California’s Statewide Pricing Pilot Summer 2003 Impact Evaluation 17 th Annual Western Conference, San Diego, California Ahmad Faruqui and Stephen S. George Charles River Associates June 25, 2004

CHARLES RIVER ASSOCIATES Work in Progress (6/8/04)—Subject to Revision 2 Outline Design of the Statewide Pricing Pilot (SPP) Methodology and data Residential results Price elasticity estimates Impact simulations

CHARLES RIVER ASSOCIATES Work in Progress (6/8/04)—Subject to Revision 3 The genesis of the Statewide Pricing Pilot (SPP) It is an outgrowth of the CPUC OIR (R ) on advanced metering and demand response The first large-scale scientific experiment focused on dynamic pricing for mass-market consumers Customer enrollment began in April 2003; new rates became effective in July 2003 and will stay in effect through December 2004 SPP addresses several policy issues: What is the price elasticity of demand for electricity by time period? Does responsiveness vary by rate type, climate zone and customer characteristics? Will customers accept time-varying and dynamic rates? Are reductions in energy use and coincident peak demand resulting from widespread use of more economically efficient pricing sufficiently large to off- set the metering costs required to implement rate reform?

CHARLES RIVER ASSOCIATES Work in Progress (6/8/04)—Subject to Revision 4 The SPP is testing several rate options Time-of-Use (TOU) rate Traditional two-part TOU rate Peak period from 2 pm to 7 pm Rates vary seasonally Critical Peak Pricing-Fixed (CPP-F) rate TOU rate 350 days a year Much higher price during peak period on up to 15 days a year, which are called the previous evening Critical Peak Pricing-Variable (CPP-V) rate Similar to CPP-F except they may be called in just 4 hours Critical peak period can vary in length from 1 to 5 hours between 2 pm and 7 pm Both treatment and control group consumers had volunteered into a smart thermostat pilot program funded by Assembly Bill 970 The above rates are layered on top of a very complex, five-tier, increasing block rate structure

CHARLES RIVER ASSOCIATES Work in Progress (6/8/04)—Subject to Revision 5 Additional SPP design features About 2,500 participants drawn from three investor-owned utilities allocated to various treatment and control groups There are multiple price levels and ratios for each rate type in order to allow for estimation of all own-price and cross-price elasticities A mandatory pilot was not politically acceptable Customers were randomly selected but not required to participate The pilot design attempts to mimic a voluntary “opt-out” pricing regime Residential sample segmented into four climate zones C&I sample segmented by size <20 kW and between 20 kW and 200 kW

CHARLES RIVER ASSOCIATES Work in Progress (6/8/04)—Subject to Revision 6 The experiment includes four climate zones

Methodology and Data

CHARLES RIVER ASSOCIATES Work in Progress (6/8/04)—Subject to Revision 8 Two demand models were used to estimate customer price responsiveness Double Log. This expresses the log of peak and off-peak usage as a function of the log of peak and off-peak prices and cooling degree hours during each period Constant elasticity-of-substitution (CES). This expresses the log of the ratio of peak to off-peak usage as a function of the log of the ratio of peak to off-peak price and of the difference in cooling degree hours during the two periods; another equation expresses daily energy usage as a function of daily price and cooling degree hours Both functional forms are estimated using the “fixed-effects” estimation procedure The appendix contains more methodological information

CHARLES RIVER ASSOCIATES Work in Progress (6/8/04)—Subject to Revision 9 Data Characteristics For the CPP-F and CPP-V rates, the demand models were estimated using values that were averaged over all days in the following three time periods Pretreatment period (June only) Non-CPP days in the treatment period CPP days in the treatment period Consequently, there are three time-series observations for each customer, with treatment customers facing a different price for each time period and control customers facing the same price each time The demand models were estimated using pooled time-series, cross- section data A binary variable was used to test whether the price elasticity varies between CPP and non-CPP days for each climate zone. No statistically significant difference was found in zones 2, 3 and 4 using the double-log model and in zones 1, 3 and 4 using the CES model For the TOU models, the regressions were run using values averaged over two time periods Pretreatment period (June only) All weekdays during the treatment period

