Colorado PUC E-Filings System

Transcription

1 BEFORE THE PUBLIC UTILITIES COMMISSION OF THE STATE OF COLORADO PROCEEDING NO. A- E IN THE MATTER OF THE APPLICATION OF BLACK HILLS/COLORADO ELECTRIC UTILITY COMPANY, LP FOR () APPROVAL OF ITS 0 ELECTRIC RESOURCE PLAN, AND () APPROVAL OF ITS 0-0 RES COMPLIANCE PLAN. DIRECT TESTIMONY OF Colorado PUC E-Filings System DANIEL G. HANSEN, Ph.D. ON BEHALF OF BLACK HILLS/COLORADO ELECTRIC UTILITY COMPANY, LP June, 0

2 SECTION Table of Contents PAGE I. INTRODUCTION AND QUALIFICATIONS... II. PURPOSE OF TESTIMONY... III. FORECAST OVERVIEW... IV. RESIDENTIAL FORECASTS... V. COMMERCIAL FORECASTS... VI. LARGE POWER SERVICE FORECAST... VII. SYSTEM PEAK DEMAND FORECASTS... VIII. CONCLUDING REMARKS...

3 DIRECT TESTIMONY OF DANIEL G. HANSEN, Ph.D. I. INTRODUCTION AND QUALIFICATIONS Q. PLEASE STATE YOUR NAME, POSITION, AND BUSINESS ADDRESS. A. My name is Daniel G. Hansen. I am a Vice President at Christensen Associates Energy Consulting, LLC located at Suite 00, 00 University Bay Drive, Madison, Wisconsin 0. 0 Q. BRIEFLY DESCRIBE YOUR EDUCATIONAL BACKGROUND AND BUSINESS BACKGROUND. A. My employment history and expertise is provided in Appendix A. Q. HAVE YOU PREVIOUSLY TESTIFIED IN UTILITY REGULATION PROCEEDINGS? A. Yes. I have testified in Arizona, Connecticut, Minnesota, Nevada, New Mexico, Oregon, and Utah. My testimony has covered issues including utility sales forecasts, revenue decoupling mechanisms, and rate design. In these proceedings, I represented a broad range of clients, including a regulator, an environmental organization, a non-profit organization of utility investors, and investor-owned utilities. 0 Q. ON WHOSE BEHALF ARE YOU TESTIFYING? A. I am testifying on behalf of Black Hills/Colorado Electric Utility Company, LP, d/b/a Black Hills Energy (Black Hills or the Company).

4 0 II. PURPOSE OF TESTIMONY Q. WHAT IS THE PURPOSE OF YOUR TESTIMONY? A. I will describe the forecasts I helped prepare for Black Hills, including the following six models:. Use per customer for residential customers;. The number of residential customers served;. Use per customer for commercial (Small General Service (SGS) and Large General Service (LGS)) customers;. The number of commercial (SGS + LGS) customers served;. Sales to the Large Power Service customers; and. System monthly peak demands. These forecasts are part of Black Hills Electric Resource Plan (ERP) filing. I will hereafter refer to forecasts of the number of customers served as customer models or customer forecasts. 0 Q. DOES THE ERP CONTAIN DETAILED DESCRIPTIONS OF THE STATISTICAL MODELS AND RESULTING FORECASTS? A. Yes. Appendix C of the ERP contains the statistical models and detailed outcomes of the forecasting process. In my testimony, I provide an explanation of the methods contained in the ERP and a high-level summary of the resulting forecasts. Q. WHAT IS THE OUTLINE OF YOUR TESTIMONY?

5 A. In Section III, I provide a high-level overview of the forecasting process. In Section IV, I present the residential forecasts. In Section V, I present the commercial forecasts. In Section VI, I present the Large Power Service forecast. In Section VII, I present the system monthly peak demand forecasts. In Section VIII, I provide concluding remarks. 0 III. FORECAST OVERVIEW Q. WHAT WAS YOUR ROLE IN THE PREPARATION OF THE SALES AND PEAK DEMAND FORECASTS? A. I worked with Black Hills staff to develop and review the statistical models that produce the forecasts. Black Hills collected the required data, estimated the models, and developed the resulting forecasts under my guidance. Q. PLEASE DESCRIBE THE BASIC STEPS INVOLVED IN PRODUCING FORECASTS OF SALES, THE NUMBER OF CUSTOMERS SERVED, AND SYSTEM PEAK DEMANDS. A. In each case, the forecasting process consists of three steps: data collection; model estimation; and forecast development. 0 Q. PLEASE DESCRIBE THE FIRST STEP IN DEVELOPING THE FORECAST: DATA COLLECTION. A. In the data collection step, the required information is collected for each customer class (or for the system as a whole). This includes sales, the number of customers

6 served, weather conditions, economic and demographic variables, and electricity price data. The utility s databases are the source of the sales and customer counts. Historical electricity price data (expressed as annual average revenue) were taken from Black Hills FERC Form filings. Weather data were obtained for the NOAA National Climatic Data Center s (NCDC) Pueblo Airport weather station. Economic and demographic data were obtained from Woods & Poole Economics, Inc. (W&P) for Pueblo and Fremont Counties for the years through 00. The W&P data provide a consistent measure of the included variables for the historical and forecast time periods. 0 0 Q. PLEASE DESCRIBE THE SECOND STEP IN DEVELOPING THE FORECAST: MODEL ESTIMATION. A. In the second step, statistical models are developed to estimate the relationship between sales (or peak demands or the number of customers served) and the relevant drivers. For example, the monthly sales to a particular customer group are modeled as a function of weather, economic and demographic conditions, and seasonal patterns. The historical sales data are matched to the appropriate historical data for the relevant drivers and used to estimate a statistical model that results in estimated coefficients. These coefficients reflect how sales change due to changes in each driver. For example, the estimated coefficient on a weather variable indicates the extent to which sales increase as the temperature gets hotter. Q. PLEASE DESCRIBE THE THIRD STEP: FORECAST DEVELOPMENT.

