Abstract:
This thesis investigates the estimation of dynamic panel data models with fixed effects. We consider models with incidental intercepts, incidental intercepts and trends, and incidental intercepts and strictly exogenous regressors. We place particular emphasis on the Hsiao, Pesaran and Tahmiscioglu (2002) maximum likelihood approach and the Han and Phillips (2006) estimator; both of these have the important property of consistency even when the autoregressive coefficient is unity. Although econometric theory, and Monte Carlo simulations are presented our focus is on cross-country empirical applications. In chapter three, we estimate the price and income elasticities of oil consumption in three sectors (Transportation, Industrial and Other) using panels containing 73 countries and spanning 1972-2000. We find evidence of high autoregressive coefficients, which have not been the subject of much research. In chapter four we apply the same specifications and estimators to demand for green coffee beans in panels of up to 40 countries. To the best of our knowledge, dynamic panel data models have not previously been applied to coffee demand. In our results the autoregressive coefficients are lower than for oil. However, in both sets of results, we find that there is a substantial discrepancy between the results of the Arellano and Bond (1991) generalised method of moments estimator and the Hsiao et al (2002) estimator. Chapter five considers panel unit root testing. We compare the empirical performance of a test based on the Han and Phillips (2006) estimator with the popular Im, Pesaran and Shin (2003) test procedure for panels of macroeconomic variables including many form the Penn World Tables and real exchange rate series.
