GMM estimation of the autoregressive parameter in a spatial autoregressive error model using regression residuals

Abstract

This paper suggests an improved GMM estimator for the autoregressive parameter of a spatial autoregressive error model by taking into account that unobservable regression disturbances are different from observable regression residuals. Although this difference decreases in large samples, it is important in small samples. Monte Carlo simulations show that the bias can be reduced by 65 − 80% compared to a GMM estimator that neglects the difference between disturbances and residuals. The mean squared error is smaller, too.