Introduction

In panel data models it is often of interest to estimate the average long-run effects of some exogenous variables (x) on a dependent variable of interest (y). In situations where T (the number of time periods) is sufficiently large there are four procedures that can be used to estimate this average effect (Pesaran and Smith (1995)). The first involves estimating separate regressions for each group and averaging the long-run coefficients over groups, which Pesaran and Smith refer to as the mean group estimator (MGE). The second procedure is to pool the separate regressions by imposing common slopes (but allowing for fixed or random intercepts), with the long-run effects estimated using standard fixed or random effects pooled regressions. The third is to average the data over groups and estimate aggregate time series regressions based on group averages. The last is to average the data over time and estimate cross-section regression based on long-time averages. In the static case, where the regressors are strictly exogenous and the coefficients differ randomly and are distributed independently of the regressors across groups, all four procedures provide consistent (and unbiased) estimates of the average (long-run) effects (Zellner (1969)). For some time it was wrongly believed that a similar result held for dynamic panel data models, namely that all the above four procedures yield consistent estimators.