is still well below the pre-recession levels. 8

Along with the partial recovery in housing starts, it is expected that the Southern 9

California economy will further expand in the 2016-2018 period with annual non-farm employment 10

growing slightly above two percent from 2015 to 2018. As shown in Figure V-5, total non-farm 11

employment is projected to remain above its pre-recession peak level for years 2016 to 2020. 12

5



63

Figure V-5 Total Non-Farm Employment Growth in SCE Service Area,

Actual and Forecast

4. Weather Assumptions 1 SCE uses 30-year average temperature conditions to define “normal” weather. Normal 2

weather conditions are assumed throughout the forecast period. For model estimation and forecasting, 3

actual and normal temperature data are transformed into cooling degree days (a measure of summer 4

season cooling load) and heating degree days (a measure of winter heating load).68 As shown in Figure 5

V-6 below, the SCE service area experienced higher-than-normal cooling degree days for four 6

consecutive years 2012 to 2015. Because the forecast for 2016 and beyond assumes normal weather, a 7

slowing trend is apparent in the 2016 forecast, as electricity sales transition from 2015, a year of 8

warmer-than-normal summer weather conditions, to normal weather conditions in 2016. 9

68 Cooling degree days (CDD) are defined at a base of 70 degrees Fahrenheit whereas heating degree days

(HDD) are defined at a base of 65 degrees Fahrenheit.

6



64

Figure V-6 Recorded and Normal Cooling Degree Days as a Percent of Normal Days

5. Other Factors Influencing the Forecast 1 Other factors influencing total retail sales during the 2016-2020 period are electricity 2

rates, energy efficiency programs, transportation electrification load (including electric vehicle-charging 3

load), and self-generation, such as residential rooftop solar and combined-heat and power installations.69 4

a) Electricity Rates 5

SCE’s average system electricity rates were relatively constant in current dollars 6

between 2009 and 2011. Rates increased from 2012 to 2014 mainly due to the 2012 GRC and increasing 7

fuel and purchased power costs. Average rates started declining in 2015 and are expected to do so in 8

2016 again before increasing slightly in 2017. The average electricity rate is expected to return to the 9

2014 level by 2018. The forecast increase in rates in 2017 and 2018 is mainly driven by the 2015 GRC 10

Decision and the 2018 GRC request and rising fuel and purchased power costs plus the elimination of 11

69 SCE has also incorporated limited behind-the-meter energy storage in its forecast. This is based on SCE’s

2014 LCR RFO contracts.

7



65

refunds in 2016 and 2017 related to ERRA and GRC over collections. Historical and forecast system 1

average prices are shown in Figure V-7below. All other things being equal, higher electric rates result in 2

a reduction in the average electricity use per customer. 3

Figure V-7 Average System Electricity Price, Actual and Forecast

b) Energy Efficiency 4

Energy efficiency (EE) savings include savings from both utility-funded programs 5

and codes and standards approved by federal and state governments. EE has dramatically reduced 6

customers’ total consumption due to the significant savings achieved cumulatively over the recent few 7

decades. As a result, SCE’s historical retail sales data already reflect reduced energy consumption from 8

past EE programs. SCE’s retail sales would have been much higher in absence of the different EE 9

programs. SCE reflects most of the energy efficiency trends and impacts through estimating its customer 10

average usage (kWh consumption per customer) equations and using SCE’s retail sales data, which has 11

EE savings embedded already. Additionally, anticipated EE savings exceeding historical norms are 12

deducted after-the-fact on an annual incremental basis. 13

8



66

c) Thermal and Solar Photovoltaic Self Generation 1

Savings from customer on-site behind the meter bypass self-generations including 2

thermal self-generation and solar PV generation under net metering programs represent energy 3

consumption that would have taken place in the absence of the bypass self-generation. 4

Methodologically, historical savings from self-generation are added back to recorded retail sales to 5

generate a hypothetical level of “energy consumption” that would have taken place in the absence of the 6

programs. Then a forecast level of self-generation savings are deducted from the hypothetical “energy 7

consumption” amount to yield the forecast of retail sales. 8

SCE accounts for all existing thermal self-generation that are already operating 9

behind customers’ meters within SCE’s system. In addition, SCE projects future thermal self-generation 10

growth consistent with the historical trends. Combining the system level forecast with customer 11

information, SCE allocates the thermal self-generation forecast down to the circuit level.70 12

There were 154,395 total cumulative residential solar PV installations in the SCE 13

service area by the end of 2015. SCE predicts annual incremental installations of 73,008 in 2016, 92,701 14

in 2017, and 105,666 in 2018 as shown in Figure V-8 and Table V-19. SCE utilizes a generalized bass 15

diffusion process to model residential customer adoption of solar PV systems. The model utilizes the 16

solar PV system costs adjusted for the federal Investment Tax Credit as an explanatory variable. SCE 17

fits the model with the historical customer adoption data (up to the end of 2015) and uses it to predict 18

future customer adoption over the SCE service territory. In addition, SCE takes into consideration of the 19

impacts from the recent tier rate changes and California’s Zero Net Energy (ZNE) mandate.71 Based on 20

the total projected annual residential solar PV generation installations, SCE expects the annual growth 21

rate to be 47% in 2016, 41% in 2017, and 33% in 2018 across SCE’s service territory. This solar PV 22

forecast is then adapted for allocation down to the circuit level for distribution planning analysis 23

purposes.72 24

70 Refer to WP SCE-09 Vol. 1, Chapter V, p. 7. 71 The move to a two-tier rate structure dampens PV adoptions from what they otherwise might have been in

this period from 2016-2018. The Zero Net Energy (ZNE) mandate increases the forecast beginning in 2020 when it takes effect. SCE estimates this leads to an increase in residential solar PV capacity of 1,172 MW in 2016, 1,658 MW in 2017, and 2,214 MW in 2018.

72 Refer to WP SCE-09 Vol. 1, Chapter V, p. 7.

9



67

Figure V-8 Residential Solar PV Annual Incremental Installations and Cumulative Installed

Capacity; History and Forecast

Table V-19 Annual Residential Solar PV Generation, Installations, and Cumulative Capacity

Year

Annual Residential Solar PV

Generation (GWh)Incremental Installations

Cumulative Capacity (MW)

2012 340 14,776 194 2013 568 25,630 324 2014 879 33,886 501 2015 1,387 55,798 791 2016* 2,056 73,008 1,172 2017* 2,907 92,701 1,658 2018* 3,881 105,666 2,214 2019* 4,927 111,763 2,811 2020* 6,021 116,253 3,434 *forecast

10



68

d) Transportation Electrification Load 1

SCE forecasts future transportation electrification load growth for both light duty 2

EV load and other non-light duty electric transportation load. As a nascent and dynamic market affected 3

by several exogenous variables such as manufacturer supply, local, state and federal policy, and 4

technology advancement, plug-in electric vehicle (PEV) forecasting is treated separately as a positive 5

load contributor. SCE has relied on expert forecasts of plug-in electric vehicle adoption. SCE expects 6

the number of light duty EVs will reach 197,612 by 2018 as shown in Table V-20. The annual PEV load 7

growth rate is forecast to be 73% in 2016, 44% in 2017, and 35% in 2018. 8

Table V-20 Annual Plug-in Electric Vehicle Load (GWh) and Annual Cumulative Plug-in

Electric Vehicles

SCE’s forecast load from non light-duty vehicles relies heavily on E3 and ICF 9

International’s 2014 California Transportation Electrification Assessment (TEA Study).73 The forecasts 10

published in the TEA Study were narrowed down to SCE’s territory based on SCE’s share of economic 11

activity for each vehicle segment in California. The combination of these two types of electrification 12

load forecasts are then incorporated into SCE’s overall load forecast. 13

6. Total Retail Sales Forecast by Customer Class 14 Table V-21, below, presents SCE’s 2015 recorded and 2016-2020 forecast of total 15

electricity sales by customer class. The projected average annual growth in total retail sales is about 16

negative 0.4 percent per year from 2016 to 2018, compared to the positive 0.2 percent average annual 17

growth experienced between 2009 and 2015. 18

73 These non-light duty EVs include but are not limited to forklifts, truck stop electrification spaces, transport

refrigeration units (reefers), shore power, port cargo handling equipment, airport ground support equipment, and high speed rail.

