1 Econometric Load Forecasting Peak and Energy Forecast 06/14/2005 Econometric Load Forecasting Peak and Energy Forecast 06/14/2005

2 Previous ERCOT Forecast Methodology The previous load forecast methodology was based on a simplistic trending of historic ERCOT peak demand growth to develop the long-term forecast of summer peak demand, unadjusted for weather or economic conditions Several compound growth rates were calculated (10-year, 5-year, 2-year, etc). These growth rates were applied to the latest available peak to obtain forecasts, and the most reasonable forecast was selected, based on judgment

3 Forecasting Improvement Original Plan –Develop hourly load forecast for five years for smaller regions of ERCOT (either weather zones or CM areas) for use in UPlan studies –Sum of these forecasts across ERCOT would be used as the ERCOT System peak and energy forecast Revised Goal –Due to the immediate need for a more rigorous forecasting approach, the goal for the current forecast cycle was changed to develop an ERCOT system forecast directly

4 Econometric Forecasting Basics Regression Analysis Develop an equation or equations that describe the historic load as a function of certain independent variables –Variables must be logical, historically measurable and have an available forecast Statistical analysis techniques are used to calculate the appropriate coefficients on each variable and to choose the best equations –Equations are chosen that minimize sum of the squares of the differences between the actual, observed load levels and the load levels that are predicted by inserting the historic values of the independent variables into the equation Projected values for each of the variables in the equations are then inserted into the equations to produce the forecast

Forecasting Process ERCOT Total System Hourly Load Forecast Allocate Energy Allocate Energy Allocate Energy ERCOT Peak and Energy Forecast Economic Data Weather Data Calendar Data Forecasted Data Load Data Economic Data Weather Data Calendar Data Historic Data ERCOT Total System Summer Hourly Load Shape Model ERCOT Total System Winter Hourly Load Shape Model ERCOT Total System Spring/Fall Hourly Load Shape Model ERCOT Winter Monthly Energy Model ERCOT Spring/Fall Monthly Energy Model ERCOT Summer Monthly Energy Model Six Regression Equations

Forecasting Process 1.Obtain weather and economic variables (historic and forecast) 2.Develop regression equations describing the historic actual: –Monthly Energy Using a different equation for each season –Hourly Load Shape Using a different equation for each season 3.Incorporate forecasted values of economic and normalized temperatures for into Monthly Energy equation to produce forecasted monthly energy 4.Incorporate normalized temperatures for into Monthly Energy equation to produce forecasted load shape 5.Produce hourly demand forecast by fitting forecasted monthly energy under projected hourly load shape

7 Load Data Economic Data Weather Data Calendar Data Historic Data 2005 Forecasting Process Historic Data

8 Historic Data - Economic 1.Economic Data Economic data obtained from Economy.com Data includes economic and demographic data (such as income, employment, housing permits, GDP, population and migration patterns) for Texas at the state, county, Metropolitan Statistical Areas (MSAs), and national level

9 Historic Data - Weather 2. Weather Data Ten years of weather data obtained from Weather Bank for 20 weather stations The data is first weighted by individual weather stations using ERCOT’s standard weighing factor, and then for the total system using weights proportional to the load in each weather zone

10 Historic Data - Load 3. Load Data Settlement load data available on an hourly basis since July 31, 2001 Prior to 2001, we have Transmission and Distribution Service Providers (TDSP) hourly data

11 Load Data Economic Data Weather Data Calendar Data Historic Data ERCOT Total System Summer Hourly Load Shape Model ERCOT Total System Winter Hourly Load Shape Model ERCOT Total System Spring/Fall Hourly Load Shape Model ERCOT Winter Monthly Energy Model ERCOT Spring/Fall Monthly Energy Model ERCOT Summer Monthly Energy Model Six Regression Equations 2005 Forecasting Process Regression Equations

12 Develop monthly energy and hourly load shape equations for each season The general formulation of the energy equations is: Energy Month i = f {Cdd, Hdd, Income, Population, Monthly Indicators} The general formulation of the load shape equations is: Load hour i = f {Max Temps, Lagged Temps, Heat Index, Non- Linear Temp Components (square and cube), Temp Gains (diff between daily high and Low temps), Temp Build-up, Dew Point, Month*Temp Interactions, Cdd, Hdd, Hour of Day Indicators, Weekday/Weekend, Holidays, Population, income} Regression Equations

