Essays on Markets with Frictions

In this dissertation, I explore the implications of various forms of frictions on market outcomes.Specifically, I look at search frictions in two-sided markets, geographic frictions in a healthcare market, and the use of a machine learning approach in the presence of regulatory frictions. In the first chapter, I leverage the entry of a high-speed train system in South Korea as a natural experiment to establish the causal effect of competition between hospitals on health care quality and consumer welfare. Using a difference-in-differences estimator, we examine the effects of competition on hospitals depending on their proximity to train stations, notably how increased competition impacts health outcomes as measured by 30-day mortality rates. Our results suggest that increased competition leads to an improvement in the quality of clinical care. To evaluate the overall impact of the HST on patient welfare, we estimate a structural model of hospital choice, allowing for a flexible formation of patients’ consideration sets. We find that patients living near a HST station experience an improvement in welfare arising from the reduction in travel time as well as improvements in hospital quality. Patients living further away from HST stations also experience an improvement in welfare although they do not gain from the reduced travel time due to the improvement in the quality of treated hospitals. We also find that the HST can have a beneficial impact on patient health by facilitating patients’ sorting to better hospitals, even while holding quality of clinical care constant. In the second chapter, I study the impact of search frictions and preferences on the formation of a match in two-sided markets. Since agents on both sides have private preferences regarding each others’ characteristics, forming a match based on mutual compatibility requires extensive costly search. To better understand the relative impact of search frictions and preferences on match outcomes, we use data from a field experiment conducted on an online dating platform wherein randomly selected users are given the ability to know upfront a piece of information about the private preference of the opposite gender (information which otherwise should have been searched for). We find descriptive evidence suggesting that reducing search frictions through the provision of information may lead to less sorting between matched couples in terms of various characteristics such as race and education level. To investigate the relative contribution of search frictions and preferences on assortative matching, we develop and estimate a model that incorporates both costly search and preference heterogeneity across users. Identification of our model relies on the variation in information caused by the experiment as well as the exclusion restriction to separately identify preferences from costs. Our estimation results reveal that frictions play a significant role in shaping matching outcomes. Using model estimates, we simulate matches under various environments, including the Gale-Shapley protocol. We find that removing frictions leads to significantly less sorting between couples. We also find that frictions in our platform lead to significant departure from efficiency. These results highlight the importance of platform designs that aim to reduce search frictions. In addition, with one-third of the marriages in the U.S. beginning online, this paper shows how the design of an online platform can contribute to diversity, which can in turn alleviate persistent social inequality. In the third chapter, I study how we can use machine learning methods to overcome challenges faced by firms in the presence of restrictive privacy regulations. The ever-increasing volume of consumer data provide unprecedented opportunities for firms to predict consumer behavior, target customers, and provide customized service. Recent trends of more restrictive privacy regulations worldwide, however, present great challenges for firms whose business activities rely on consumer data. We address these challenges by applying the recently developed federated learning approach - a privacy-preserving machine learning approach that uses a parallelized learning algorithm to train a model locally on each individual user’s device. We apply this approach to data from an online retailer and train a Gated Recurrent Unit recurrent neural network to predict each consumer’s click-stream. We show the firm can predict each consumer’s activities with a high level of accuracy without the need to store, access, or analyze consumer data in a centralized location, thereby protecting their sensitive information