Year Annual Load Cumulative PEVs2015† 222 58,718 2016* 384 101,633 2017* 552 146,049 2018* 747 197,612 2019* 954 252,799 2020* 1,178 312,581 †estimate*forecast

11



69

Table V-21 Annual Retail Sales by Customer Class (GWh)

B. Customer and New Meter Connection Forecasts 1 Table V-22 and Table V-23 present the forecasts of total electricity customers and new meter 2

connections by customer class for 2016-2020 and 2015 recorded. Both new customers and new meter 3

connections are closely tied to activity in the residential construction sector, usually with a lag of up to 4

18 months, meaning that a change in the number of new meter connections or new customers is typically 5

a result of a change in the number of housing starts that occurred up to 18 months earlier. Our forecast 6

of new customers and new meter connections follows closely the housing market cycle described above. 7

Over the period from 2016 to 2018, total projected customer growth averages about 0.8 percent per year, 8

which is slightly higher with the historical 0.5 percent average annual growth recorded from 2009 to 9

2015. 10

Table V-22 Year-End Customers by Customer Class

New meter connections dropped to a low of 19,800 in 2011 as a result of the housing market 11

collapse. Since then, new meter connections have been increasing steadily. Total new meter connections 12

12



70

reached 31,653 in 2015. Given the housing construction outlook, SCE expects the number of new meter 1

connections to increase to 36,336 in 2016, 40,519 in 2017, and 48,848 in 2018. This in line with 2

Moody’s Analytics’ housing market (including starts) outlook.74 3

Table V-23 New Meter Connections

74 Housing starts were forecast to grow at about 1.9% compound annual growth rate (CAGR) from 2015 to 2018

according to Moody’s Analytics.

13

2018 General Rate Case Index of Workpapers

SCE-09, Vol. 01, Chapter V

DOCUMENT PAGE(S)Introduction 2 Forecast Assumptions and Drivers 3-8 Circuit Level Distributed Energy Resource (DER) Forecast 6-8 Historic Forecast Performance 8-9 Weather Adjustment Procedures 9-10 Forecast Uncertainty 11-12 Flow Diagram for Electric Use and Customer Modeling & Forecasting 13-14 Model Statistics – Electricity Use Models 15-26 Electricity Use Model Variable Description 27-32 Model Statistics – Customer Models 33-45 Customer Model Variable Description 46-48 Model Statistics – Residential Meter Connection Models 49-51 Residential Meter Connection Model Variable Description 51 Non Residential Meter Connection Models 52-53 Non Residential Meter Connection Models Variable Descriptions 54 Sales and Customer Model Data 55-503


Exhibit No. SCE-09 / Vol. 01 / Chapter V Witness: H. Sheng

1

SCE 2018 GRC Retail Sales and Customer Forecast Methodology

SCE-09, Volume 01, Chapter V,

Work Papers

Southern California Edison

1



2

1) Introduction

SCE uses econometric models to develop its retail sales forecast – a forecast of monthly retail electricity sales (billed recorded sales measured at the customer meter) by customer class. Retail sales are final sales to bundled, Direct Access (DA), and Community Choice Aggregate (CCA) customers. DA and CCA sales are subtracted from the retail sales forecast in order to derive to the forecast of SCE bundled customer sales. Retail sales exclude sales to public power customers, contractual sales, resale city sales, municipal departing load and inter-changes with other utilities.

The retail sales forecast represents the sum of sales in six customer classes:

residential, commercial, industrial, public authority, agriculture and street lighting. Each customer class forecast is itself the product of two separate forecasts: a forecast of electricity consumption and a forecast of the number of customers1. Customer class data are used because they have been defined in a consistent manner throughout the sample period used in the econometric estimation.

In addition to the categorization by customer class, residential sales are further

modeled and forecasted according to geographical region. The SCE service area encompasses several distinct climate zones. Accordingly, we model residential electricity consumption in part to capture regional variation in the weather/consumption relationship. Additionally, the commercial customer class is modeled and forecast according to a small and large customer criteria2. We find that small and large commercial customers respond differently, from an electricity usage standpoint, to changes in weather, rates and economic conditions.

The electricity consumption per customer or per square foot forecasts are

produced by statistical models that are based upon measured historical relationships between electricity consumption and various economic factors that are thought to influence electricity consumption. The estimation procedure used to construct these statistical models is ordinary least squares (OLS). Another set of econometric equations are used to forecast customers by customer class (in most cases customer additions are modeled (the change in the number of customers in the current month and the previous month) and converted into a forecast of total customers).

The regression equations, combined with forecasts of various economic drivers, such as employment and output, along with normal weather conditions and normal number of days billed, are used in combination to predict sales by customer class. Model-generated forecasts may be modified based on current trends, judgment, and events that are not specifically modeled in the equations. Direct Access and Community Choice Aggregate By the end of 2013, DA reopening to non-residential customers was completed. In the near term, SCE is expecting no near-term increases in DA load. SCE had its first departing Community Choice Aggregate (CCA) load starting in May 2015 in the form of Lancaster Choice Energy (LCE). SCE has incorporated its best estimate of the migrating CCA load to this application based on the best information SCE had received at the time 1 Electricity usage of residential, agriculture, commercial, and streetlights service accounts is forecasted by consumption per customers. Electricity usage of industrial and public authority (OPA) service accounts is forecasted by usage per square footage. 2 Small customers are generally those in the GS-1 and GS-2 rate categories while large customers are typically time-of-use (TOU) rate class customers.

2



3

that this forecast was made. As a result, SCE’s bundled sales growth has been reduced relative to retail sales growth. 2) Forecast Assumptions and Drivers

The underlying assumptions regarding the economy, weather, electricity prices, conservation and self-generation are all significant factors affecting the sales forecast. Each of these important variables is discussed briefly below. Employment

Changes in employment are an important source of explanatory power in measuring and predicting variation in electricity consumption. Changes in employment cause both seasonal variations in electricity consumption and changes in the long-term rate of growth in consumption over the forecast period.

SCE matches employment on a sectoral basis with electricity consumption by

customer class. Specifically, private commercial services employment in counties served by SCE is assumed to explain changes in SCE commercial class electricity sales. Manufacturing employment contributes to the explanation of changes in industrial class electricity sales. Government employment (federal, state and local) is used to model public authority (federal, state, or local government) customer class electricity sales. Agriculture employment is used to help explain changes in agriculture customer class sales.

Historical employment data by county is obtained from the California Economic

Development Department (EDD). Moody’s Analytics (MA) provides forecast employment data for California. The EDD historical data used is non-seasonally adjusted. The MA employment forecast data is converted from seasonally adjusted to non-seasonally adjusted.

