Estimation and Inference in High-Dimensional Panel Data Models

点击翻译按钮获取中文摘要

The analysis of many modern economic panel data sets requires to account for both unobserved heterogeneity and high-dimensional data structures. Nevertheless, only few econometric methods have been developed to analyze high-dimensional panel data models with unobserved heterogeneity. The main purpose of the project is to devise novel estimation and inference techniques for such models. We will focus attention on models with interactive fixed effects, which are a very flexible and popular framework to take into account unobserved heterogeneity. A very common way to estimate the unknown parameters in a low-dimensional panel model with interactive fixed effects is the so-called common correlated effects (CCE) estimator introduced by Pesaran in 2006. However, this very popular estimator breaks down in high dimensions, and naive extensions to the high-dimensional case fail dramatically. In the project, we will develop a novel CCE-type estimator which does work in high dimensions. The theoretical part of the project will be concerned with deriving the asymptotic properties of the proposed estimator. In a first step, we will concentrate on estimation theory and derive the convergence rate of the estimator. In a second step, we will turn to distribution theory and analyze inferential procedures based on it. The methodological and theoretical analysis of the project will be complemented by simulation studies and empirical applications.