Dynamic Pricing: Transitioning from Experiments to Full Scale Deployments

Transcription

1 Dynamic Pricing: Transitioning from Experiments to Full Scale Deployments Michigan Retreat on Peak Shaving to Reduce Tasted Energy Sanem Sergici, Ph.D. August 06, 2014 Copyright 2014 The Brattle Group, Inc.

2 Why does price responsive demand (PRD) matter? Avoided or deferred resource costs Reduced wholesale market prices Fairness in retail pricing Customer bill reductions Facilitating deployment of distributed generation resources Environmental benefits 1 brattle.com

3 More than 200 time-varying rate tests have been conducted, with a broad range of outcomes Results of Recent Time Varying Pricing Pilots 2 brattle.com

4 When the impacts are grouped by rate type, a pattern begins to emerge Results of Recent Time Varying Pricing Pilots Note: Chart shows only price treatments and excludes treatments with enabling technology 3 brattle.com

5 Enabling technologies significantly boost customer price response Price Response with and without Enabling Technology Notes: Chart shows only treatments testing price and technology side by side Technology included automation (e.g. smart thermostats) and/or information (e.g., in home displays) 4 brattle.com

6 The peak-to-off-peak price ratio further explains some of the variation in impacts 60% 50% Price Only Price+Tech Peak Reduction 40% 30% 20% 10% 0% Peak to Off Peak Price Ratio Note: 2 Price only outliers were removed from the regression 5 brattle.com

7 We ve learned a lot from these pilots Customers respond to prices, but there is much more to the equation Impacts persist across several years and consecutive events Enabling technologies boost price responsiveness Responsiveness dependent on temperature and humidity Decisions made during implementation planning can dramatically impact results Simplicity of program design Activities raising awareness in customers, i.e., pre and post event messaging Whether a program is opt out or opt in 6 brattle.com

8 Case Study: Baltimore Gas & Electric s Smart Energy Pricing (SEP) Pilot Scientifically valid sample design and M&V method Continued for four consecutive summers ( ) The Brattle Group carried out the impact evaluation for each of the summers and analyzed whether price responsiveness persists over time More than 11 different treatments tested over the course of four years Nearly 950 treatment customers at its height Yielded invaluable information for the design of BGE s full scale pricing program 7 brattle.com

9 SEP pilot tested 11 treatments over the course of four years Peak Time Rebate (Price Only) X X X X Peak Time Rebate + Energy Orb X X Peak Time Rebate + Energy Orb + AC Switch Peak Time Rebate + Energy Orb + Smart Thermostat Peak Time Rebate + Smart Thermostat Dynamic Peak Price (DPP)* Peak Time Rebate + Change in Notification Period X X Peak Time Rebate + Change in Event Window Peak Time Rebate + In Home Display/Portal X X Peak Time Rebate + Legacy DLC Program Legacy DLC Program Control Group X X X X * DPP = CPP + TOU. This was also combined with an Energy Orb and AC Switch X X X X X X X

10 Findings informed BGE s full scale deployment of dynamic pricing Customers were responsive to the price signals Customers responded similarly to the CPP and PTR rates Elasticity of substitution ranged from to (depending on weather conditions) The peak impacts ranged from 23 to 34 percent at a 10:1 price ratio (depending on the weather) Daily price elasticity was estimated at 0.05 Customers who were on the price only treatment for four years showed persistence in their price responsiveness BGE rolled out PTR rates to 315,000 customers in the Summer of 2013 Default tariff for residential customers 82% of the customer engaged 9 brattle.com

11 Residential dynamic pricing is transitioning to a new phase: full scale deployment Several utilities are achieving significant participation through aggressive opt in programs Time of use (TOU) rates at APS and SRP Variable peak pricing (VPP) at OG&E Others are rolling out default programs for the mass market Pepco BGE Sacramento Municipal Utility District (SMUD) The Province of Ontario, Canada 10 brattle.com

12 Residential TOU Enrollment Rates Source: The Brattle Group 11 brattle.com

13 Residential Dynamic Pricing Enrollment Rates Source: The Brattle Group 12 brattle.com

14 Case Study: Ontario s Residential TOU Program Besides Italy, Ontario is the only region in the world to deploy Time of Use (TOU) rates for generation charges to all customers who stay with regulated supply TOU rates were deployed in Ontario to incentivize customers to curtail electricity usage during the peak period and possibly to reduce overall electricity usage The Brattle Group was retained by Ontario Power Authority to undertake the impact evolution of the TOU program Three year assignment; the 1 st Year Impact Evaluation results are presented here, the 2 nd year study is underway 13 brattle.com