7 0 A. In the third and final step, the estimated coefficients from the second step are combined with forecasts of each driver to produce the forecast sales (or peak demand, or the number of customers served). Forecast weather conditions are set to reflect normal weather based on average temperatures across the previous 0 years. W&P provides forecasts for all of the economic and demographic variables. The retail electricity price is forecast according to the projected average retail price of electricity contained in the Energy Information Administration s (EIA) Annual Energy Outlook (AEO). Specifically, Black Hills retail price is assumed to increase by the same percentage as the forecast retail electricity price in the AEO. Q. WHAT TIME PERIODS ARE INCLUDED IN BLACK HILLS RETAIL ELECTRICITY SALES, CUSTOMER, AND PEAK DEMAND FORECASTS? A. The statistical models are estimated using monthly data from January 00 through December 0. Monthly forecasts are developed for 0 through Q. HOW DO YOU CREATE THE SALES FORECAST FOR EACH CUSTOMER CLASS? A. For the Large Power Service customers, the class-level sales are forecast directly. For the residential and commercial classes, the class-level sales forecasts were created by combining the forecasts of use per customer (UPC) and the customer

8 forecast. Specifically, for each forecast month, the sales forecast equals the product of the UPC forecast and the customer forecast. 0 Q. WHY DO YOU SEPARATE THE SALES FORECAST INTO THE UPC AND CUSTOMER COMPONENTS FOR THE RESIDENTIAL AND COMMERCIAL CLASSES? A. By dividing the sales forecast into the UPC and customer components, I am able to better distinguish between the effect of drivers on customer-level usage versus the number of customers served. For example, I would expect variations in weather conditions to explain some of the variation in average customer usage levels (e.g., the average customer uses more when the weather is hotter, all else equal), but I don t expect the weather variations to be a significant driver of the number of customers served. By separating the sales forecast into UPC and customer models, we are better able to isolate the effect of weather on UPC. A similar argument applies to other explanatory variables. Customers may use more electricity as economic conditions improve and/or more customers may be attracted to the service territory by improved economic conditions. These potential effects can be separately estimated using these methods. 0 IV. RESIDENTIAL FORECASTS Q. PLEASE DESCRIBE BLACK HILLS RESIDENTIAL FORECAST. A. Two statistical models are estimated for Black Hills residential customers: a UPC model and a customer model.

9 0 Q. PLEASE DESCRIBE THE RESIDENTIAL UPC MODEL. A. The residential UPC model includes the following variables (or drivers):. Cooling degree days with a 0-degree threshold (CDD0);. Heating degree days with a 0-degree threshold (HDD0);. The natural log of the -month moving average of real household personal income; and. Monthly indicator variables. The dependent variable in the model is the natural log of use per customer (sales divided by the number of customers served) in each month. 0 Q. WHY DO YOU TAKE THE NATURAL LOG OF THE UPC AND HOUSEHOLD INCOME VARIABLES? A. The use of logs allows the model to estimate an income elasticity, which is the percentage change in UPC in response to a percentage change in household income. This has two benefits. First, it allows for a more straightforward interpretation of the estimated coefficient because an elasticity can be compared across studies (e.g., values for different utilities and/or time periods), which facilitates model validation. Second, it represents a more appropriate relationship to apply over a long period of time, during which income may exhibit significant changes. I also examined models that included the natural log of the retail electricity price, with average class revenue per kwh from FERC Form data serving as the proxy for price. The models did not find a statistically significant relationship between price and UPC.

10 0 Q. HOW IS THE CDD0 VARIABLE CALCULATED? A. CDD0 is first calculated for each day as follows: CDD0 = MAX{0, (MaxT + MinT) / 0} That is, the average daily temperature is calculated as the average of the maximum and minimum daily temperatures. The threshold value (0 F) is subtracted from the average temperature and CDD0 is set to be the maximum of the result and zero. The daily CDD0 values are then added up across the days of the billing month (approximated as the th of the previous month to the th of the current month). Q. HOW IS THE CDD0 VARIABLE INTERPRETED? A. CDD0 is intended to reflect the demand for cooling (i.e., air conditioner use). It is developed under the assumption that there is no cooling load below the threshold value (set to the daily average temperature of 0 F here) and that cooling load increases with temperature above the threshold level. Q. HOW IS THE HDD0 VARIABLE CALCULATED? I found that using a 0-degree Fahrenheit threshold for both the CDD and HDD variables produced a slightly better model fit (higher R-squared) than a -degree threshold. However, the choice of a 0- or - degree threshold has very little effect on the forecast. For example, the forecast annual growth in residential sales from 0-00 changes from 0. percent to 0.0 percent as the threshold is changed from 0 to degrees.