The short-run elasticity value for the impact of employment growth on the

electricity consumption of small commercial customer class is about 3.1, for the large commercial customer class about 4.3, manufacturing employment is about 0.6, government is about 9.4, and agriculture is about 8.7.

Weather

SCE uses 30 year average temperature conditions to characterize normal

weather. Normal weather conditions are assumed throughout the forecast period. For purposes of model estimation and forecasting, daily actual and normal temperature data are transformed into monthly cooling degree days (CDD), that is the summer season from April to October, and heating degree days (HDD), meaning the winter season from November to March. A base temperature of 70 degrees F is used to calculate monthly cooling degree days and a base temperature of 65 degrees F is used to calculate monthly heating degree days. The CDD and HDD variables used in model estimation are based on daily temperatures that are a weighted average of 10 stations located in the SCE service area. The station locations are Ontario, Thermal, Long Beach, Riverside, Burbank, Santa Ana, Oxnard, Fresno, Lancaster and Los Angeles International Airport.

An important aspect in the calculation of CDD/HDD is the weights attached to the

weather stations. The weather station weights reflect the historical geographical customer distribution. SCE customer growth is increasing faster in the areas experiencing higher temperatures in the summer and lower temperatures in the winter and thereby have a higher frequency of cooling and heating appliances.

3



4

In the residential models, the stations selected represent temperatures in the counties served by SCE. For example, the residential Orange County model uses a customer weighted average of temperatures recorded by the Santa Ana, Long Beach and Riverside weather stations. The non-residential identical sales models are estimated with customer and appliance weighted CDDs/HDDs. Commercial, industrial, and public authority sales models are estimated using only the customer adjusted CDDs/HDDs.

Since normal weather is assumed throughout the forecast, weather variation

generates a seasonal pattern to electricity use but has only a small influence on the trend. More detail on weather normalization is provided below.

Billing Days

We define billing days as the sum of the number of calendar days between meter

reads for each of the meter read cycles. There are typically 21 meter reading cycles to a month. The number of days for which a customer is billed can vary depending upon meter reading schedules in a month and the number of holidays and week end days in a month. Recorded sales will therefore vary with the number of days billed. The average number of billing days in a month turns out to be a very important source of explanatory power in all the electricity use models. For purposes of the forecast, we assume the historical average number of billing days in each month. Like weather, billing days explains variation in use over the months in a year, but does not contribute to trend growth in electricity consumption.

Electricity Prices

It is typically difficult to estimate a statistically significant relationship between changes in electricity consumption and changes in electricity prices. There are a number of reasons for this. First, electricity prices are regulated and therefore may vary only infrequently. Second, price signals between electric utilities and consumers can be obscured by lags in the transmission of price information and the complexities inherent in tariff structures. We attempt to simplify these issues by using an average unit revenue price with a one period lag (with the exception of the industrial electricity consumption model, which do use current period rates). Finally, electricity consumption is considered to be a necessity good, which means that consumption is relatively unresponsive to changes in price, at least in the short-run. In other words, the short-run residential price elasticity of demand, as derived from our forecast models, is generally in the range of --0.11 to -0.01. For purposes of model estimation, electricity prices are derived as monthly utility revenue divided by kWh consumption (i.e., unit revenue prices) and deflated by a consumer purchasing index in order to express rates in constant dollars.

Real Output

Real output serves much the same purpose in the residential electricity consumption model that employment does in the commercial and industrial electricity consumption models: Changes in output per capita explain a significant amount of the variation in residential electricity consumption that is due to changes in economic conditions. This was particularly true during the 2003 to 2007 period – a period of robust economic growth, and the period 2008 to the present, which saw a sharp decline in real output due to high levels of unemployment and depressed real estate prices. Although changes in real output explain some of the seasonal variation in residential electricity consumption, it is really a major determinant of the long-run growth trend in residential electricity consumption. Real output elasticities are typically in range of 0.8 to 2.0. We use historical and forecast real income per capita by metropolitan statistical area from Moody’s Analytics in our regional residential OLS forecasting models. In the case of Riverside and San

4



5

Bernardino counties, MA’s combined Inland Empire MSA forecast was divided using a 10-year compound average growth rate based on historical county employment data obtained from CA EDD.

Self-Generation

The forecast of customer on-site bypass self-generation is calculated from two lists of customers operating generating systems interconnected to the SCE grid for the purpose of meeting their own energy requirements: a thermal list and a solar list. Both customer lists identify those customers that have behind-the-meter systems on-line, under construction or current plans to install. The description of each facility includes designation of customer class, nameplate capacity in kilowatts (KW), probable bypass KW, capacity factor and on-line date. Separate forecasts are developed for thermal and solar/renewable systems and then combined for use in the sale forecast. Both lists are used to estimate annual energy production by customer class, which is allocated to the months in the year. For thermal generation, the annual energy is calculated using the bypass capacity and a high capacity factor for all hours of the year. The annual energy is distributed to the months using a thermal load shape based on typical TOU-8 customer load shape, modified to be fully online during the on-peak periods from June into October of each year. The hourly loads are summed by month in order to produce a thermal by-pass consumption variable. There will be approximately 158,000 operational behind-the-meter solar systems by the end of 2015 ranging in size from 1KW to more than 1,000 KW within the SCE service area. For the solar generation, the annual energy is calculated using the bypass capacity and annual capacity factors. The capacity factors are taken from the CPUC Self-Generation Incentive Program, Fifth Year Impact Evaluation, Draft-Final Report prepared by in February 2007 by Itron for PG&E and the Self-Generation Incentive Working Group. Annual energy is distributed to the months of the year using a load shape based on hourly distribution. The monthly thermal and solar by-pass variables are summed for a single by-pass variable suitable for inclusion in the sales forecasting models. Residential Solar Photovoltaic SCE models the residential adoption of solar photovoltaic through a generalized Bass diffusion model.3 The Bass diffusion model is a standard technology adoption model originally developed in 1969.4 The SCE model uses percentage changes in the price-per-Watt-AC of installation, adjusted for the Federal Investment Tax Credit, as its explanatory variable. Bloomberg New Energy Finance (BNEF) provided SCE’s historical and forecast solar installation price series from 2010-2020.5 The compound monthly growth rate was used to extend this series back to 2000. Residential solar photovoltaic adoption history comes from SCE’s internal net energy meter (NEM) database. As this model is essentially a regression, expected policy changes in the future which are not reflected in the history require post-model adjustment. To reflect the zero net energy mandate, which requires new homes be designed to use zero net energy from the grid starting in 2020, we add half of the forecast of new single-family homes to the model forecast. We only apply fifty percent of new single-family houses for two reasons. The 3 Bass, Frank M., Trichy V. Krishnan, Dipak C. Jain. “Why the Bass model fits without decision variables.” Marketing science. Vol. 13, No. 3, Summer 1994. 4 Bass, Frank. “A new product growth for model consumer durables.” Management science. Vol. 15, Issue 5, 1969. 5 “H2 2015 US PV market outlook: Brace yourself.” Bloomberg new energy finance. 9 November 2015.