15 Impact Evaluation Challenges Non experimental design Sheer size of data (hourly data on four million customers over three years) Presence of more than 70 LDCs in the province Mild peak and off peak price ratio (1.5:1) Multiple pricing periods Customers could opt out of regulated TOU tariff and switch to retail providers 14 brattle.com

16 Impacts Measured in the Study For each LDC analyzed in the Study, we quantified the following impacts and elasticities by customer class and by season: Load shifting impacts Peak demand impacts Conservation impacts Elasticity of substitution Overall conservation elasticity These results and the details of the study are discussed in, Impact Evaluation of Ontario s TOU Rates: First Year Analysis, report prepared for Ontario Power Authority, November 's_time of Use_Rates First_Year_Analysis_Faruqui_et_al_Nov_26_2013.pdf? brattle.com

17 An ideal impact evaluation study has two important design elements 1 A control group to serve as a proxy for the behavior of the treatment group customers in the absence of a treatment 2 Pre treatment period data on both control and treatment customer groups to net out the pre existing differences in between the two groups Control Group Treatment Group Pre Treatment C1 T1 Post Treatment C2 T2 In this construct, the true impact would be measured by: (T2 T1) (C2 C1) 16 brattle.com

18 Full-scale deployment of TOU rates poses two challenges for impact evaluation Problem 1 The TOU rollout was not designed as a randomized controlled experiment, there was no control group 2 For some LDCs, the TOU rates were deployed very shortly after the AMI deployment. This implies that there is a very short window with pre TOU data available Solution In the 1 st year study, we took advantage of the phased nature of the TOU roll out within the LDCs to construct a proxy control group Eligible customers were allocated into two groups using the median TOU start date Treatment and control groups were randomly selected the from these two buckets In the sample design process, we defined eligible customers to be included in the study as those who had: at least 6 months of pre TOU data at least12 months of post TOU data 17 brattle.com

19 Overview of the Methodology We employ a two pronged approach to measure TOU rate impacts Estimate an advanced model of consumer behavior called the addilog demand system to discern load shifting effects that are triggered by the TOU rates and to estimate inter period elasticities of substitution Estimate a simple monthly consumption model to understand the overall conservation behavior of the customers and estimate an overall price elasticity of demand By using the parameter estimates from these two models and solving them together, we calculate the impact that TOU rates on energy consumption by pricing period and for the month as a whole 18 brattle.com

20 Our models were estimated over several pricing periods Summer Time Periods (May October) Winter Time Periods (January April, November and December Period Hours TOU Window 1 Weekends & Holidays 2 9 pm 7 am Off peak 3 7 am 11 am Mid peak 4 11 am 5 pm Peak 5 5 pm 7 pm Mid peak 6 (*) 7 pm 9 pm Off peak Period Hours TOU Window 1 Weekends & Holidays 2 9 pm 7 am Off peak 3 7 am 11 am Peak 4 11 am 5 pm Mid peak 5 5 pm 7 pm Peak 6 (*) 7 pm 9 pm Off peak Privileged and Confidential Prepared at the Request of Counsel 19 brattle.com

21 Overview of Residential Class Results There is significant evidence of load shifting across all LDCs Reduction in usage in the peak and mid peak periods (generally highest in the peak periods), increase in usage in the off peak periods Load shifting is higher in the summer rate period than the winter Summer peak period impacts range from 2.6% to 5.7% Winter peak period impacts range from 1.6% to 3.2% Peak period substitution elasticities range from 0.12 to 0.27 Evidence on conservation is limited 20 brattle.com

22 Ontario residential TOU impacts compared to TOU pilots from around the globe 30% TOU Only Arc with OPA Summer Impacts (N = 42) Peak Reduction 20% 10% LDC#4 LDC#2 LDC#1 0% LDC# Peak to Off Peak Price Ratio 21 brattle.com

23 General Service Class Conclusions There is some evidence of load shifting across all LDCs Reduction in usage in the peak and mid peak periods, generally highest in the peak periods. Small increase in usage in the offpeak periods A few odd results, most likely an artifact of the heterogeneity in the General Service class Impacts far smaller for General Service than Residential class No clear pattern of winter versus summer load shifting impacts Summer peak period impacts range from 0% to 0.6% Winter peak period impacts range from 0.2% to 1% Evidence on conservation is negligible 22 brattle.com