11 A. HDD0 is calculated in the same manner as CDD0, except that the daily average temperature is subtracted from the threshold value (rather than the other way around for CDD0), as follows: HDD0 = MAX{0, 0 (MaxT + MinT) / } Q. HOW IS THE HDD0 VARIABLE INTERPRETED? A. The HDD0 variable is similar to the CDD0 variable, except that it reflects the demand for electric heating. 0 Q. WHAT IS THE PURPOSE OF THE HOUSEHOLD INCOME VARIABLE? A. This variable reflects how customers change their electricity usage in response to changes in economic conditions that directly affect their livelihood. Q. WHAT IS THE PURPOSE OF THE MONTHLY INDICATOR VARIABLES? A. These variables reflect seasonal patterns in electricity usage that are not captured by the other included variables. For example, lighting demand may vary seasonally due to changes in the number of daylight hours, which would not be well reflected by other included variables, such as CDDs and HDDs. 0 Q. ARE THE ESTIMATED COEFFICIENTS FROM THE RESIDENTIAL UPC MODEL REASONABLE? A. Yes. The estimates can be summarized as follows:

12 0 Residential UPC is positively related to both CDDs and HDDs (i.e., UPC increases when the weather is hotter); UPC is more sensitive to CDDs than HDDs, which reflects a larger effect of temperatures on cooling-related load than heating-related load; The estimated income elasticity is positive, indicating a positive relationship between UPC and income (i.e., people use more electricity when the economy improves); Seasonal patterns are relevant explanatory factors (i.e., the coefficients on the indicator variables are statistically significant); and The overall explanatory model of the power is very good, with an R-squared value of 0.. (R-squared ranges from 0 to and reflects the share of variation in the dependent variable that is explained by the included variables.) 0 Q. PLEASE DESCRIBE THE RESIDENTIAL CUSTOMER MODEL. A. The dependent variable in the residential customer model is the natural log of the number of residential customers served. The model includes only one explanatory variable: the number of households in the W&P database. Specifically, this information is included as the natural log of the -month moving average of the number of households. This variable is intended to reflect a combination of economic and demographic factors that affect the number of residential customers Black Hills is expected to serve. Q. DOES THE MODEL PRODUCE REASONABLE ESTIMATES? 0

13 A. Yes, the coefficient on the households variable is positive and statistically significant, indicating that the number of residential customers increases with the number of households (and vice-versa). The R-square value of.0 indicates that this variable (and the included constant term) explains virtually all of the variation in the number of customers served. 0 Q. PLEASE SUMMARIZE THE FORECAST THAT RESULTS FROM THE RESIDENTIAL UPC AND CUSTOMER MODELS. A. The forecast results in modest growth during the forecast period. From 0 to 00, residential UPC is forecast to grow by 0. percent per year; the number of residential customers served is forecast to grow by 0. percent per year; and total residential sales is forecast to grow by 0. percent per year. The sales growth rate declines during the forecast period, from 0. percent in the early years to 0. percent in later years. 0 V. COMMERCIAL FORECASTS Q. PLEASE DESCRIBE BLACK HILLS COMMERCIAL FORECAST. A. The SGS and LGS customer groups were combined in order to develop Black Hills commercial forecast. As with the residential forecast, two statistical models are estimated: a UPC model and a customer model. Q. PLEASE DESCRIBE THE COMMERCIAL UPC MODEL. A. The residential UPC model includes the following variables (or drivers):

14 . CDD0;. HDD0;. The natural log of the -month moving average of gross regional product (GRP);. The natural log of the average commercial revenue per kwh sold ( price ); and. Monthly indicator variables. The dependent variable in the model is the natural log of use per customer (sales divided by the number of customers served) in each month. 0 Q. HOW DO YOU INTERPRET THE COEFFICIENTS ON THE GRP AND PRICE VARIABLES? A. As with the income variable in the residential UPC model, the use of natural logs for the GRP and price variables results in estimates of elasticities. The coefficient on the ln(grp) variable is interpreted as the percentage change in UPC associated with a given percentage change in GRP. Similarly, the coefficient on the ln(price) variable is interpreted as the percentage change in UPC associated with a given percentage change in price (or average revenue). 0 Q. WHAT IS THE PURPOSE OF THE CDD AND HDD VARIABLES? A. These variables account for weather-related changes in commercial customer electricity use per customer.

15 Q. WHAT IS THE PURPOSE OF THE GRP VARIABLE? A. This variable reflects how commercial customers change their electricity usage in response to changes in economic conditions. Q. WHAT IS THE PURPOSE OF THE PRICE VARIABLE? A. This variable reflects how commercial customers change their electricity usage in response to changes in the price of electricity. 0 Q. WHAT IS THE PURPOSE OF THE MONTHLY INDICATOR VARIABLES? A. These variables account for seasonal changes in commercial customer s electricity use that are not related to other included variables (weather and economic conditions). 0 Q. ARE THE ESTIMATED COEFFICIENTS FROM THE COMMERCIAL UPC MODEL REASONABLE? A. Yes. The estimates can be summarized as follows: Commercial UPC is positively related to CDDs; Commercial UPC has a relatively weak relationship with HDDs; Commercial UPC is positively related to economic conditions, as reflected by GRP; Commercial UPC is negatively related to electricity prices (i.e., UPC decreases when the electricity price increases);

16 Seasonal patterns are relevant explanatory factors (i.e., the coefficients on the indicator variables are statistically significant); and The overall explanatory model of the power is very good, with an R-squared value of Q. PLEASE DESCRIBE THE COMMERCIAL CUSTOMER MODEL. A. The dependent variable in the commercial customer model is the natural log of the number of commercial customers served. The model includes the following explanatory variables: The natural log of the -month moving average of GRP; A monthly time trend variable; and Monthly indicator variables. Q. WHAT IS THE PURPOSE OF THE GRP VARIABLE? A. The GRP variable accounts for the effect of changes in economic conditions on the number of customers served. 0 Q. WHAT IS THE PURPOSE OF THE TIME TREND VARIABLE? A. The time trend variable is intended to identify (and account for) any trend in the number of commercial customers served, controlling for the other included variables (GRP and the monthly indicators).