5



6

first reason is that the mandate does not require all new houses install solar photovoltaic, and by 2020 there may be other forms of distributed generation that are competing for market adoption. The second reason is we cannot, at this time, be certain of the compliance rate from the home building industry. Our new single-family housing forecast is supplied by Moody’s Analytics. From 2016 to 2018, the annual incremental adoptions were decreased by 4.55% to reflect the effect of the implementation of a two-tier rate scheme. Transportation Electrification Beginning in 2016, SCE updated their light-duty PEV forecast methodology to incorporate actual market adoption information from the previous five years and narrowed the number of studies used to forecast PEV population numbers through 2035.6 SCE uses average year over year growth rates from five expert studies.7 Using growth rates, as opposed to discrete population forecasts allows SCE to evenly weight each study, eliminate variances in starting years and better account for the underlying assumptions in each forecast. This single averaged annual growth rate was then applied to actual adoption that had been realized in SCE territory from 2012 to 2015. Once population numbers are determined for each year, several variables are then applied to determine hourly, daily, and annual electricity load shapes.8 Electricity Conservation Programs

SCE no longer takes the position that energy efficiency (EE) should be explicitly included in the econometric estimation of kWh consumption per customer. Instead, EE is omitted from econometric estimations and is deducted after the fact on an incremental basis as needed. Other EX Post Modifications to the Sales Forecast SCE makes some additional adjustments to the customer class sales forecast produced by the econometric models. The primary reason for this is that these components are all relatively new phenomena and thus cannot be explicitly modeled in the econometric equations. These components include PEV charging, other new electric technologies such as high speed rail and other electrified rail transport, shipping port electrification, industrial uses such as electrified forklifts and truck stops. 3) Circuit Level Distributed Energy Resource (DER) Forecast

In general, the process that SCE used to perform this circuit allocation is based on

DER potential. We identified the types of customers who have the greatest economic potential and/or interest in installing DERs, inventoried the dispersal of these customers across SCE’s individual distribution circuits, and then allocated the quantity of DERs to distribution circuits in proportion to the amount of customers with DER potential on these circuits. As a result, the DER allocations described in this section are unconstrained by any limitations of the existing distribution grid to accommodate the DERs. Areas with limited integration capacity and high DER potential may preclude development of some of the DERs projected in the forecast, or

6 SCE has realized more than 60,000 plug-in electric vehicles in its service territory since 2010. 7 Navigant Q2 2015, EPRI July 2011, Gartner Research Jan 2012, Bloomberg May 2012, CEC 2012 IEPR 8 SCE develops assumptions for electric vehicle miles traveled per day (eVMT), vehicle-type mix (e.g., battery electric, plug-in hybrid 15, plug-in hybrid 40), vehicle and charger efficiencies, customer TOU adoption, and customer charging behavior.

6



7

alternatively may identify areas where additional distribution investment is needed to accommodate DER growth. System-wide DER Forecasts

As a starting point, the system wide distributed solar and combined heat and power forecasts were obtained from SCE’s internal 2016 winter forecasts. For distributed solar and combined heat and power, the system wide forecasts were baselined to the total existing interconnections for each respective DER. The existing interconnections includes actual approved and pending generation applications.

DER Circuit Allocation Distributed Solar

The overall projections of solar photovoltaic (PV) systems were allocated by first splitting the projections between residential and commercial installations using an internal forecast. SCE forecasts the residential and commercial market segments separately, using similar techniques. The models predict customer solar PV adoption using a set of input variables, such as historical adoption and economic potential. For residential customers, economic potential was developed using a study of individual customer potential savings performed for SCE by Caltech9. Based on their study, savings (economic) potential was the main predictor of Solar PV adoption.

For commercial customers, SCE used a similar approach, relying on historical adoption and economic potential. The commercial customers were grouped by historical usage and North American Industry Classification System (NAICS) code10. The process involved two parts: first identifying optimal PV sizing for each customer and then applying this size to the savings equation to determine the economic potential. Once the system-wide forecast was determined, the system-wide forecast was allocated down to the circuit level based on the underlying customer type dispersion across the distribution circuits. The solar PV load shape was based on an internal information (e.g., module type, inverter type, number of modules per string, angle of system) inputted into the software package PVSyst11. The resulting hourly solar PV output represents a typical pattern of energy production. Combined Heat and Power (CHP)

The CHP forecast allocation was done in a two-step process. The first step involved obtaining Standard Industrial Classification (SIC) codes associating certain customer types with CHP potential and total CHP potential in MW by customer type in SCE’s territory from the study, ‘Combined Heat and Power: Policy Analysis and Market Assessment 2011-2030’1213. When obtaining the total CHP potential by customer type from the study, it was assumed that distributed connected CHP would be less than 5 MW in size. A normalized ratio was obtained that would allocate a system wide CHP forecast to specific customer types as identified by SIC code.

The second step determined where the customer types reside on SCE’s system. Internal

customer data was used to determine the customer type identified by SIC code, circuit location, and peak load. These three pieces of information allowed for the circuit allocation by customer type of CHP potential. It was assumed that peak load would correspond to the size of CHP installation. Combining the circuit level allocation with the allocation to specific customer types 9 “A Model for Residential Adoption of Photovoltaic Systems,” http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=7286226 10 The North American Industry Classification System (NAICS) is the standard used by Federal statistical agencies to classify businesses on their primary economic activity. As an example, code 44511 represents supermarkets and other grocery stores. 11 PVSyst models complete photovoltaic systems. More information on the capabilities of PVSyst can be found at: www.pvsyst.com 12 Standard Industrial Classification (SIC) code was the standard used by Federal statistical agencies to classify businesses on their primary economic activity. SIC codes are superseded by NAIC codes. 13 http://www.energy.ca.gov/2012publications/CEC-200-2012-002/CEC-200-2012-002-REV.pdf

7



8

allowed for the development of a final allocation that distributes a system wide CHP forecast to each circuit based on the CHP potential of current customers.

The final step to obtain circuit level DER forecasts was to utilize these circuit level DER allocations in combination with the system-wide DER forecasts. Since the DER allocation provides a ratio of applicable DERs by circuit, the system-wide forecast is multiplied by the ratio for each circuit to obtain the circuit level DER forecasts. 4) Historic Forecast Performance

SCE examines model statistics as one aspect of assessing forecast

reasonableness. If the model statistics suggest a well specified model and estimated parameters conform to economic theory, we place some degree of confidence that the model will produce a reasonable forecast. For example, we generally accept a statistical relationship between electricity use and a variable thought to influence it only if the estimated parameter is at least twice the magnitude of its standard error. Also, we compare elasticities derived from the model and compare these to elasticities published in various studies or reported by other utilities.

We also perform in-sample simulations. That is, we test the models forecast performance over a period of time where simulated electricity use can be compared to actual electricity use.

Our forecasts are regularly and constantly evaluated with respect to accuracy. The basic evaluation is straightforward: the forecast prediction for a particular time period is compared to actual data, adjusted for weather variation as that data becomes available.

The basic metrics used in the evaluation are the Root Mean Squared Error

(RMSE) and the Mean Absolute Percent Error (MAPE). The definitions of RMSE and MAPE are as follows: Suppose the forecast sample is j = T + 1, T + 2, …,T + h Let SF,t represent predicted sales in period t and SN,t represent actual adjusted sales in period t; then:

RMSE = SQRT( ∑ t= T+ 1 (SF,t - SN,t)2 / h )

MAPE = 100 ● ∑ t= T+ 1 ABS((SF,t - SN,t)/ SN,t) / h The validation process with respect to the long term sales forecast is undertaken monthly as each successive month’s actual billed sales becomes available. As part of the validation process, the new month’s billed sales is converted into weather and billing day adjusted values in order to eliminate variation in weather and billing days from the evaluation calculations. An analysis of the September 2014 forecast compared to actual weather adjusted monthly sales for the period January 2015 to December 2015 reveals the following:

8



9

SCE Sales Forecast Evaluation for 2015

Actual (Weather Adj.)

