Topics in Time Series Regressions: Pre-Test Estimators and Panel Data Analysis

Chung, Seoung

Graduate Thesis Or Dissertation

Topics in Time Series Regressions: Pre-Test Estimators and Panel Data Analysis Public Deposited

Analytics

Citations

Citeable URL: https://scholar.colorado.edu/concern/graduate_thesis_or_dissertations/h128nd74f

Abstract

This dissertation consists of three essays on time series econometrics, examining some implications of non-stationary data for standard estimation and testing procedures. The first chapter studies the asymptotic distributions of the ordinary least squares (OLS), fixed effects (FE), and first-difference (FD) estimators in the panel cointegration model. We show that all the estimators are asymptotically normal and only OLS estimator is consistent. We use Monte Carlo experiments to examine and compare finite sample properties of OLS, FE, and FD estimators when endogeneity of the regressor and serial correlation in the error terms are included, and evaluate the effectiveness of the autoregressive transformations in panel cointegraion regression. We find that OLS is preferred in terms of bias and size and autoregressive transformation is not a reliable procedure in terms of reducing bias. The second chapter expands on Entorf's results [Journal of Econometrics, 80 (1997) 287-296] on panel spurious regressions. We replicate Entorf's experiments and emphasize the role of drifts and the relative number of time series (T) and cross sectional observations (n) in the estimations and tests. Generally, the spurious regressions phenomenon arises even for large n and small T. The third chapter investigates the properties of the pre-test estimation and hypothesis test after pre-testing for a unit root in an ARMA(1,1) model. We examine the recursive bootstrap and moving-block methods in estimation after pre-test to approximate the distribution of pre-test estimators and compare bootstrap inference with conventional asymptotic inference.

Creator