UC Berkeley

Electricity is an essential input in the modern economy and a necessity of modern life in the developed world. However, the production of electricity and many of its end uses can be subject to market failures. In this dissertation I study two such market failures related to electricity: (i) power plants exercising market power in response to congestion of the electricity transmission grid; and (ii) electricity consumption for groundwater extraction, which imposes a spatial externality on neighbors. I finally develop new methods for designing experiments and performing power calculations for use with panel data, which is particularly relevant in electricity research that uses high-frequency data on electricity production or consumption.

In the first chapter, I examine how congestion of the electricity transmission grid affects the market in which a power plant competes and the market power exercised by that power plant. Economic theory tells us that market structure is the primary determinant of a firm's ability to exercise market power. However, it is challenging to empirically estimate the causal effect of market structure on market power because a firm rarely experiences exogenous variation in its market's structure. In this paper, I exploit a novel source of exogenous variation in market size within the Texas electricity market --- congestion of electricity transmission lines due to ambient temperature shocks --- to estimate the causal effect of market size on the exercise of market power. When transmission lines congest, this statewide market splits into smaller localized markets. I find that a 10% reduction in market size causes firms to more than double markups. The direction of this effect is consistent with a model of oligopoly competition in which firms set markups in response to residual demand, which is less elastic in a smaller market. My results imply that the markups induced by transmission congestion at high temperatures generate $7.1--21.5 million of deadweight loss annually. These markups also create large transfers --- $2.1 billion per year --- from consumers to producers, which raise important equity concerns.

In the second chapter, coauthored with Fiona Burlig and Louis Preonas, we examine an economically important electricity end use: groundwater extraction in California agriculture. Agricultural and water pumping demand for electricity comprises about 8 percent of California's total electricity usage. Groundwater pumping, a significant share of this usage, creates even more external costs than typical electricity consumption. When one farm pumps groundwater, it lowers the level of the aquifer and increases pumping costs for neighboring farms---who now must pull groundwater a greater vertical distance. We study farmers' electricity usage for groundwater pumping and estimate the extent to which one farm's pumping behavior hurts neighboring farms. We assemble a novel dataset that combines (i) detailed data on farmers' electricity consumption, (ii) rich data from technical audits of these farmers' groundwater pumps, and (iii) publicly available measurements of groundwater depths in California aquifers. Using changes in agricultural electricity prices, we estimate farmers' pumping behavior to be far more price responsive in the medium-term than previously thought. However, we find no evidence that farmers respond to within-day price changes, suggesting limited opportunities for demand response from the agricultural sector. We then calculate the extent to which one farm's water pumping from a shared aquifer raises costs for neighboring farms. Our results suggest that the magnitude of the "pumping cost" spillover effect is fairly small compared to each farm's private cost of pumping groundwater.

In the third chapter, coauthored with Fiona Burlig and Louis Preonas, we answer the question: how should researchers design panel data experiments, particularly when using high-frequency data common in electricity research? We analytically derive the variance of panel estimators, informing power calculations in panel data settings. We generalize Frison and Pocock (1992) to fully arbitrary error structures, thereby extending McKenzie (2012) to allow for non-constant serial correlation. Using Monte Carlo simulations and real-world panel data, we demonstrate that failing to account for arbitrary serial correlation ex ante yields experiments that are incorrectly powered under proper inference. By contrast, our "serial-correlation-robust" power calculations achieve correctly powered experiments in both simulated and real data. We discuss the implications of these results, and introduce a new software package to facilitate proper power calculations in practice.