(MWh)

Forecast September 2014 Vintage

(MWh) MAPE

Calculation Jan-15 7,022 6,697 0.0463 Feb-15 5,968 5,991 0.0037 Mar-15 6,870 6,555 0.0458 Apr-15 6,185 6,390 0.0331

May-15 6,510 6,835 0.0499 Jun-15 6,815 7,090 0.0403 Jul-15 7,801 8,229 0.0549

Aug-15 7,993 8,419 0.0533 Sep-15 8,233 7,909 0.0394 Oct-15 7,535 7,098 0.0581 Nov-15 6,957 6,519 0.0630 Dec-15 6,773 6,850 0.0114

Jan-Dec Total

(GWh) 84,664 84,582

Simple Error 0.05% MAPE Error 4.20%

The analysis shows that the 2014 SCE billed monthly retail sales tracked actual weather-adjusted retail sales closely, with the exception of the months of October and November. The yearly MAPE was 4.2 percent due largely to relatively high errors in October and November. October 2015 CDDs were 118 percent above normal and November CDDs were 225 percent above normal. 5) Weather Adjustment Procedures

SCE has developed the weather and billing cycle adjustment model for the

purpose of comparing recorded and weather adjusted sales on a monthly basis. Weather and the calendar have the most significant impact on the monthly and annual variations in electricity sales. The Weather Modeling System (WMS) is a SAS based program that calculates heating- and cooling-degree days (HDD/CDD) that correspond to the monthly billing cycle schedule rather than a calendar month.

The annual billing cycle consists of 12 schedules of 21 meter reading days

distributed across the year. A monthly billing cycle consists of 21 meter read days. The 12 monthly billing cycles while approximating a calendar month are not required to be coincident with the calendar month. In addition the number of days for between each meter read varies depending on the days in the month and the number of weekend days and holidays. The MWS, using daily temperatures and the number of days between each meter read, calculates the number of HDD/CDD for the 252 (12 x 21) meter read days in a year.

9



10

The electricity sales for each monthly billing cycle are disaggregated into each meter read. The electricity sales for the meter reads are statistically adjusted as a function of the difference between actual HDD/CDD for recorded number of days in the meter read. The adjusted electricity sales are then aggregated back into a monthly billing cycle.

The HDD/CDD is also adjusted for the changing distribution of customers within

the service area. The WMS calculates customer-weighted average HDD/CDD using daily temperatures for the ten weather stations listed above. A further refinement is that the HDD/CDD are also adjusted according to the changing saturation of space conditioning appliances. Finally, separate sets of HDD/CDD are calculated for residential and non-residential electricity sales. A corresponding set of normal HDD/CCD, based on thirty years of history (1978 to 2007) are also calculated in the same manner.

The weather and billing day adjustment process is as follows:

Let YA,t = actual billed sales per customer and YN,t = adjusted sales per customer

Then YAt = β0 + β1● CDDA,t + β2 ● BDaysA,t and

YNt = β0 + β1● CDDN,t + β2 ● BDaysN,t

Where CDDA,t is actual measured cooling degree days in the current time period, BDaysA,t is actual measured billing days in the current time period, CDDN,t is normal cooling degree days and BDaysN,t is normal billing days; β1 and β2 are coefficients that measure the relationship between a change in CDD and BDays respectively and a change in sales per customer.

The weather adjustment is:

Wt = (YA,t – YN,t) ● Custt and Weather Adjusted sales are: SN,t = SA,t – Wt

10



11

6) Forecast Uncertainty Suppose the "true" regression model is given by: Yt = xt′ ß + et where et is an independent, and identically distributed, mean zero random disturbance, and ß is a vector of unknown parameters. The true model generating Y is not known, but we obtain estimates b of the unknown parameters. Then, setting the error term equal to its mean value of zero, the (point) forecasts of Y are obtained as: yt = xt′ b Forecasts are made with error, where the error is simply the difference between the actual and forecasted value: et = yt − xt′ b Assuming that the model is correctly specified, there are two sources of forecast error: residual uncertainty and coefficient uncertainty. Residual Uncertainty The first source of error, termed residual or innovation uncertainty, arises because the innovations e in the equation are unknown for the forecast period and are replaced with their expectations. While the residuals are zero in expected value, the individual values are non-zero; the larger the variation in the individual errors, the greater the overall error in the forecasts. The standard measure of this variation is the standard error of the regression. Residual uncertainty is usually the largest source of forecast error. Coefficient Uncertainty The second source of forecast error is coefficient uncertainty. The estimated coefficients b of the equation deviate from the true coefficients ß in a random fashion. The standard error of the estimated coefficient, given in the regression output, is a measure of the precision with which the estimated coefficients measure the true coefficients. The effect of coefficient uncertainty depends upon the exogenous variables. Since the estimated coefficients are multiplied by the exogenous variables in the computation of forecasts, the more the exogenous variables deviate from their mean values, the greater is the forecast uncertainty. Forecast Variability The variability of forecasts is measured by the forecast standard errors. For a single equation without lagged dependent variables or ARMA terms, the forecast standard errors are computed as: se = s √ 1+ xt′ (X′X)-1 xt

11



12

where s is the standard error of regression. These standard errors account for both innovation uncertainty (the first term) and coefficient uncertainty (the second term). Point forecasts made from linear regression models estimated by least squares are optimal in the sense that they have the smallest forecast variance among forecasts made by linear unbiased estimators. Moreover, if the innovations are normally distributed, the forecast errors have a t-distribution and forecast intervals can be readily formed. A two standard error band provides an approximate 95% forecast interval. In other words, if you (hypothetically) make many forecasts, the actual value of the dependent variable will fall inside these bounds 95 percent of the time. SCE constructs 95% confidence bands around its base case forecast based on the uncertainties described above.

12



13

7) Flow Diagram for Electric Use and Customer Modeling and Forecasting Electricity Consumption (kWh per Customer) Modeling and Forecasting

Economic and natural resource indicators (e.g.,

employment, precipitation)

Floor space by business sector

Weather – CDD and HDD

Billing Days

Electric rates by customer class

Primary inputs –explanatory variables

Pre-model adjustment to dependent variable Variable to be explained Forecast Techniques

Add Self Generation byCustomer Class

Monthly ElectricConsumption byCustomer Class

Econometric Modelsof Monthly Electricity

Consumption byCustomer class

Output

Forecast MonthlyElectricity Consumption

By Customer class

13



14

Electric Customer Modeling and Forecasting

Floor space by businesssector

Floor space by business sector

Residential buildingstarts

Residential building starts

Customers by customerclass

Customers by customer class

Econometric models ofresidential and non-

residential customers bycustomer class using

forecast of explanatoryvariables

Econometric models of residential and non-

residential customers by customer class using

forecast of explanatory variables

Forecast electricitycustomers by customer

class

Forecast electricity customers by customer

class

Primaryr inputstt –expx lanatoryr variables

Primary inputs –explanatory variables VaVV riable to be expx lainedVariable to be explained Forecast TeTT chniquesForecast Techniques Outpt utOutput

Economic indicators(e.g., employment)

Economic indicators (e.g., employment)

esidential customerforecast for non-

residential customerclasses

Residential customer forecast for non-

residential customer classes

Electric Retail Sales Modeling and Forecasting

Primaryr inputstt –expx lanatoryr variables

Primary inputs –explanatory variables VaVV riable to be expx lainedVariable to be explained Final Outpt utFinal Output

Adjusted futuremonthly electricity

consumption bycustomer class

Adjusted future monthly electricity

consumption by customer class

Future monthlyelectricity

customers bycustomer class andsectoral floor space

Future monthly electricity

customers by customer class and sectoral floor space

Monthly retail electricitysales by customer class

(pre-adjustment)

Monthly retail electricity sales by customer class

(pre-adjustment)X =

Post-model adjd ustmentPost-model adjustment

Adjusted forecastmonthly electricity

consumption bycustomer class

Adjusted forecast monthly electricity

consumption by customer class

Subtract DG andadd EV load,

EE exceedenceadjustment

Subtract DG and add EV load,

EE exceedence adjustment

14



15

8) Model Statistics – Electricity Use Models

The statistical details of the electricity consumption models are shown below. A glossary of variable names follows in Section 8.

