Essays on nonlinear panel data models and conditional quantiles

This dissertation is comprised of three essays based on nonlinear panel data models on conditional quantiles. The second chapter considers the problem of efficient estimation of a partially linear dynamic panel data model with fixed effects. Using the series method, we develop the generalized method-of-moments (GMM) estimator for the linear component. Asymptotic normality for the estimator of the parametric portion and consistency for the nonparametric component are established. Monte Carlo experiments are implemented to investigate the finite sample of the GMM estimator. As an extension, we relax the assumption of partial linearity so that the model becomes a nonparametric additive model in dynamic and exogenous components. We obtain the estimators of the nonparametric functions and derive the asymptotic results and bias correction. Finally, to illustrate the performance of the proposed estimators in a real data setting, we apply them to unemployment dynamics at the U.S. state level.;Chapter 3 extends a linear stochastic production frontier model with time-varying individual effects to a nonparametric model in which the functional form of the production frontier is unspecified. We derive the kernel estimator for such a frontier in fixed effects framework and implement Monte Carlo simulations to investigate finite sample performances of our estimator. Lastly, we apply the estimator proposed in this chapter to estimate the production function and time-varying technical efficiency of private manufacturing establishments in Egypt over the period 1988 to 1996.;In chapter 4, we explore a two-step conditional quantile estimation of predictions which have substantially lower bias compared to the usual nonparametric conditional quantile estimator. The two-step conditional quantile estimation is proposed in a multivariate setting. We apply the two-step estimation method to predict the tail conditional quantiles of daily stock returns from Europe. Our results indicate that substantial gains in the tail quantile prediction are obtained with respect to different loss functions over alternative multivariate and univariate quantile estimators.