17 Q. WHAT IS THE PURPOSE OF THE MONTHLY INDICATOR VARIABLES? A. The monthly indicator variables are intended to account for seasonal variations in the number of commercial customers served. 0 Q. DOES THE MODEL PRODUCE REASONABLE ESTIMATES? A. Yes. The findings can be summarized as follows: The number of commercial customers served is positively related to economic conditions (i.e., more customers are served when GRP increases); Controlling for economic conditions, there is a downward trend in the number of commercial customers served; There is seasonal variation in the number of commercial customers served; and The model explains a very high proportion of the variation in customers served, with an R-squared value that rounds to Q. PLEASE SUMMARIZE THE FORECAST THAT RESULTS FROM THE COMMERCIAL UPC AND CUSTOMER MODELS. A. The forecast results in somewhat higher growth than was forecasted for the residential class. From 0 to 00, commercial UPC is forecast to grow by.0 percent per year; the number of commercial customers served is forecast to grow by 0. percent per year; and total commercial sales is forecast to grow by. percent per year. The rate of sales growth declines during the forecast period, ranging from.0 percent in the early years to 0. percent in later years.

18 0 VI. LARGE POWER SERVICE FORECAST Q. PLEASE DESCRIBE BLACK HILLS LARGE POWER SERVICE FORECAST. A. The Large Power Service forecast consists of one model: a sales model. A Large Power Service customer model was also developed as a placeholder in case the Company adopts a UPC approach for this customer class in the future. This customer model is not incorporated in the 0 ERP load forecast because the UPC model plus customer model approach is not as effective for classes consisting of relatively few large customers. Specifically, UPC can vary considerably across the customer within such a class, such that the removal or addition of one (or a few) customers can have a substantial effect on class-level UPC. In this case, it is more straightforward to directly forecast class-level sales. 0 Q. PLEASE DESCRIBE THE LARGE POWER SERVICE SALES MODEL. A. The Large Power Service sales model includes the following variables (or drivers):. CDD0;. The natural log of the -month moving average of gross regional product (GRP); and. Monthly indicator variables. The dependent variable in the model is the natural log of class sales in each month.

19 Q. WHAT IS THE PURPOSE OF THE CDD VARIABLE? A. This variable reflects how Large Power Service customers change their electricity usage in response to changes in weather conditions, specifically cooling load that occurs as temperatures increase. I also tested whether heating load is a relevant factor for this class, but did not find a statistically significant relationship between sales and HDDs. 0 Q. WHAT IS THE PURPOSE OF THE GRP VARIABLE? A. This variable reflects how Large Power Service customers change their electricity usage in response to changes in economic conditions. Q. WHAT IS THE PURPOSE OF THE MONTHLY INDICATOR VARIABLES? A. These variable account for seasonal changes in Large Power Service sales that are not related to other included variables (weather and economic conditions). 0 Q. ARE THE ESTIMATED COEFFICIENTS FROM THE LARGE POWER SERVICE SALES MODEL REASONABLE? A. Yes. The estimates can be summarized as follows: Large Power Service usage has a weakly positive relationship to CDDs (i.e., class sales are not very sensitive to weather conditions); Large Power Service usage has a weakly positive relationship with economic conditions, as reflected by GRP. The failure to estimate a stronger relationship

20 may be due to GRP being an imperfect proxy for the demand for the Large Power Service s products; Seasonal patterns are not as important as they are for other classes. However, note that the seasonal indicator variables are highly collinear with the CDD variable (i.e., they change together across time). If the CDD variable is removed from the model, the monthly indicator variables are statistically significant and vice versa; and The overall explanatory model of the power is good, with an R-squared value of Q. PLEASE SUMMARIZE THE FORECAST THAT RESULTS FROM THE LARGE POWER SERVICE SALES MODEL. A. Large Power Service sales are forecast to increase by.0 percent per year. This value declines somewhat over the forecast period, but not as much as it does for the other customer classes. The growth rate is. percent per year early in the forecast period and 0. percent per year at the end of the forecast period. 0 VII. SYSTEM PEAK DEMAND FORECASTS Q. PLEASE DESCRIBE BLACK HILLS MONTHLY PEAK DEMAND FORECAST. A. Black Hills monthly peak demands are modeled using a single equation that accounts for the effect of weather, economic, and seasonal factors on maximum demands. In order to increase the sample size, the model includes all hours that

21 were within percent of the month s maximum demand. This has the effect of increasing the sample from 0 (the number of months from January 00 through December 0) to. 0 Q. HOW WAS INFORMATION REGARDING LARGE CUSTOMER LOAD ADDITIONS AND REDUCTIONS INCLUDED IN THE PEAK DEMAND FORECAST? A. Black Hills largest customer load is omitted from the peak demand statistical model. This customer has a large and highly variable load that is not easily modeled. In addition, Black Hills did not want to assume that the demand associated with this large load would change at the same rate as the rest of Black Hills load. In place of statistical modeling, Black Hills engaged in discussions with the customer regarding its plans. This provides more detailed information that can be difficult to capture in a statistical model. Therefore, the statistical model omits this customer and the customer-specific forecast is added back into the forecast at the end of the process. 0 Q. WHAT VARIABLES ARE INCLUDED IN THE PEAK DEMAND MODEL? A. The dependent variable is the natural log of the month s peak demand (or hours within percent of the peak). The explanatory variables are: CDD0 on the peak day; HDD0 on the peak day;