Residential Electricity Use Model – L.A. County

Dependent Variable: LAUSE

Number of Observations Read 132 Number of Observations Used 132

Analysis of Variance

Source DF Sum of Mean F

Value Pr > F Squares Square Model 19 0.84549 0.0445 157.86 |t| Estimate Error Intercept 1 -0.24395 0.42628 -0.57 0.5683 LA_CDDSUMSEASLASIZE 1 6.49E-07 4.27E-08 15.2



16

Residential Electricity Use Model – Orange County

Dependent Variable: ORUSE Number of Observations Read 132 Number of Observations Used 132



Value Pr > F Squares Square Model 19 0.77397 0.04074 110.78 |t| Estimate Error Intercept 1 -0.29764 0.26358 -1.13 0.2612 OR_CDDSUMSEASORSIZE 1 6.21E-07 4.20E-08 14.78



17

Residential Electricity Use Model – Riverside County

Dependent Variable: RVUSE Number of Observations Read 144 Number of Observations Used 144



Value Pr > F Squares Square Model 19 6.64654 0.34982 299.55 |t| Estimate Error Intercept 1 -0.54895 0.24726 -2.22 0.0282 RV_CDDSUMSEASRVSIZE 1 7.99E-07 5.63E-08 14.19



18

Residential Electricity Use Model – San Bernardino County

Dependent Variable: SBUSE Number of Observations Read 132 Number of Observations Used 132



Value Pr > F Squares Square Model 19 3.28446 0.17287 288.57 |t| Estimate Error Intercept 1 -0.42233 0.29792 -1.42 0.1591 SB_CDDSUMSEASSBSIZE 1 7.26E-07 4.59E-08 15.84



19

Residential Electricity Use Model – Ventura/Santa Barbara Counties

Dependent Variable: VSBUSE Number of Observations Read 132 Number of Observations Used 132



Value Pr > F Squares Square Model 19 0.42407 0.02232 83.62 |t| Estimate Error Intercept 1 0.01375 0.29033 0.05 0.9623 VSB_CDDSUMSEASVSBSIZE 1 4.46E-07 4.19E-08 10.63



20

Residential Electricity Use Model – Other (Rural) Counties

Dependent Variable: OTHUSE Number of Observations Read 144 Number of Observations Used 144



Value Pr > F Squares Square Model 17 3.85946 0.22703 303.46 |t| Estimate Error Intercept 1 -0.01318 0.17632 -0.07 0.9405 OTH_CDDSUMSEASOTHSIZE 1 5.73E-07 4.92E-08 11.64



21

Commercial Electricity Use Model – Large Customers

Dependent Variable: COMLUSE Number of Observations Read 168 Number of Observations Used 168



Value Pr > F Squares Square Model 17 18335 1078.51258 46.47 |t| Estimate Error Intercept 1 -423.48094 60.60036 -6.99



22

Commercial Electricity Use Model – Small Customers

Dependent Variable: COMSUSE Number of Observations Read 96 Number of Observations Used 96



Value Pr > F Squares Square Model 17 14.98792 0.88164 62.48 |t| Estimate Error Intercept 1 -7.17495 7.25283 -0.99 0.3256 COMCDDCOMSIZESUMSEAS 1 5.76E-07 1.34E-07 4.3



Industrial Electricity Use Model

Dependent Variable: INDUSE Number of Observations Read

168

Number of Observations Used

168



Value Pr > F Squares Square Model 16 9.27582 0.57974 75.62 |t| Estimate Error Intercept 1 0.16989 0.28351 0.6 0.5499 COMCDDSUMSEAS 1 -0.00002023 0.00026915 -0.08 0.9402 CUMBDAYS 1 0.00215 0.00029207 7.38



Other Public Authority Electricity Use Model

Dependent Variable: OPAUSE Number of Observations Read

96

Number of Observations Used

96



Value Pr > F Squares Square Model 12 2.87855 0.23988 46.72 |t| Estimate Error Intercept 1 -10.26236 3.77488 -2.72 0.008 COMCDD 1 0.00172 0.00020595 8.35



Agriculture Electricity Use Model

Dependent Variable: AGRUSE Number of Observations Read 168 Number of Observations Used 168



Value Pr > F Squares Square Model 16 827.73026 51.73314 135.15 |t| Estimate Error Intercept 1 -48.60287 5.54065 -8.77



Street Light Electricity Use Model

Dependent Variable: STRLUSE Number of Observations Read 96 Number of Observations Used 96



Value Pr > F Squares Square Model 4 0.95351 0.23838 97.99 |t| Estimate Error Intercept 1 -0.4113 0.82917 -0.5 0.6211 CUMBDAYS 1 0.00105 0.00014314 7.36



9) Electricity Use Model Variable Description

Residential Electricity Use Model ResUse Recorded residential class monthly electricity consumption in kWh per

customer. Source: SCE. CDD Cooling degree-days. Sources: SCE and National Weather Service HDD Heating degree-days. Sources: SCE and National Weather Service ResRate Residential constant $2009 dollar price of electricity in cents per kWh.

Source: SCE and IHS Global Insight CUMBDAYS Average number of days in monthly billing statement multiplied by the

number of billing cycles in month. Source: SCE GeoGDP Regional output in 2009 dollars. Compiled from Moody’s Analytics data. JAN-DEC Binary variable set equal to 1 for the designated month and zero

otherwise. GeoSIZE Average residential household size in square feet. Compiled from Dodge

Data & Analytics data. SUMSEAS A binary equal to 1 during the summer months April to October and

zero otherwise. WINSEAS A binary equal to 1 during the winter months November to March and

zero otherwise. DUMMY_YYYYMMMM Binary variables equal to one on a particular month and year, and zero

otherwise, that are designed to capture billing irregularities in customer data.

LA Prefix in front of variable name to denote Los Angeles County. OR Prefix in front of variable name to denote Orange County. SB Prefix in front of variable name to denote San Bernardino County. RIV Prefix in front of variable name to denote Riverside County. VEN Prefix in front of variable name to denote Ventura and Santa Barbara

Counties. OTH Prefix in front of variable name to denote Rural Counties (Fresno, Inyo,

Kern Kings, Mono and Tulare)

27



Commercial Electricity Use Model COMUSE Recorded commercial class monthly electricity consumption in MWh per

commercial customer. Source: SCE COMCDD Non-residential cooling degree-days, dynamic population share weighted.

