Efficient Estimation of Spatial Autoregressive Models

This paper considers estimating general spatial autoregressive models using the generalized method of moments (GMM). I propose nonparametric estimates of the optimal instruments based on conditional second moment restrictions. I show that these instruments are optimal over all possible instruments, especially over those usually suggested in a spatial context. I provide a nonparametric estimator of sample autocovariances function for irregularly spaced spatial processes, and I show that this estimator converges in probability. I then derive the consistency in norm L_2 of the resulting asymptotic variance matrix estimator. Finally, the asymptotic distribution of the GMM estimator is stated.