In the first chapter I propose a semiparametric estimator that allows for a flexible form of heteroskedasticity for multinomial discrete choice (MDC) models. Despite being semiparametric, the rate of convergence of the smoothed maximum score (SMS) estimator is not affected by the number of alternative choices. I show the strong consistency and asymptotic normality of the proposed estimator. The rate of convergence of the SMS estimator for MDC models can be made arbitrarily close to the inverse of the square root of the sample size, which is the same as the rate of convergence of Horowitz's (1992) SMS estimator for the binary response model. Monte Carlo experiments provide evidence that the proposed estimator has a smaller mean squared error than both the conditional logit estimator and the maximum score estimator when heteroskedasticity exists. I apply the SMS estimator to study the college decisions of high school graduates using a subset of Chilean data from 2011. The estimation results of the SMS estimator differ significantly from the results of the conditional logit estimator, which suggests possible misspecification of parametric models and the usefulness of considering the SMS estimator as an alternative for estimating MDC models. Many MDC applications include potentially endogenous regressors. To allow for endogeneity, in the second chapter I propose a two-stage instrumental variables estimator where the endogenous variable is replaced by a linear estimate, and then the preference parameters in the MDC equation are estimated by the SMS estimator described in the first chapter. In neither stage do I specify the distribution of the error terms, so this two-stage estimation method is semiparametric. This estimator is a generalization of the estimator proposed by Fox (2007). Fox suggests applying the maximum score estimator in the second stage of estimation. This chapter is the first to derive the statistical properties of an estimator allowing for endogeneity in this semiparametric setting. The two-stage instrument variables estimator is consistent when the linear function of instrument variables and other covariates can rank order the choice probabilities. The second chapter also provides results of some Monte Carlo experiments.

We propose a tractable semiparametric estimation method for dynamic discrete choice models. The distribution of additive utility shocks is modeled by location-scale mixtures of extreme value distributions with varying numbers of mixture components. Our approach exploits the analytical tractability of extreme value distributions and the flexibility of the location-scale mixtures. We implement the Bayesian approach to inference using Hamiltonian Monte Carlo and an approximately optimal reversible jump algorithm. For binary dynamic choice model, our approach delivers estimation results that are consistent with the previous literature. We also apply the proposed method to multinomial choice models, for which previous literature does not provide tractable estimation methods in general settings without distributional assumptions on the utility shocks. In our simulation experiments, we show that the standard dynamic logit model can deliver misleading results, especially about counterfactuals, when the shocks are not extreme value distributed. Our semiparametric approach delivers reliable inference in these settings. We develop theoretical results on approximations by location-scale mixtures in an appropriate distance and posterior concentration of the set identified utility parameters and the distribution of shocks in the model.