Finite sample properties of alternative GMM estimators for random effects models with spatially correlated errors

摘要

For panel data models with error components that are spatially correlated, the finite sample properties of alternative generalized method of moments (GMM) estimators are determined. We suggest using a continuously updated GMM estimator which is invariant to curvature altering transformations and which should improve small sample efficiency. A Monte Carlo study using a wide range of settings compares the small sample efficiency of various GMM approaches and the maximum likelihood estimator (MLE). The GMM estimators turn out to perform comparably to the MLE approach and even outperform the latter for complex weighting matrices and non-normally distributed errors.