Essays on Econometric Models for Games and Censored Dat

This dissertation consists of two chapters on nonparametric approach to econometric models.;The first chapter studies identification and inference in game theoretic models with incomplete information and random coefficients. These models allow for strategic interactions in the presence of incomplete information while incorporating payoff parameter heterogeneity across games. The possibility of multiple Bayesian Nash Equilibria presents challenges to identification, and potential discontinuities in the equilibrium selection rule require attention in estimation and inference. In this work, I establish conditions for point identification of the structural parameters, including the joint distribution function of random coefficients, the equilibrium selection mechanism, and the support of random coefficients. I also suggest the Penalized Sieve Minimum Distance (PSMD) estimator assuming piecewise smooth equilibrium selection, derive asymptotic normality for the distributional parameters of random coefficients, and construct a pointwise confidence interval of the parameter based on Chen and Pouzo (2012, 2015). Empirical applications include an entry game between Walmart and Kmart in the context of Jia (2008), and a labor supply game of husbands and wives motivated by Heckman (1978). Both illustrations show that random coefficients capture heterogeneous entry behavior across markets and the heterogeneous work decisions of married couples with young children.;The second chapter provides estimation and inference methods for a nonparametric generalization of Honore (1992)'s classic censored regression model with fixed effects. I consider an unknown nonparametric form of the structural function and first establish identification of this unknown function under the standard assumptions from the previous literature. Next I consider a sieve estimation version of Honore (1992)'s seminal trimmed LAD approach and show the resulting approach yields consistent, asymptotically normal estimates of the structural function. The performance of pointwise confidence intervals for the structural function based on a consistent asymptotic variance estimator and a weighted bootstrap approach are compared in a Monte Carlo simulation. The result verifies a benefit of using the nonparametric estimator when the structural function is nonlinear and the percentage of censored data is modest to moderate. Finally, an empirical application for a simple intertemporal labor supply model examines the potential nonlinear relation of labor supply and hourly wage.