24 Concluding remarks Residential customers responded to the TOU rates by shifting their usage from peak to off peak and mid peak periods The load shifting impacts for general service customers were far smaller than those estimated for the residential customer class and results are not as distinct, with some odd substitution patterns Evidence on energy conservation was negligible and generally insignificant in both the residential and general service class We found no evidence of self selection bias associated with customers opting out of the TOU rates into retail rates 23 brattle.com

25 Bibliography Ahmad Faruqui and Sanem Sergici, Dynamic pricing of electricity in the mid Atlantic region: econometric results from the Baltimore gas and electric company experiment,, Journal of Regulatory Economics, Volume 40: Number 1, August Ahmad Faruqui and Sanem Sergici, Household response to dynamic pricing of electricity a survey of 15 experiments, Journal of Regulatory Economics (2010), 38: Ahmad Faruqui, Sanem Sergici and Ryan Hledik, Piloting the Smart Grid, with Ryan Hledik and Sanem Sergici, The Electricity Journal, Volume 22, Issue 7, August/September 2009, pp Peter Fox Penner, Smart Power Climate Change, the Smart Grid, and the Future of Electric Utilities, Island Press, brattle.com

26 APPENDIX 25 brattle.com

27 Determining Sample Sizes We expected the peak and conservation impacts flowing from the TOU rates to be small due to the mild ratio of peak to offpeak prices This implied that we would need larger sample sizes to be able to detect a statistically significant impact than other studies which had used higher price ratios We conducted statistical power calculations to determine the minimum treatment and control group sizes to achieve a predetermined statistical precision level Roughly 106,000 residential customers and 150,000 general service customers were sampled 26 brattle.com

28 The Econometric Estimation of the Addilog system We estimate the Addilog system using the seemingly unrelated regression (SUR) estimation routine SUR employs random effects estimator in the context of unbalanced panels (time invariant fixed effects are accounted for first differences) Parameter estimates from the Addilog system readily yield elasticity of substitution for all five periods relative to the 1 st period Other elasticities (such as own price and cross price elasticities), can also be derived from the estimated addilog system but that is not a trivial task 27 brattle.com

29 The Monthly Conservation Model Where: X may refer to non TOU variables such as weather, socio demographic variables, etc. e refers to We estimate the monthly conservation model using fixed effects estimation corrected for the 1 st order autocorrelation Parameter estimates from this equation yield the overall price elasticity of demand 28 brattle.com

30 Generalized Addilog System Where: X may refer to non TOU variables such as weather, census characteristics, etc. v refers to random disturbance 29 brattle.com

31 Presenter Information SANEM SERGICI Senior Associate Cambridge Dr. Sanem Sergici is a Senior Associate in The Brattle Group s Cambridge, MA office with expertise in electricity markets, applied econometrics, and industrial organization. At Brattle, the focus of Dr. Sergici s work has been on assisting electric utilities, regulators, and wholesale market operators in their strategic questions related to energy efficiency, demand response, and customer behavior in the context of Smart Grid. Dr. Sergici has significant expertise in the design and evaluation of a variety of demand response and behavior based energy efficiency programs; development of load forecasting models; ratemaking for electric utilities; and energy litigation. She has recently completed several long term resource planning projects that involve the development of scenarios and strategies for an electric system to meet long range electric demand while considering the growth of renewable energy, energy efficiency, other demand side resources. She has spoken at several industry conferences and published in several industry journals. The views expressed in this presentation are strictly those of the presenter(s) and do not necessarily state or reflect the views of The Brattle Group, Inc. 30 brattle.com

32 About Brattle The Brattle Group provides consulting and expert testimony in economics, finance, and regulation to corporations, law firms, and governments around the world. We aim for the highest level of client service and quality in our industry. We are distinguished by our credibility and the clarity of our insights, which arise from the stature of our experts, affiliations with leading international academics and industry specialists, and thoughtful, timely, and transparent work. Our clients value our commitment to providing clear, independent results that withstand critical review. 31 brattle.com

33 Our Practices PRACTICES Antitrust/Competition Commercial Damages Environmental Litigation and Regulation Intellectual Property International Arbitration International Trade Product Liability Regulatory Finance and Accounting Risk Management Securities Tax Utility Regulatory Policy and Ratemaking Valuation INDUSTRIES Electric Power Financial Institutions Health Care Products and Services Natural Gas and Petroleum Telecommunications and Media Transportation Privileged and Confidential Prepared at the Request of Counsel 32 brattle.com

34 Offices NORTH AMERICA Cambridge New York San Francisco Washington, DC EUROPE London Madrid Rome 33 brattle.com