Essays on Regulations in Peer-To-Peer Markets

Abstract

Peer-to-Peer (P2P) markets have shown tremendous growth in the post-recession era, and three reasons contribute to their success. First, the economic surplus is generated for both buyers and sellers from the trade of a large selection of traditional products and services. Second, there exists easy instant access to the market through modern devices and internet technologies. Third, these markets self-correct failures by removing information asymmetry through reputation systems. Despite increasing efficiency in the markets of traditional goods and services, many P2P businesses and their markets face scourge from lawmakers because of two reasons. First, the P2P markets enable many users to operate where their operation conflicts with the existing regulatory regime. On the other hand, lobbying efforts by the incumbents in the industry might influence the regulators. Chapter 1 examines how a Peer-to-Peer market such as Airbnb - a short-term rental marketplace - responds to supply restrictions prompted by a series of regulatory threats and incumbent pressure. The growth of online P2P marketplaces, their potential to influence traditional industries, and their operational dissent with outdated (yet existing) laws have made these markets a subject of substantive policy debate. In anticipation of regulation, Airbnb had purged thousands of rentals from its New York City market over the past few years. I use unique daily data of over 100 thousand rentals that appeared on the Airbnb website between September 2014 and March 2017. I identify changes in price and quantity of accommodation due to the purge by exploiting the patterns of rental loss from the platform observed in the data. Next, I estimate the residual demand faced by each rental, and under assumed monopolistic competition market structure, I recover the marginal costs. These estimates are used to compute the welfare of producers (hosts/rental owners) and consumers (guest/travelers), which I find out to be 392M USD and 566M USD respectively in a 31 month period. Using a counterfactual setting where purge was not carried out, I recompute the welfare and find that rental owners lost nearly 26M USD while consumers lost 25M USD as a result of the purge. Chapter 2 revisits the residual demand estimation problem in the presence of high dimensional confounding demand signals. I refrain from subjectively selecting the variables entering the model and automate this process by using Machine Learning (ML) techniques. I apply the Double Machine Learning (DML) approach proposed by Chernozhukov et al. (2018) and Chernozhukov et al. (2017), to a massive data of Airbnb rentals in the New York City (NYC) and estimate the demand facing each rental. The paper exploits the intrinsically high-dimensional text data from rental amenities, descriptions, and reviews, which confound the true demand relationship producing biased estimates. In converting text data into meaningful features and constructing a computationally manageable data, I make several design choices. The findings in this paper show a huge improvement on the previous estimates of demand, which completely ignored the time-varying reviews as controls. The demand estimate jumped to -14.364 with the standard error of 1.283, when we included a host of features constructed from reviews. The second stage results are stable across different choices of Penalized Regression methods. The findings also indicate that a good model fit in the first stage is crucial to the validity of the final estimates. Finally, the chapter shows how biases from ignoring high dimensional features in estimating residual demand, transmit themselves to the supply side and in the welfare calculation.