Residential Analysis Results

CHARLES RIVER ASSOCIATES Work in Progress (6/8/04)—Subject to Revision 11 Key findings for the CPP-F rate Double Log: Own-price elasticities for peak period energy use are statistically significant in zones 2, 3 and 4 Range from low of –0.08 to high of –0.21 Higher in the warmer zones 3 and 4 and lower in the cooler zones 1 and 2 Own-price elasticities for off-peak energy use are statistically significant in zones 1 and 2 Zone 1 elasticity is –0.17 and –0.10 Cross-price elasticities are typically small and often insignificant CES: All elasticities of substitution are small but statistically significant; they range between –0.04 to –0.16, with higher values being observed in the warmer zones The daily price elasticities in zones 3 and 4 are statistically significant and equal –0.06; those in zones 1 and 2 are not significant

CHARLES RIVER ASSOCIATES Work in Progress (6/8/04)—Subject to Revision 12 Key findings for the CPP-V rate Double Log: The own price elasticity for peak period energy use is –0.21 and statistically significant The own price elasticity for off-peak energy use is not statistically significant Cross-price elasticities are typically small or not statistically significant CES: The elasticity of substitution is –0.21 The daily price elasticity is not statistically significant It is important to note that these elasticities pertain to customers who had already volunteered into an earlier pilot program involving smart thermostats and therefore cannot be generalized to the population as a whole

CHARLES RIVER ASSOCIATES Work in Progress (6/8/04)—Subject to Revision 13 Key findings for the TOU rate Double Log: None of the own-price or cross-price elasticities are statistically significant; Interestingly, CPP-F customers on non- CPP days,who face a TOU rate, display significant price elasticities CES: The elasticity of substitution for zones 2 and 3 is statistically significant and ranges between –0.11 and –0.28 None of the daily price elasticities are significant There are a couple of reasons why the TOU price elasticities are not as statistically significant as the CPP-F price elasticities The sample sizes of the TOU customers are smaller than those for the CPP-F rate Without a CPP rate that is exercised a few days each month, customers may forget they are on a time-varying rate

CHARLES RIVER ASSOCIATES Work in Progress (6/8/04)—Subject to Revision 14 Rate impacts can be simulated by using the estimated demand models The estimated demand models can simulate the impact for the rates used in the experiment and for a variety of other rates that are generally similar to the ones used in the experiment They should NOT be made using the point elasticities in the previous slides, since point elasticities will tend to exaggerate the impact of large price increases and underestimate the impact of large price decreases It is important to note that impact simulations require both own and cross-price effects, another reason why relying on just the point estimates of the own-price elasticity will mislead rather than inform policy analysis

CHARLES RIVER ASSOCIATES Work in Progress (6/8/04)—Subject to Revision 15 A Double-Log example of how point elasticities over- state the impact of large price increases

CHARLES RIVER ASSOCIATES Work in Progress (6/8/04)—Subject to Revision 16 Impact analysis is based on the weighted average prices for treatment customers in each climate zone Control Group Average Price 13.3 cents/kWh

CHARLES RIVER ASSOCIATES Work in Progress (6/8/04)—Subject to Revision 17 Peak period impacts are larger in the hotter climate zones than in the cooler zones These estimates come from a sub-sample of customers who volunteered into a smart thermostat pilot and may not be generalizable to the general population of residential customers

CHARLES RIVER ASSOCIATES Work in Progress (6/8/04)—Subject to Revision 18 Impacts are considerably smaller on non- CPP Days, which have lower peak prices These estimates come from a sub-sample of customers who volunteered into a smart thermostat pilot and may not be generalizable to the general population of residential customers

CHARLES RIVER ASSOCIATES Work in Progress (6/8/04)—Subject to Revision 19 Impacts vary somewhat by experimental rate These estimates come from a sub-sample of customers who volunteered into a smart thermostat pilot and may not be generalizable to the general population of residential customers

CHARLES RIVER ASSOCIATES Work in Progress (6/8/04)—Subject to Revision 20 Impacts also vary somewhat by model specification These estimates come from a sub-sample of customers who volunteered into a smart thermostat pilot and may not be generalizable to the general population of residential customers

