Essays on Retail Electricity Markets and Methods for Analyzing Matched Data

Abstract

My dissertation consists of two chapters in energy economics and two in econometrics. Chapter 1 studies the relationship between retail electricity prices and household income in Connecticut. Using novel data sets constructed from files published in the Connecticut regulatory authority’s online docket system, I show that customers living in the state’s poorest zip codes paid significantly more for competitive retail electric service than customers living in above median income zip codes between 2011 and 2018. These differences may be explained by the fact that low-income customers purchased from more expensive retailers, which, coupled with preferences for value-add products such as renewable energy, led them to choose contracts with higher prices. Chapter 2 examines the role of regulatory policy in deregulated retail electricity markets in the United States using hand-collected data on whether and when a market adopted three pro-competitive policies. While all three policies increased the share of customers choosing to purchase competitive retail electric service, the estimated effect size of policies differ. The policies also had different effects on competitive retailers’ entry decisions and prices, such that only one policy plausibly increased the consumer welfare of residential customers. Chapter 3, which is co-authored with Bo Honoré and Adriana Lleras-Muney, proposes methods for analyzing merged data sets where observations are linked to multiple possible matches. We show that it is straightforward to modify standard econometric models to accommodate such data, and that, when observations are weighted according to the number of matches, the resulting estimator performs well under a variety of data generating processes. Chapter 4 evaluates the combined performance of methods for merging data sets and estimating linear regression models when the variables of interest are recorded in separate data files. I study four matching methods, that either do or do not allow for multiple matches per observation, and that are either deterministic or probabilistic in nature. Simulation results demonstrate that allowing for multiple matches per observation can produce more precise estimates, provided the appropriate estimation method is used, and that the choice between using a deterministic or probabilistic method in such cases is less important.