Three Essays on Dynamic Panel Data Estimation

Abstract

This dissertation consists of three essays, first two of which consider a new estimation method of dynamic panel data models and the last one considers an application of these models. The first essay (Chapter 1) offers empirical likelihood (EL) estimation of dynamic panel data models, which provide great flexibility to empirical researchers. EL estimation method is shown to have great advantages in usual settings, however little is known on the relative merits of these estimators in panel data models. With this essay, we try to fill that gap by establishing the asymptotic properties of the EL estimator for a dynamic panel model with individual effects when both the time and the cross-section dimensions tend to infinity. We give the conditions under which this estimator is consistent and asymptotically normal. In the second essay (Chapter 2), via a Monte Carlo study, we assess the relative finite sample performances of EL, generalized method of moments, and limited information maximum likelihood estimators for an autoregressive panel data model when there are many moment conditions. We also extend our results to the many weak moments settings. Our results suggest that when the overall performances are concerned, in terms of median, interquartile range and median absolute error of the estimators, in both strong and weak moments settings, EL is more reliable. In the final essay (Chapter 3) we consider an application of dynamic panel data models to examine the determinants of the allocation of state highway funds using panel data for North Carolina's 100 counties for the years 1990 to 2005. We make two main contributions with this essay. First, although there have been numerous studies of highway funding at the state level, to our knowledge, there is no analysis at the sub-state or county levels. Second, by using dynamic panel data models and sophisticated methods to estimate them, we account for any potential persistence in the process of adjustment toward an equilibrium, besides, unlike most of the previous studies, we control for the unobserved county heterogeneity and time effects that explain spatial differences, which may cause omitted variable problem if ignored.