Essays in Theoretical and Applied Econometrics

Abstract

Despite their wide use in empirical applications, traditional econometric tools may perform poorly in applied work, as the difficulties faced by researchers in applied work are often overlooked through reasonings that depend on restrictive conditions. This dissertation consists of three connected chapters on essential issues in conducting robust estimation and causal inference for key economic parameters under different setups.

The first chapter discusses identification and estimation issues on the treatment effect with anticipation, a generalization of widely used stringent assumptions. Potential outcomes frameworks with assumptions motivated by economic models are provided and bounds for treatment effects are achived. Correpsonding estimation and inference procedures are provided, as well as generalizations to incorporate complicated situations to achieve improvement over current practice.

The second chapter provides estimation and inference procedures robust to high-dimensional covariates in an important class of broadly applied cluster models. Robustness is achieved through either generalization of heteroskedasticity consistent estimators or the leave-one-out procedure.

The third chapter studies a strategic trading model between a market maker who behaves like an ``econometrician'' and uses econometric tools to price and a well-informed inside trader. We focus on the application of econometric tools in estimating unknown parameters in a model that is robust to information ambiguity. Unique linear equilibrium exhibits the underreaction phenomenon. We also show the equivalence between a robust linear strategy and a specific two-way learning procedure regardless of the statistical models chosen by the market maker.