An Experimental Study of Monthly Electricity Demand (In)elasticity.

• Date: 01 March 2021
• Author: Nauze, Andrea La

1. INTRODUCTION

   The price elasticity of demand for electricity is a key parameter for analyzing the costs of climate change mitigation, the incidence of carbon pricing, market power, and electricity market design. This paper reports results from a field experiment designed to estimate this elasticity at the monthly level. In our experiment, we partnered with an electricity retailer to randomly provide households discounts of up to 95% off of their per-unit electricity price for two months. Combining our experiment with billing, smart meter, and survey data, we find that residential electricity demand is unresponsive to large reductions in both marginal and average prices. Our preferred own-price elasticity of demand estimate is -0.003. The estimate is statistically insignificant, but economically important as it points to perfectly inelastic demand. With 95% confidence, we can rule out households in our sample having a price elasticity more negative than -0.04.

   In some ways our results are surprising. The non-experimental literature documents that households are relatively inelastic at this time horizon, but not as inelastic as we find them to be. Many of these studies struggle to control for endogeneously-set prices, endogeneous lagged consumption, and measurement error in prices (Alberini and Filippini, 2011). Some key exceptions are Ito (2014), who uses variation in relative prices along different increasing block rates across two adjacent utilities to find a monthly demand elasticity of -0.05 with respect to average price, and Deryugina et al. (2020) who use matching estimators to estimate one, three, and ten year elasticities of -0.16, -0.27, and -0.30 to -0.35, respectively, when prices decrease due to utility reform. The rich experimental literature avoids these identification issues, but has to-date focused on the implementation of hourly changes in prices, like time-of-use, critical peak pricing, or real-time pricing interventions.

   In this paper we describe a unique experiment with an electricity retailer partner who was willing to randomize prices for two months and let us publish the results. We provided different levels of discounts, some very large, which allows us to document highly inelastic demand at the monthly level across the entire range of the demand curve. We believe that the most likely explanation for our findings is that an intermediate, month-long time horizon is both too long to allow for significant inter-temporal smoothing of electricity use, and too short to change habits or household appliance stock. Over this time horizon, adjustment costs in behavior and technology can rationally justify non-response.

   We exploit the richness of our dataset to address several potential alternative explanations for these findings. Working with our partner, we ensured that our discounts were credible. All correspondence with households in the experiment was sent directly by the retailer via email, in the same format as all previous on-going correspondence with the retailer. All other retailers in the country similarly use email for corresponding with customers. Furthermore, our experimental sample is constructed from a set of households who showed willingness to engage with the retailer by replying to a baseline survey before the experiment started.

   We also provide evidence that non-response is not due to households being unaware of the discounts. We obtained server logs that confirm that relevant emails were indeed sent to all of the treated households and not to control households. Using an email tracking platform, we confirmed that the vast majority of treated households opened the treatment emails. Restricting our sample to households we are sure opened treatment emails, we still find next-to-no demand response to our experimental price shocks.

   Our price discounts were promised and applied for a month, and then refreshed at different rates for a second consecutive month. Households were told that they would receive the discounts on their first quarterly bill after the experimental period was over. Some households received a non-discounted bill during the experimental period. To address concerns that our results could be driven by household confusion, we check and confirm that all of our findings are robust to restricting our sample to experimental household-days before any intermediate bills were received.

   Households were notified about electricity bill savings-to-date after the first month of the experiment. We confirm that households did not change their behavior directly after that communication. We also confirm that they did not change electricity usage levels after the discount was applied to their post-experiment bill.

   To what extent is the responsiveness of our customers representative of Victorian customers? On one hand, our partner retailer is a mid-sized retailer that entered the market during retail deregulation. Customers of our retailer are more likely to be price-aware than the two-thirds of Victorian customers still with the large incumbent firms from the pre-deregulation era. That said, we restrict our sample to customers who are not on pay-on-time discounts, who could be less price-responsive. But it is unclear as to how pay-on-time discounts translate to price sensitivity: the discounted price looks attractive, but the Australian Competition and Consumer Commission (ACCC) documents that households on these plans end up paying the much higher non-discounted prices 44% of the time (ACCC, 2018). Our experimental sample pays rates that are representative of actual prices paid by the retailer's average customer and slightly lower rates than the average customer in the state (Table 2).

   Did prices disproportionately lack salience for the households that we study? This would have been the case if our households were disproportionately enrolled in automatic payment plans, and hence less likely to pay attention to their bills (Sexton, 2015). However, only 35% percent of the households in our sample have automatic payment plans. When we restrict our sample to households without automatic payment, we still find no evidence of a demand response to our experimental price shocks.

   Households' appliance stocks are also unlikely to explain our findings as most households have appliances that allow for varied electricity use. Specifically, 87% of households have air conditioning, and many have electric heating (47%), clothes dryers (27%), and/or electric hot water (16%). We find statistical-zero treatment effects when we restrict our sample to these households that we expected to be relatively more price elastic.

   Our results also cannot be explained by household absenteeism: over 92% of the household-days in our sample involve levels of use that exceed 4 kWh/day. We further find no evidence of a demand response when we restrict our analysis to hours of the day when we can infer from smart meter data that someone is likely to be home.

   One possible explanation for the observed lack of response is experimenter-goodwill. If households are thankful to receive discounts they could be reluctant to increase electricity use in an experimental setting. We discount this explanation for a few reasons. First, the discounts were framed as a recognition for having answered a baseline survey, so they were earned, not given. Second, all emails were sent directly from the retailer, not the University. Finally, even if a goodwill effect did exist, such an effect would have had to grow exactly in proportion to discount size to explain our results.

   It is possible that our result differs from findings in the literature due to asymmetry in price responsiveness. With the notable recent exception of Deryugina et al. (2020), many prior studies describe how consumers react to increases rather than decreases in electricity prices. We unfortunately did not design the experiment in a way that would allow us to confirm this hypothesis.

   Related literature

   Our findings contribute most directly to the literature of residential electricity demand that describes how households respond to monthly changes in prices. This large literature typically uses variation in prices induced by geographic and increasing block discontinuities (Ito, 2014), changes in regulatory policies (Deryugina et al., 2020), electricity crises (Bushnell and Mansur, 2005; Reiss and White, 2008; Costa and Gerard, 2015; Alberini et al., 2019), and some combination of the above with structural models (Reiss and White, 2005), and panel and time-series methods (Alberini and Filippini, 2011; Cuddington and Dagher, 2015; Ros, 2017; Burke et al., 2018). Most of this literature relies on data that is aggregated across households, for instance at the state or municipal-level.

   Our paper differs from the recent body of experimental evidence primarily in our choice of time horizon. We consider prices shocks that are much longer in duration than is common in the experimental literature. Experimental studies typically examine the effects of price variation or shocks that last for several hours, with a focus on prices that are set ahead of time (time-of-use) as well as a few examples that fluctuate with market conditions (critical-peak-pricing or real-time-pricing). See, for example, Fowlie et al. (2017), Wolak (2011), Allcott (2011a), and Ito et al. (2018) as well as experiments in Australia summarized in Arup et al. (2014). Faruqui and Sergici (2010) and Faruqui and Sergici (2011) provide a survey of this literature that includes experiments involving private companies and utilities.

   While many randomized control trials find responses at the hourly-level, recent experiments find that households can also be unresponsive to changes in price. Although Fowlie et al. (2017) finds price elasticities of -0.075 and -0.31 to critical peak pricing and time-of-use prices, respectively, among opt-in participants, the authors find much smaller effects among the much...