22 Cooling degree hours in the peak hour, calculated as: o CDH = MAX(Temp. in F 0, 0) Heating degree hours in the peak hour, calculated as: o HDH = MAX(0 Temp. in F, 0) The month s CDD0 per day; Monthly indicator variables; The natural log of GRP; and The natural log of non-farm employment. 0 Q. WHAT IS THE FUNCTION OF THE VARIOUS WEATHER VARIABLES INCLUDED IN THE MODEL? A. The CDD and CDH variables account for the effect on cooling-related usage on system peak demand. CDD reflects overall daily conditions (the maximum and minimum temperatures) while CDH reflects the temperature during the peak hour. The HDD and HDH variables account for the effect on heating-related usage on system peak demand. The month s CDD per day variable attempts to capture the buildup of heat across days. For example, an isolated hot day may have a smaller effect on system demand than a similar day that was preceded by a number of hot days. 0 Q. WHY ARE THE TEMPERATURE THRESHOLDS FOR THE CDH AND HDH VARIABLES DIFFERENT FROM THOSE OF THE CDD AND HDD VARIABLES? 0

23 0 A. The CDH variable has a higher threshold than the CDD variable to reflect the difference between the hourly and daily perspectives of the variables. That is, the threshold is intended to reflect the temperature at which cooling load begins to occur. The CDD variable is based on the daily average temperature whereas the CDH variable is based on a single hour s temperature. A day with a relatively low average temperature may include hours that are hot enough to induce cooling load. For example, a F daily average temperature may be based on a low of 0 F and a high of 0 F. Customers may not have cooling load when the temperature is as low as F in individual hours, but they may have cooling load on days with that average daily temperature (which include significantly warmer hours). Q. WHAT IS THE FUNCTION OF THE TWO ECONOMIC VARIABLES INCLUDED IN THE MODEL? A. The GRP and employment variables account for the longer-term effects (relative to weather) of economic conditions on system peak demand. My expectation is that improved economic conditions lead to higher system demands over time. 0 Q. WHAT IS THE PURPOSE OF THE MONTHLY INDICATOR VARIABLES? A. These variables account for season variations in system peak demand that are not captured by the other included variables (e.g., weather).

24 0 Q. ARE THE ESTIMATED COEFFICIENTS FROM THE SYSTEM PEAK DEMAND MODEL REASONABLE? A. Yes. The estimates can be summarized as follows: All of the weather variables are positively related to system peak demand. The coefficients on the CDD and CDH variables are higher than the corresponding coefficients on the HDD and HDH variables, indicating a greater influence of cooling loads relative to heating loads on system peak demand. The coefficients on the monthly indicator variables indicate seasonal variation in peak demand, controlling for weather conditions. Both economic variables (GRP and non-farm employment) are positively related to peak demand, as expected. Note that these variables determine the forecast growth rate of peak demand. The weather variables and seasonal indicators help explain and adjust for short-term patterns in demand, but the economic variables determine how demand is forecast to change over time. (The same normal weather conditions and seasonal effects are included in each forecast year.) The model explains a very high proportion of the variation in peak demand, with an R-squared value of Q. PLEASE SUMMARIZE THE FORECAST THAT RESULTS FROM THE SYSTEM PEAK DEMAND MODEL. A. From 0 to 00, system demand is forecast to grow by 0. percent per year. The demand growth declines during the forecast period, ranging from 0. percent in the early years to 0. percent in later years.

25 0 Q. DID YOU HELP PREPARE A CONFIDENCE INTERVAL AROUND THE PEAK DEMAND AND SALES FORECASTS? A. Yes. The base forecasts I have described so far represent the expected (or average) forecast values, which are based on the W&P forecast of economic and demographic variables and normal weather conditions. However, the future values of these drivers are not known with certainty. For example, there may be a recession that is not reflected in the W&P forecast. The confidence interval attempts to capture some of the forecast uncertainties to form a range of outcomes, which are shown as high and low scenarios. 0 Q. WHAT IS THE CONFIDENCE INTERVAL INTENDED TO REFLECT? A. As prepared here, the confidence interval is intended to reflect the effect longerterm variations in economic conditions (in this case, GRP and non-farm employment) on system peak demands. The confidence interval excludes shortterm effects, such as deviations from normal weather conditions. Specifically, the confidence interval is based on the variation in 0-year average growth rates in economic variables from 0 through 0, which are converted into variations in demand using the corresponding estimates from the peak demand forecast model (e.g., the estimate of the effect of changes in GRP on system peak demand).

26 0 Q. PLEASE DESCRIBE HOW THE CONFIDENCE INTERVALS WERE PREPARED. A. The preparation of the confidence intervals was complicated by the fact that the Company had a relatively short historical period over which to observe the variability of Black Hills sales and peak demands (00 to 0). However, a much longer historical time series of economic variables is available in the W&P data. Therefore, the W&P data was combined with the estimates of the effect of changes in the economic variables on system peaks to develop high and low scenarios. The confidence interval incorporates the variability in 0-year average growth rates in economic conditions from to 0 (e.g., the value is the average growth rate from 0 to, the 0 value reflects to 0, etc.). Each of these 0-year growth rates is multiplied by our estimate of the effect of the variable on system peak demand. For example, a percent 0-year growth rate in GRP is multiplied by an estimated coefficient of 0. to estimate a 0. percent growth rate in peak demand due to the increase in GRP. 0 Q. HOW ARE THE 0-YEAR GROWTH RATES USED TO DEVELOP THE CONFIDENCE INTERVALS? A. The 0-year average growth rates in peak demand due to changes in GRP and non-farm employment are used as follows. First, assume that the percentage change in peak demand (or sales) is normally distributed with a mean equal to our forecast value and a standard deviation that reflects the variability in the historical economic data (after normalizing to reflect differences between the historical and

