UC Berkeley

The coming century will bring numerous environmental challenges and understanding the strategic decisions involved in energy production and consumption will be central to addressing them effectively. In this dissertation, I use methods from applied econometrics, behavioral economics, and industrial organization to investigate various lines of inquiry around this broader motivation. In Chapter 1, I study how residential electricity consumers respond to increasingly complicated incentives that are meant to improve allocative efficiency and test whether their behavior is consistent with standard models. In Chapter 2, I estimate the impact of temperature on high school students' standardized test performance in order to understand how environmental factors affect educational outcomes. In Chapter 3, I evaluate a targeting strategy meant to improve the efficiency of an electricity pricing program and develop a theoretical framework to ground the findings.

The first chapter studies whether consumers are attentive to time-varying incentives to reduce electricity consumption. Dynamic pricing models typically assume that consumers respond to marginal incentives. I use a field experiment to assess the impact of dynamic pricing on residential electricity consumption and find strong evidence of inattention. I propose a model to interpret the results which suggests that the benefits of dynamic pricing may be substantively undermined by inattention. I also explore the role of automation in dynamic pricing, which holds the promise of reducing the cognitive choice frictions that cause inattention and lowering the effort cost of responding to price changes. I report three primary findings. First, households---both with and without automation---significantly respond to a short term price increase by reducing consumption. Second, responses are very insensitive to the size of the price change. A price increase of 31 percent causes consumption to fall by 12 percent on average, whereas a price increase of 1,875 percent causes an average reduction of 14 percent. Third, automation causes responses that are more than three times larger than the average effect, but are still insensitive to the price level. The results suggest that households use simplifying heuristics when facing dynamic prices and that automation reduces effort costs, but does not resolve inattention. I apply the model to recover bounds on the price elasticity of demand and shed light on the potential attention costs of dynamic pricing.

The second chapter, coauthored with Maximilian Auffhammer and Catherine Wolfram, studies the impacts of extreme temperature on over 5 million students standardized test performance. We exploit plausibly exogenous year-to-year within-school daily weather variation in order to measure the contemporaneous effect of maximum outdoor temperature on aggregate student performance. The exam studied is the California High School Exit Exam, a state-wide standardized test that evaluates high school students' mathematics and English-language arts aptitude and was a requirement for receiving a diploma from 2006-2015. We document a nonlinear relationship between temperature and performance. Temperatures above 27.5$^\circ$C show statistically significant negative impact on pass rates in both subjects and scores in the math assessment. We also document heterogeneity in the effect by income in the area surrounding the school and find more pronounced effects for schools in the lowest income quartile.

The third chapter, coauthored with Maximilian Balandat and Datong Zhou, evaluates the effect of targeting based on heterogeneous treatment effects using an experiment. We provide a theoretical framework for how various factors undermining external validity affect targeting and the how experimental evaluation of targeting can be used to parse competing mechanisms. Our theoretical framework distinguishes between group-level heterogeneity as defined by covariates and subject-level effects we call individual treatment effects (ITEs). ITEs can only be gleaned through observing program participation using panel data, but capture additional effect heterogeneity within the group-level effects. We partnered with a energy technology company in order to examine the impact using ITEs to target in the field. We find our targeting strategy reduces the costs of the partner by 52 percent and the results are highly significant. The strategy also reduces revenue by 24 percent, indicating an overall increase in profit on the order of 28 percent. We also examine the persistence of the effects and find the cost savings begin to diminish only 60 days after deployment of the targeting strategy. These findings suggest significant potential for reducing the cost of the program, but only in the short-term. Importantly, the experimental evaluation allows us to understand its performance without having to rely on the common practice of conducting ex-post simulations.