Generalized least squares pooling averaging: A simulation study

In this paper we propose a generalized least squares (GLS) pooling averaging estimator to address the bias-efficiency trade-off dilemma in estimating heterogeneous linear panel data regression models. By means of a Monte Carlo simulation study, we provide insights in the performance of our GLS pooling averaging estimator compared to the existing ordinary least squares (OLS) pooling averaging in different situations, specifically in the presence of cross-unit error correlations. Our simulation results show that GLS pooling averaging convincingly outperforms OLS pooling averaging in panels with a large number of observations per unit and high error correlations, particularly under moderate noise. We further find that pooling averaging is more robust than individual OLS estimation under endogeneity.