Sources: SCE and National Weather Service COMRATE Commercial class constant $2009 dollar price of electricity in cents per

kWh. Sources: IHS Global Insight and SCE SCEGDP SCE regional output in 2009 dollars. Compiled from Moody’s Analytics

data. SCENFEMP SCE non-farm employment. Compiled from Moody’s Analytics data. COMSIZE Average commercial building size in square feet. Sources: Dodge Data &

Analytics and SCE CUMBDAYS Average number of days in monthly billing statement multiplied by the

number of billing cycles in month. Source: SCE JAN-NOV Binary variable set equal to 1 for the designated month and zero

otherwise SUMSEAS A binary equal to 1 during the summer months May to October and zero

otherwise NONRES_CAC An index measuring the average efficiency of commercial air conditioning

equipment. Compiled from Energy Information Administration data. COMCUSDUMMY Binary variables equal to one on multiple periods, and zero otherwise,

that are designed to capture irregularities in customer data S A symbol after a variable name to denote small commercial class

customers or rate (generally those in the GS-1 and GS-2 rate groups) L A symbol after a variable name to denote large commercial class

customers or rate (generally those in the TOU rate groups)

28



Industrial Electricity Use Model INDUSE Recorded industrial class monthly electricity consumption in kWh per

industrial building square feet. Sources: SCE and Dodge Data & Analytics COMCDD Non-residential cooling degree-days static population weighting. Sources:

SCE and National Weather Service INDRATE Industrial class constant $2009 dollar price of electricity in cents per kWh.

Sources: SCE and IHS Global Insight SCEMFGEMP SCE regional manufacturing sector monthly employment. Compiled from

Moody’s Analytics data. CUMBDAYS Average number of days in monthly billing statement multiplied by the

number of billing cycles in a month. Source: SCE JAN-NOV Binary variable set equal to 1 for the designated month and zero

otherwise SUMSEAS A binary equal to 1 during the summer months May to October and zero

otherwise INDUSE_TREND Linear counter variable designed to capture secular trend in industrial

class electricity consumption not otherwise captured in the model

29



Other Public Authority Electricity Use Model OPAUSE Recorded other public authority class monthly electricity consumption in kWh per

government building square feet. Sources: SCE and Dodge Data & Analytics COMCDD Non-residential cooling degree-days, static population weighted. Sources: SCE

and National Weather Service OPARATE Other public authority class constant $2009 dollar price of electricity in cents per

kWh. Sources: SCE and IHS Global Insight SCEGOVEMP Government employment. Compiled from Moody’s Analytics data. CUMBDAYS Average number of days in monthly billing statement multiplied by the number of

billing cycles in month. Source: SCE DAYSHRS Number of hours of daylight in a month in Southern California (a proxy for office

lighting use). Source: SCE LIGHTINDX An index of commercial building lighting efficiency, Compiled from Energy

Information Administration data. MAR-AUG Binary variable set equal to 1 for the designated month and zero otherwise. NONRES_CAC An index measuring the average efficiency of commercial air conditioning

equipment. Compiled from Energy Information Administration data.

30



Agriculture Electricity Use Model AGRUSE Recorded agriculture class monthly electricity consumption in MWh per

agriculture customer. Source: SCE CUMBDAYS Average number of days in monthly billing statement multiplied by the

number of billing cycles in month. Source: SCE RUNOFF Full natural flow of San Joaquin River at Friant Dam in cubic feet of flow

per second. Sources: U.S Department of the Interior and SCE PRECIP Fresno monthly precipitation level in inches. Sources: National Oceanic

and Atmospheric Administration and SCE JAN-NOV Binary variable set equal to 1 for the designated month and zero

otherwise. AGRDUMMY1215 Binary variables equal to one from 2012 to 2015 and zero otherwise, that

are designed to drought-impacted usage data.

31



Street Light Electricity Use Model STLUSE Recorded street light class electricity monthly consumption in MWh per street light

customer. Source: SCE RESPRSTLT Number of residential customers per street lighting customer. Source: SCE. CUMBDAYS Average number of days in monthly billing statement multiplied by the number of

billing cycles in month. Source: SCE. DAYSHRS Number of hours of daylight in a month in Southern California (a proxy for office

lighting use). Source: SCE LIGHTINDEX An index of commercial building lighting efficiency. Compiled from Energy

Information Administration data.

32



10) Model Statistics – Customer Models

The statistical details of the residential and non-residential customer models are shown below, while a glossary of terms follows in Section 10. The residential customer models are constructed on the basis that new customers are determined mainly by housing starts (with a lag extending from 8 to 12 months depending upon the region). The housing start forecast is from Moody’s Analytics.

Note that in the case of the industrial and other public authority (OPA) customer classes,

the sales forecasts are constructed as the product of electricity consumption per square foot and total building square feet. Thus the forecasts of Industrial class customers and OPA customer are independent of industrial and OPA customer class sales. An independent forecast of building square feet by building type is provided by Dodge Data & Analytics.

33



Residential Electricity Customer Model – L.A. County

Dependent Variable: D_LACUS Number of Observations Read 168 Number of Observations Used 168



Value Pr > F Squares Square Model 15 13673435 911562 5.42 |t| Estimate Error Intercept 1 202.02232 133.84022 1.51 0.1333 LASTRT18 1 0.01317 0.00561 2.35 0.0202 DUMMY_REC0809 1 -469.36953 91.64406 -5.12



Residential Electricity Customer Model – Orange County

Dependent Variable: D_ORCUS Number of Observations Read 168 Number of Observations Used 168



Value Pr > F Squares Square Model 15 4404878 293659 4.16 |t| Estimate Error Intercept 1 214.92066 85.21616 2.52 0.0127 ORSTRT15 1 0.02884 0.00729 3.96 0.0001 ORCUSDUMMY 1 548.47528 125.49211 4.37



Residential Electricity Customer Model – Riverside County

Dependent Variable: D_RVCUS Number of Observations Read 144 Number of Observations Used 144



Value Pr > F Squares Square Model 13 60362570 4643275 50.78 |t| Estimate Error Intercept 1 166.28743 91.96777 1.81 0.0729 RVSTRT3 1 0.07235 0.00291 24.9



Residential Electricity Customer Model – San Bernardino County

Dependent Variable: D_SBCUS Number of Observations Read 168 Number of Observations Used 168



Value Pr > F Squares Square Model 14 26775033 1912502 27.03 |t| Estimate Error Intercept 1 52.36271 75.86253 0.69 0.4911 SBSTRT3 1 0.04238 0.0024 17.69



Residential Electricity Customer Model – Ventura/Santa Barbara Counties

Dependent Variable: D_VSBCUS Number of Observations Read 168 Number of Observations Used 168



Value Pr > F Squares Square Model 13 3871016 297770 20.54 |t| Estimate Error Intercept 1 21.62301 34.89783 0.62 0.5364 VSBSTRT12 1 0.05051 0.00571 8.84



Residential Electricity Customer Model – Other (Rural) Counties

Dependent Variable: D_OTHCUS Number of Observations Read 144 Number of Observations Used 144



Value Pr > F Squares Square Model 13 2911648 223973 18.33 |t| Estimate Error Intercept 1 -69.34312 36.01458 -1.93 0.0564 OTHSTRT3 1 0.11702 0.00784 14.92



Commercial Customer Model – Large Customers

Dependent Variable: D_COMLCUS Number of Observations Read 168 Number of Observations Used 168