27 forecast average growth rates). Next, find the 0 th and 0 th percentile values from this distribution, which are -0. percent and +. percent, respectively. These are the low and high growth rates that serve as the confidence interval. 0 Q. DO YOU APPLY THIS METHODOLOGY TO BOTH PEAK DEMAND AND SALES FORECASTS? A. Yes, the values listed above are applied to both the peak demand and sales forecasts. I did not develop a separate confidence interval using the sales data because of the complications associated with combining values across customer groups (and therefore across statistical models). Q. WHERE CAN I FIND MORE INFORMATION ABOUT THE METHODS USED TO PRODUCE THE HIGH AND LOW FORECAST SCENARIOS? A. Section. of the ERP contains a detailed description of the methods used to calculate the high and low forecasts. 0 VIII. CONCLUDING REMARKS Q. IS THE METHODOLOGY USED TO DEVELOP THE SALES AND DEMAND FORECASTS REASONABLE? A. Yes. In the sections above, I have explained how each of the estimated models produces reasonable estimated coefficients. Tables DGH- and DGH- demonstrate the reasonableness of the resulting forecasts through comparisons of historical and forecast growth rates in both the forecast outcomes (e.g., sales) and

28 the underlying factors (i.e., the driver variables, such as the number of level intuition about each forecast. For example, the residential UPC growth rate is somewhat higher in the forecast period relative to the historical period (0. percent vs. 0. percent). This corresponds with a higher forecast growth rate in household income, which is the economic variable included in that forecast model. Table DGH-: Historical and Forecast Growth Rates of Forecast Outcomes Class Residential Commercial (SGS+LGS) households). By combining information across the two tables, one can get a high- Annual Growth Rate Outcome UPC 0.% 0.% Customers 0.% 0.% Sales 0.% 0.% UPC -0.%.0% Customers 0.0% 0.% Sales -0.%.% Large Power Service Sales 0.%.0% System MW Month-specific from -.0 to +.; annual =.0% 0.% Note that Table DGH- calculates growth rates using data directly included in and forecast from the methods described in my testimony. In her testimony, Black Hills witness Ms. Seaman describes the adjustments that are made to the forecasts, such as adding in the largest customer s demand and accounting for forecast impacts from demand side management (DSM) programs. These adjustments may lead to different annual growth rates than shown in Table DGH-.

29 0 Table DGH-: Historical and Forecast Growth Rates of Driver Variables Variable Annual Growth Rate Household Income 0.%.% Households.0% 0.% GRP.%.% Commercial Price.% 0.% Non-farm Employment 0.%.% Q. WHAT OTHER INSIGHTS ABOUT THE FORECAST MODELS CAN BE DERIVED FROM TABLES DGH- AND DGH-? A. The forecast of residential customers grows more slowly in the forecast period (0. percent vs. 0. percent in the historical period), which is matched by a reduction in the growth rate in the number of households (from.0 percent to 0. percent). The forecast with the largest difference between the historical and forecast period growth rates is commercial UPC (and, by extension, sales), which declined by 0. percent per year in the historical period but is forecast to grow.0 percent per year from 0 to 00. This appears to be driven by a sharp reduction in the growth rate of the average retail price paid, which increased. percent per year from 00 to 0, but is forecast to grow by only 0. percent per year from 0 to 00. (A small increase in the GRP growth rate contributes to a lesser extent.) Q. DOES THIS CONCLUDE YOUR TESTIMONY? A. Yes.

30 Appendix A Daniel G. Hansen RESUME May 0 Address: 00 University Bay Drive, Suite 00 Madison, WI 0 Telephone: 0.. Fax: dghansen@caenergy.com Academic Background: Ph.D., Michigan State University,, Economics M.A., Michigan State University,, Economics B.A., Trinity University,, Economics and History Positions Held: Vice President, Laurits R. Christensen Associates, Inc. 00 present Senior Economist, Laurits R. Christensen Associates, Inc., 00 Economist, Laurits R. Christensen Associates, Inc., Professional Experience: I work in a variety of areas related to retail and wholesale pricing in electricity and natural gas markets. I have used statistical models to forecast customer usage, estimate customer load response to changing prices, and estimate customer preferences for product attributes. I have developed and priced new product options; evaluated existing pricing programs; evaluated the risks associated with individual products and product portfolios; and developed cost-of-service studies. I have conducted evaluations and provided testimony regarding revenue decoupling and weather adjustment mechanisms.

31 MAJOR PROJECTS: Proceeding No. A- E Assisted a utility in forecasting the load impacts from a new residential peak-time rebate program. Evaluated residential demand response pilot programs with programmable-controllable thermostats. Developed long-term forecasting models for an electric utility. Conducted a review of an electric utility s load forecasting methods. Conducted an independent evaluation of a revenue decoupling mechanism for an electric utility. Estimated load impacts for commercial and industrial demand response programs. Evaluated a straight-fixed variable rate design for a natural gas utility. Estimated the load impacts from a residential peak-time rebate program. Worked with a state's regulatory staff to evaluate alternative electricity pricing structures for residential, commercial, and industrial customers. Assisted a utility in meeting regulatory requirements regarding the allocation of distribution services. Evaluated a residential electricity pricing pilot program. Evaluated the cost effectiveness of automated demand response technologies. Evaluated and modified short- and long-term electricity sales and demand forecasting models. Created a short-term electricity demand forecasting model. Prepared testimony regarding the return on equity effects associated with natural gas revenue decoupling mechanisms. Conducted an independent evaluation of two natural gas revenue decoupling mechanisms Created forecasts of load impacts from electricity demand response programs. Estimated historical the load impacts from electricity demand response programs. Prepared testimony regarding a proposed natural gas decoupling mechanism. Prepared testimony regarding the weather normalization of test year sales and revenues. Participated on a regulatory proceeding panel to discuss decoupling mechanisms. Prepared testimony regarding a proposed electricity decoupling mechanism. Prepared a report and testimony regarding a natural gas decoupling mechanism.