13 Variable Selection and Regression Estimation Details Multiple Regression Analysis was use to develop the forecasting equations Initial selection of variables came from a stepwise procedure to determine those that were the most statistically significant A subset of those variables was carefully chosen on the basis of empirical results and judgment –Variables had to be logical, historically measurable and have an available forecast Ordinary Least Squares techniques with some of the models including autoregressive error terms were used to calculate the appropriate coefficients on each variable and to choose the best equations

14 Load Shape Model Fit

15 Model Fit Detailed SAS output is included at the end of the presentation, showing: –Model Variables –Coefficients –Statistical analysis

16 Load Data Economic Data Weather Data Calendar Data Historic Data ERCOT Total System Summer Hourly Load Shape Model ERCOT Total System Winter Hourly Load Shape Model ERCOT Total System Spring/Fall Hourly Load Shape Model ERCOT Winter Monthly Energy Model ERCOT Spring/Fall Monthly Energy Model ERCOT Summer Monthly Energy Model Six Regression Equations Economic Data Weather Data Calendar Data Forecasted Data 2005 Forecasting Process Forecasted Data

17 Economic Forecast ERCOT obtains the economic forecasts used in the models from Economy.com Economy.com is a leading national provider of economic data and forecasts, with over 500 clients worldwide including AEP, LCRA and Entergy Forecasts and data received includes economic and demographic data (such as income, employment, housing permits, GDP, population and migration patterns) for Texas at state, county and MSA, and some national economic data

18 Economic Forecast Growth Rates

19 Economic Forecast Growth Rates 1.95% 1.18% 1.33% 2.52%

20 Forecasted Weather Data Weather Forecast Assumptions Calculation of the normalized temperature profile involves the following steps: 1)Compute an overall system temperature for every year by combining the weather zone temperatures and weighing them according to the load in each zone 2)Rank the hourly temperatures for each year from highest to lowest 3)Determine the median temperature from all years for every hour 4)Calculate the sum of the absolute values of the difference of the median and the hourly temperatures for all hourly temperatures in each year 5)Determine the year with the minimum summed value and select this year as the typical year profile 6)Use this year’s profile to resort the median temperatures

21 Allocate Energy Allocate Energy Allocate Energy Load Data Economic Data Weather Data Calendar Data Historic Data ERCOT Total System Summer Hourly Load Shape Model ERCOT Total System Winter Hourly Load Shape Model ERCOT Total System Spring/Fall Hourly Load Shape Model ERCOT Winter Monthly Energy Model ERCOT Spring/Fall Monthly Energy Model ERCOT Summer Monthly Energy Model Six Regression Equations ERCOT Total System Hourly Load Forecast ERCOT Peak and Energy Forecast Economic Data Weather Data Calendar Data Forecasted Data 2005 Forecasting Process Hourly Forecast

22 Hourly Forecast The forecasted hourly shape from the load shape equations is scaled to produce the final hourly forecast –Each hour’s load is scaled so that the amount of energy under the load shape for a month is equal to the amount of energy projected for that month by the energy forecast from the energy equations –The percent of a month’s energy that is contained in an each hour from the load shape equation is maintained The peak forecast is the highest hourly load from this final hourly forecast

23 ERCOT Peak Forecast 1.83% Avg. Growth

24 ERCOT Energy Forecast 2.10% Avg. Growth

25 Forecast Growth Rates - Annual

26 Forecast Growth Rates - Annual 1.44% 1.81% 1.34% 2.10% AVG=56.16%

27 Questions?

28 Appendix Model Parameters and Statistics SAS Output

29 Summer Season Load Shape Equation

30 Summer Season Energy Equation

31 Winter Season Load Shape Equation

32 Winter Season Load Shape Equation

33 Winter Season Load Shape Equation

34 Winter Season Energy Equation

35 Spring/Fall Season Load Shape Equation

36 Spring/Fall Season Load Shape Equation

37 Spring/Fall Season Load Shape Equation

38 Spring/Fall Season Energy Equation