Three Essays on Nonparametric and Semiparametric Methods and Their Applications

Abstract

This dissertation contains three essays on nonparametric and semiparametric regression methods.

In the first essay, we consider the problem of nonparametric regression with mixed discrete and continuous covariates using the k-nearest neighbor (k-nn) method. We derive the asymptotic normality of the proposed estimator and use Monte Carlo simulations to demonstrate its finite sample performance. We apply the method to estimate corn yields in Iowa as a function of agricultural district, temperature, and precipitation.

In the second essay, we consider the problem of testing error serial correlation in fixed effects panel data models in a nonparametric framework. We show that our test statistic has a standard normal distribution under the null hypothesis of zero serial correlation. The test statistic diverges to infinity at the rate of √N under the alternative hypothesis that errors are serially correlated, where N is the cross-sectional sample size. We propose a bootstrap version of the test which we show to perform well in finite sample applications.

In the third essay, we consider estimation of varying-coefficient single-index models with an endogenous regressor. We propose a multi-step instrumental variables procedure to estimate the coefficient function and the corresponding index parameters. We prove the consistency of the estimators, and we present Monte Carlo simulations demonstrating their finite sample performance. We then apply the proposed method to examine the determinants of aggregate illiquidity in the U.S. stock market.