32 Evaluated a model that estimated the costs associated with removing and relicensing hydroelectric facilities. Assisted an electric utility in evaluating new rate options for commercial and industrial customers. Designed and evaluated time-of-use and critical-peak pricing rates for an electric utility. Reviewed cost-of-service study for a municipal electric utility. Produced a report on rate design methods that provide appropriate incentives for demand response and energy efficiency. Assisted in wholesale power procurement process. Evaluated a weather-adjustment mechanism for a natural gas utility. Assessed weather-related fixed cost recovery risk for an electric utility. Evaluated a revenue decoupling mechanism for a natural gas utility. Estimated price responsiveness of real-time pricing customers. Evaluated the need for electricity transmission and distribution standby rates for a utility. Developed a market share simulation model using conjoint survey results of electricity distributors. Conducted conjoint surveyed of electricity distributors regarding rate structure preferences. Developed a method to calculate a retail forward contract risk premium. Prepared a report on the performance of Financial Transmission Rights (FTRs) in the PJM electricity market. Reviewed a retail pricing model for use in a competitive electricity market. Provided support in a natural gas rate case filing. Simulated outcomes associated with alternative wholesale rate offers to electricity distributors. Developed a business case to support a natural gas fixed bill product. Assessed the accuracy of a natural gas fixed bill pricing algorithm. Audited an evaluation of the costs associated with implementing a renewable portfolio standard. Developed a model to value interruptible provisions in a long-term customer contract. Performed a study on the determinants of electricity price differences across utilities and regions. Developed long-term demand and energy forecasts. Conducted market research to assess customer interest in new product options.

33 Recommended new retail pricing products for commercial and industrial customers. Prepared a report on the fundamentals of retail electricity risk management. Prepared a report that presented a taxonomy of retail electricity pricing products. Presented at a workshop in Africa regarding deregulated electricity markets. Prepared a report on the effectiveness of distributed resources in mitigating price risk. Performed a valuation of energy derivatives consistent with FAS. Created an electricity market share forecasting model. Developed standby rates for an electric utility. Developed an electricity wholesale price forecast. Forecasted retail customer loads for an electric utility. Assisted in mediating a new product development process with a utility and its industrial customers. Developed a model that simulates wholesale market price changes due to retail load response. Developed a pricing model for an innovative financial product. Estimated changes in wholesale electricity prices due to customer load response. Oversaw creation of software that estimates customer satisfaction with utilities. Developed a model to economically evaluate a capital addition to a generator. Developed a wholesale version of the Product Mix Model. Evaluate Risk Implications of New Product Offering. Mixed Logit Estimation of Customer Preferences. Estimation of Customer Price Responsiveness. Product Mix Model Workshops. Unbundling and Rate Design. Development of a Computer Program. Large Commercial and Industrial Customer Rate Analysis. Residential Customer Rate Analysis. Survey of Power Marketers. Development of Multi-Period Analysis Tool.

34 Evaluating the Effect of Alternative Rates on System Load. Estimating the Persistence of Weather Patterns. Electricity Customer Survey Data Analysis. Product Mix Analysis for Small Customers. Survey of Postal Facilities. Professional Papers: Proceeding No. A- E 0 Load Impact Evaluation of Pacific Gas and Electric Company s Residential Time-Based Pricing Programs: Ex-post and Ex-ante Report, with Steven Braithwait and David Armstrong, 0. 0 Load Impact Evaluation of Pacific Gas and Electric Company s Mandatory Time-of-Use Rates for Small, Medium, and Agricultural Non-residential Customers: Ex-post and Ex-ante Report, with Marlies Patton, 0. 0 Load Impact Evaluation of California Statewide Demand Bidding Programs (DBP) for Non-Residential Customers: Ex-post and Ex-ante Report, with Michael Ty Clark, 0. 0 Load Impact Evaluation of California Statewide Base Interruptible Programs (BIP) for Non-Residential Customers: Ex-post and Ex-ante Report, with Tim Huegerich, 0. Statewide Time-of-Use Scenario Modeling for 0 California Energy Commission Integrated Energy Policy Report, with Steven Braithwait and David Armstrong, 0. 0 Statewide Load Impact Evaluation of California Aggregator Demand Response Programs: Ex-post and Ex-ante Load Impacts, with Steven Braithwait and David Armstrong, 0. 0 Load Impact Evaluation of California Statewide Demand Bidding Programs (DBP) for Non-Residential Customers: Ex-post and Ex-ante Report, with Steven Braithwait and David Armstrong, 0. 0 Load Impact Evaluation of California Statewide Base Interruptible Programs (BIP) for Non-Residential Customers: Ex-post and Ex-ante Report, with Tim Huegerich, 0. 0 Load Impact Evaluation of Southern California Edison s Mandatory Time-of-Use Rates for Small and Medium-Sized Business and Agricultural Customers: Ex-post and Ex-ante Report, with Marlies Patton, 0. 0 Load Impact Evaluation of Pacific Gas and Electric Company s Mandatory Time-of-Use Rates for Small and Medium Non-residential Customers: Ex-post and Ex-ante Report, with Marlies Patton, 0. FirstEnergy s Smart Grid Investment Grant Consumer Behavior Study, with EPRI (B. Neenan) and Marlies Patton, 0.

