"Chapter 1: The Role of Reputation/Feedback Contents in NYC Airbnb Market: Evidence from Hedonic Price Regressions. Economists have found that reducing information asymmetry is crucial for online marketplaces to overcome market failure due to adverse selection. Reputation/feedback systems and multi-media web contents from sellers are known to be popular disclosure devices for this purpose. This paper employs hedonic price regressions to provide empirical evidence that the recent success of a sharing economy platform, Airbnb also relies on such publicly available information on product quality. Machine learning selectors were employed to reduce high-dimensionality in the attribute space. To process consumer review texts and sellers' advertisement texts, word/phrase extraction and sentiment analysis were introduced. I propose a GMM estimation to produce more accurate implicit price estimates, that was designed to control for time-varying unobservables. 'Superhost' designation by the platform and consumer reviews showed greater impacts than seller side advertisement texts. Chapter 2: Demand Estimation for NYC Airbnb Market: Value of Reputation/Feedback Contents and Voluntary Disclosures. The success of online marketplaces has often been attributed to reputation/feedback systems, in that they reduce adverse selection due to information asymmetry by disclosing enforced or verifiable ex-post information on product quality. This paper tries to quantify the value of such information contents in NYC Airbnb market with a newly constructed dataset containing the actual 708,308 vacation rental reservations from Airbnb tourists. A three level nested logit model was employed to capture consumers' choice set formation behaviors during web search on the platform using Google Maps API. High-dimensional attribute space due to extreme product heterogeneity necessitates variable selection using machine learning methods based on sparsity assumption. Though model selection procedures by LASSO and exact inference for post selection parameter estimates were proposed, structural modeling and endogeneity control turn out to be essential for successful identification. Text processing techniques were introduced to extract variables from sellers' advertisement texts and consumer reviews. The results confirm a key insight from information economics: enforced quality certifications and ex-post verified consumer reviews generate greater welfare impacts than non-verified seller side voluntary disclosures. Chapter 3: Estimation for the Distribution of Random Coefficients with Heterogeneous Agent Types: Monte-Carlo Simulation. This paper is a simple Monte-Carlo extension for Fox, Kim, Ryan, and Bajari (2011), which gives a direct estimator for the distribution of random coefficients in diverse settings including logit demand models. The estimator is a simple inequality constrained least squares, and this study examines its behaviors given there are hundreds of consumer types, which could be an interesting case for various marketplaces. High-dimensional metrics are then introduced to reduce the dimensionality of design matrices the rank of which is the number of consumer types. The approximation performances to the cumulative distribution of random coefficients of such post lasso estimators are compared to those of baseline estimator."--Pages ii-iii.
Read