CHARLES RIVER ASSOCIATES Work in Progress (6/8/04)—Subject to Revision 21 Conclusions Customers show significant response to both the CPP-F and CPP-V rates Impacts are higher in the hotter zones for both CPP and non CPP days Responses are substantially higher on CPP days than on non-CPP days For all zones, the CPP day impact is -12% and the non-CPP day impact is –2.3% CPP day impacts differ slightly between the two experimental rates within the CPP-F rate Results are generally similar across the two functional forms tested in this study Customers respond to TOU rates in zones 2 and 3; and they respond to the TOU portion of the CPP-F rate in zones 3 and 4

CHARLES RIVER ASSOCIATES Work in Progress (6/8/04)—Subject to Revision 22 Next Steps Assess whether price responsiveness varies with weather conditions Assess whether price responsiveness varies with customer characteristics such as the ownership of central air-conditioners Analyze the response of residential customers in the Information Only treatment cells of Track A (Zones 2 and 3) Analyze the response of residential customers in Track B who face an enhanced community-based information treatment Evaluate price responsiveness during Winter Evaluate the response of small C&I customers These results will become available by the end of July

Appendix

CHARLES RIVER ASSOCIATES Work in Progress (6/8/04)—Subject to Revision 24 Double-Log demand model specification This model is estimated with a “fixed-effects” estimation procedure that includes customer-specific constant terms. ln(Q i ) =  +  i d i +  i ln(P i ) +  j ln(P j ) +  (CDH i ) +  where Q i = average daily energy use in the i th period P i = average price during the i th period CDH i = cooling degree hours during the i th period d i = a binary variable equal to 1 for the i th customer, 0 otherwise; these are the fixed effects that don’t vary over time and are customer specific  ii = the own price elasticity of demand for energy during the ith period  ij = the cross price elasticity of demand for energy during the ith period with respect to the price during the jth period  = regression error term

CHARLES RIVER ASSOCIATES Work in Progress (6/8/04)—Subject to Revision 25 Constant Elasticity-Of-Substitution (CES) demand model specification The CES demand model involves two equations, one for assessing the rate of substitution between peak and off-peak energy use and the other for measuring the impact of price changes on daily energy use. The substitution equation has the following form: ln(Q i /Q j ) =  +  i D i +  ln(P i /P j ) +  (CDH i - CDH j ) +  where Q i = average daily energy use per hour in the i th period P i = average price during the i th period CDH i = cooling degree hours per hour during the i th period D i = a binary variable equal to 1 for the i th customer, 0 otherwise  = the elasticity of substitution between peak and off-peak energy use  = regression error term

CHARLES RIVER ASSOCIATES Work in Progress (6/8/04)—Subject to Revision 26 CES demand model specification (continued) The daily demand model specification included in the CES system has the following form. ln(Q d ) =  +  i D i +  p ln(P d ) +  (CDH d ) +  where Q d = average daily energy use per hour P d = average daily price CDH d = average daily cooling degree hours d i = a binary variable equal to 1 for the i th customer, 0 otherwise  p = the own price elasticity of demand for daily energy use  = regression error term

CHARLES RIVER ASSOCIATES Work in Progress (6/8/04)—Subject to Revision 27 The price elasticities of the CES function  ii  =   d z i + w j   ij  =  d z j - w i 

CHARLES RIVER ASSOCIATES Work in Progress (6/8/04)—Subject to Revision 28 Price elasticities and elasticities of substitution for residential energy use for CPP-F tariff Values in bold type are statistically significant at the 95% confidence level ; these point estimates should not be used for making impact simulations Ahmad Faruqui: Footnote elasticities Ahmad Faruqui: Footnote elasticities

CHARLES RIVER ASSOCIATES Work in Progress (6/8/04)—Subject to Revision 29 Price elasticities and elasticities of substitution for residential energy use for CPP-V tariff Values in bold type are statistically significant at the 95% confidence level; these point estimates should not be used for making impact simulations; these estimates come from a sub-sample of customers who volunteered into a smart thermostat pilot and may not be generalizable to the general population of residential customers Ahmad Faruqui: Two footnotes Ahmad Faruqui: Two footnotes

CHARLES RIVER ASSOCIATES Work in Progress (6/8/04)—Subject to Revision 30 Price elasticities and elasticities of substitution for residential energy use for TOU tariff Values in bold type are statistically significant at the 95% confidence level; these point estimates should not be used for making impact simulations