Value Pr > F Squares Square Model 14 64750 4625.03529 2.82 0.0009 Error 153 250609 1637.96841 Corrected Total 167 315360 Root MSE 40.47182 R-Square 0.2053 Dependent Mean 9.91071 Adj R-Sq 0.1326 Coeff Var 408.36433

Parameter Estimates

Variable DF Parameter Standard

t Value Pr > |t| Estimate Error Intercept 1 -26.18342 13.21653 -1.98 0.0494 D_SCENFEMP12 1 0.41132 0.18934 2.17 0.0314 D_COMSQF12 1 0.00878 0.0024 3.65 0.0004 COMLDUMMY 1 -45.72548 15.34462 -2.98 0.0034 JAN 1 82.57368 41.95546 1.97 0.0509 FEB 1 -0.77625 15.9584 -0.05 0.9613 MAR 1 6.04692 15.65857 0.39 0.6999 APR 1 12.31322 15.32454 0.8 0.4229 MAY 1 23.45319 15.37566 1.53 0.1292 JUN 1 22.20206 15.53825 1.43 0.1551 JUL 1 47.31782 23.47818 2.02 0.0456 AUG 1 12.72159 15.347 0.83 0.4084 SEP 1 0.39731 16.20018 0.02 0.9805 OCT 1 -25.61364 17.31279 -1.48 0.1411 NOV 1 8.92587 15.851 0.56 0.5742

The D_ indicates the first difference. SCENFEMP12 indicates SCE non-farm employment lagged 12 periods. COMSQFT12 indicates SCE commercial floorspace lagged 12 periods. Last sample observation was December 2015. First forecast period was January 2016.

40



Commercial Customer Model – Small Customers

Dependent Variable: D_COMSCUS Number of Observations Read 168 Number of Observations Used 168



Value Pr > F Squares Square Model 15 29669017 1977934 28.26 |t| Estimate Error Intercept 1 -21.69462 79.41812 -0.27 0.7851 D_COMSQF 1 0.17654 0.0185 9.54



Industrial Customer Model

Dependent Variable: D_INDCUS Number of Observations Read 96 Number of Observations Used 96



Value Pr > F Squares Square Model 14 86033 6145.2047 5.31 |t| Estimate Error Intercept 1 -75.5054 15.896 -4.75



Other Public Authority Customer Model

Dependent Variable: D_OPACUS Number of Observations Read 96 Number of Observations Used 96



Value Pr > F Squares Square Model 13 4177.84215 321.37247 1.64 0.0899 Error 82 16035 195.54461 Corrected Total 95 20213 Root MSE 13.98373 R-Square 0.2067 Dependent Mean -30.125 Adj R-Sq 0.0809 Coeff Var -46.41901

Parameter Estimates


t Value Pr > |t| Estimate Error Intercept 1 -22.87208 6.08067 -3.76 0.0003 D_OPASQFSCEGOVEMP 1 -1.55456 1.67452 -0.93 0.3559 OPADUMMY 1 -14.72231 4.64941 -3.17 0.0022 JAN 1 -5.69715 9.053 -0.63 0.5309 FEB 1 -4.15163 8.10412 -0.51 0.6098 MAR 1 -9.09202 8.35171 -1.09 0.2795 APR 1 -10.39688 7.65658 -1.36 0.1782 MAY 1 -7.04044 8.47161 -0.83 0.4084 JUN 1 -2.4712 8.36603 -0.3 0.7684 JUL 1 -11.16933 7.41578 -1.51 0.1359 AUG 1 -7.18675 7.69119 -0.93 0.3528 SEP 1 -7.35793 7.19743 -1.02 0.3096 OCT 1 -2.46392 6.99197 -0.35 0.7254 NOV 1 7.88779 7.03681 1.12 0.2656

The D_ indicates the first difference. Last sample observation was December 2015. First forecast period was January 2016.

43



Agriculture Customer Model

Dependent Variable: D_AGRCUS Number of Observations Read 96 Number of Observations Used 96



Value Pr > F Squares Square Model 13 15585 1198.83496 1.95 0.0355 Error 82 50328 613.7547 Corrected Total 95 65913 Root MSE 24.77407 R-Square 0.2364 Dependent Mean -10.61458 Adj R-Sq 0.1154 Coeff Var -

233.39657

Parameter Estimates


t Value Pr > |t| Estimate Error Intercept 1 -21.96362 8.78688 -2.5 0.0144 D_AGREMP6 1 0.0007781 0.00082345 0.94 0.3475 AGRCUSDUMMY 1 -37.34788 18.77752 -1.99 0.05 JAN 1 29.8696 16.63691 1.8 0.0763 FEB 1 23.4402 12.83888 1.83 0.0715 MAR 1 20.05259 12.39608 1.62 0.1096 APR 1 17.82266 12.60708 1.41 0.1612 MAY 1 43.41416 12.9581 3.35 0.0012 JUN 1 19.30476 13.39006 1.44 0.1532 JUL 1 17.84222 12.46821 1.43 0.1562 AUG 1 5.47839 12.5486 0.44 0.6636 SEP 1 1.24243 13.42154 0.09 0.9265 OCT 1 -20.62789 19.11202 -1.08 0.2836 NOV 1 -11.46758 15.60458 -0.73 0.4645

The D_ indicates the first difference. AGREMP6 indicates SCE agricultural employment lagged 6 periods. Last sample observation was December 2015. First forecast period was January 2016.

44



Street Light Customer Model

Dependent Variable: D_STLCUS Number of Observations Read 96 Number of Observations Used 96



Value Pr > F Squares Square Model 13 207842 15988 19.32 |t| Estimate Error Intercept 1 34.91089 10.21368 3.42 0.001 D_SCERESCUS18 1 0.0103 0.00213 4.84



11) Customer Model Variable Description Residential Customer Models RESCUS Recorded number of residential class customers. Source: SCE LA Prefix in front of variable name to denote Los Angeles County OR Prefix in front of variable name to denote Orange County. SB Prefix in front of variable name to denote San Bernardino County. RV Prefix in front of variable name to denote Riverside County. VSB Prefix in front of variable name to denote Ventura and Santa Barbara

Counties. OTH Prefix in front of variable name to denote Rural Counties (Fresno, Inyo,

Kern Kings, Mono and Tulare) Jan-Nov Binary variable set equal to 1 for the designated month and zero

otherwise. DUMMY_YYYYMMMM Binary variables equal to one on a particular month and year, and zero


GEOCUSDUMMY Binary variables equal to one on multiple periods, and zero otherwise,

that are designed to capture irregularities in customer data. DUMMY_CRASH0708 Binary variables equal to one during the housing crash, and zero

otherwise, that are designed to capture recessionary period. DUMMY_REC0809 Binary variables equal to one during the Great Recession, and zero

otherwise, that are designed to capture recessionary period. Commercial Customer Models ComCus Recorded number of commercial class customers. Source: SCE COMSQF Commercial building total square footage. Compiled from Dodge Data &

Analytics data. DUMMY_YYYYMMMM Binary variables equal to one on a particular month and year, and zero


COMLCUSDUMMY Binary variables equal to one on multiple periods, and zero otherwise,

that are designed to capture irregularities in customer data. DUMMY_REC0809 Binary variables equal to one during the Great Recession, and zero

otherwise, that are designed to capture recessionary period.

46



L

Workpapers - Southern California Edison · 2016-09-01 · Workpaper – Southern California Edison...

Documents

Transcript of Workpapers - Southern California Edison · 2016-09-01 · Workpaper – Southern California Edison...