Essays on Applied Econometrics

Abstract

In Chapter 1, I propose a two-step instrumental variable approach to identify group pattern with modal regression (modal IV) with endogeneity. The proposed modal IV estimator and its confidence interval are bias-corrected. Simulation studies shown that when the clusters in the data are clear, modal IV regression can successfully identify the heterogeneous group coefficients and the confidence interval has a decent coverage rate. Modal IV regression can possibly apply to many existing literature which allows heterogeneous parameters. Chapter 2 evaluating how policy impacts vary by different context. A national policy can involve various contexts where participant reactions to the same treatment are distributed differently. We propose a modal regression approach to find grouped patterns and how policy effects depend on them. Then, we apply it to evaluating the New Rural Cooperative Medical Scheme, a health insurance program targeting all rural residents in China. Results suggest that: there were three representative rural contexts across China; the health program reduced the out-of-pocket medical spending for the majority of rural residents in only one of the contexts; the contexts where the majority of participants did not financially benefit from the program involves households with higher medical expenditure. Chapter 3 explores a well studied economic question under a new context. We studies the effect of competition on product variety in a novel market: a live streaming platform. We propose a multi-task graph convolutional neural network approach to measure market competition intensity. We leverage a natural experiment, an unexpected account suspension of the most popular streamer, to identify the effect. Our results shows that, market competition increases product variety, and retail prices are sticky and do not adjust to the exogenous competition shock.