Asymptotics of improved generalized moments estimators for spatial autoregressive error models

This paper considers linear models with a spatial autoregressive error structure. Extending Arnold and Wied (2010), who develop an improved GMM estimator for the parameters of the disturbance process to reduce the bias of existing estimation approaches, we establish the asymptotic normality of a new weighted version of this improved estimator and derive the efficient weighting matrix. We also show that this efficiently weighted GMM estimator is feasible as long as the regression matrix of the underlying linear model is non-stochastic and illustrate the performance of the new estimator by a Monte Carlo simulation and an application to real data