35 Proceeding No. A- E An Evaluation of Portland General Electric s Decoupling Adjustment, Schedule, with Robert J. Camfield and Marlies C. Hilbrink, 0. "Evaluation of the Straight-Fixed Variable Rate Design Implemented at Columbia Gas of Ohio," with Marlies C. Hilbrink, 0. "The Effect on Electricity Consumption of the Commonwealth Edison Customer Application Program Pilot," with EPRI and CA Energy Consulting staff, 0. "The Effects of Critical Peak Pricing for Commercial and Industrial Customers for the Kansas Corporation Commission," with David A. Armstrong, 0. Meeting Commonwealth Edison s Distribution Allocation Requirements from Illinois Commerce Commission Order 0-0, with Michael O Sheasy, A. Thomas Bozzo, and Bruce Chapman, 0. "Residential Rate Study for the Kansas Corporation Commission," with Michael T. O'Sheasy, 0. "An Evaluation of the Conservation Incentive Program Implemented for New Jersey Natural Gas and South Jersey Gas," with Bruce R. Chapman, 00. A Review of Natural Gas Decoupling Mechanisms and Alternative Methods for Addressing Utility Disincentives to Promote Conservation, June 00. Evaluation of the Klamath Project Alternatives Analysis Model: Reply to Addendum A of the Consultant Report Prepared for the California Energy Commission Dated March 00, May 00, with Laurence D. Kirsch and Michael P. Welsh. Evaluation of the Klamath Project Alternatives Analysis Model, March 00, with Laurence D. Kirsch and Michael P. Welsh. A Review of the Weather Adjusted Rate Mechanism as Approved by the Oregon Public Utility Commission for Northwest Natural, October 00, with Steven D. Braithwait. A Review of Distribution Margin Normalization as Approved by the Oregon Public Utility Commission for Northwest Natural, March 00, with Steven D. Braithwait. Analysis of PJM s Transmission Rights Market, EPRI Report #00, December 00, with Laurence Kirsch. Using Distributed Resources to Manage Price Risk, EPRI Report #00, November 00, with Michael Welsh. Hedging Exposure to Volatile Retail Electricity Prices, The Electricity Journal, Vol., number, pp., June 00, with A. Faruqui, C. Holmes and B. Chapman. Weather Hedges for Retail Electricity Customers, with C. Holmes, B. Chapman and D. Glyer. In papers for EPRI International Pricing Conference 000. Worker Performance and Group Incentives: A Case Study, Industrial and Labor Relations Review, Vol., No., pp., October.

36 Worker Quality and Profit Sharing: Does Unobserved Worker Quality Bias Firm-Level Estimates of the Productivity Effect of Profit Sharing? Working Paper, May. Supervision, Efficiency Wages, and Incentive Plans: How Are Monitoring Problems Solved? Working Paper, November, presented at the Western Economics Association Meetings,. Has Job Stability Declined Yet? New Evidence for the 0 s, with David Neumark and Daniel Polsky, The Journal of Labor Economics,. Testimony and Reports before Regulatory Agencies: UNS Electric, Inc., Arizona Docket No. E 00A--0: Testimony supporting a residential demand charge proposed by UNS Electric on behalf of the Arizona Investment Council, 0. Public Service Company of New Mexico (PNM), New Mexico Case No. -00-UT: Testimony supporting a revenue decoupling mechanism on behalf of PNM, 0. Public Service Company of New Mexico (PNM), New Mexico Case No. -00-UT: Testimony supporting a revenue decoupling mechanism on behalf of PNM, 0. Xcel Energy, Inc, Minnesota E00/GR--: Testimony supporting a revenue decoupling mechanism on behalf of Xcel Energy, 0. Arizona Public Service Company, Arizona Docket No. E 0A 0: Testimony supporting a revenue decoupling mechanism proposed by APS on behalf of the Arizona Investment Council, 0. Southwest Gas Corporation, Arizona Docket No. G 0A 0 0: Testimony supporting a revenue decoupling mechanism contained in a settlement agreement on behalf of the Arizona Investment Council, 0. Otter Tail Power Company, Minnesota Docket No. E 0/GR 0 : Testimony regarding the weather normalization of test year sales in a general rate case on behalf of Otter Tail Power Company, 00. Southwest Gas Corporation, Nevada Docket No : Testimony regarding a the return on equity effects associated with a proposed revenue decoupling mechanism on behalf of Southwest Gas Corporation, 00. Southwest Gas Corporation, Arizona Docket No. G 0A 0 00: Testimony regarding a proposed revenue decoupling mechanism on behalf of the Arizona Investment Council, 00. Otter Tail Power Company, Minnesota Docket No. E 0/GR 0 : Testimony regarding the weather normalization of test year sales and revenues in a general rate case on behalf of Otter Tail Power Company, 00. Massachusetts Department of Public Utilities, Docket No. DPU 0 0: Participation in a panel regarding an Investigation into Rate Structures that will Promote Efficient Deployment of Demand Resources, on behalf of Environment Northeast, 00.

37 Connecticut Light & Power Company, Docket No : Testimony regarding a proposed electricity revenue decoupling mechanism on behalf of Environment Northeast, 00. Questar Gas Company, Docket No. 0 0 T0: Testimony regarding the effectiveness of a natural gas revenue decoupling mechanism on behalf of the Utah Division of Public Utilities, 00. PacifiCorp, FERC Docket No. 0: Evaluation of the Klamath Project Alternatives Analysis Model: Reply to Addendum A of the Consultant Report Prepared for the California Energy Commission Dated March 00, May 00, with Laurence D. Kirsch and Michael P. Welsh. PacifiCorp, FERC Docket No. 0: Evaluation of the Klamath Project Alternatives Analysis Model, March 00, with Laurence D. Kirsch and Michael P. Welsh. Northwest Natural Gas Company, Oregon Docket UG : Testimony relating to an investigation regarding possible continuation of Distribution Margin Normalization, May 00. Northwest Natural Gas Company, Oregon Docket UG : Submitted a report in compliance with a requirement to evaluate the functioning of the Weather Adjusted Rate Mechanism, October 00.