In spatial econometrics, the typical alternative of spatial autocorrelation is expressed in the form of a spatial autorregressive process. While the bulk of the literature is devoted to specification tests and estimation methods for these models, alternatives have been suggested as well. In this paper, we consider alternatives that take the form of the spatial error components formulation proposed by Kelejian and Robinson. We consider a number of specification tests against this alternative, based on both a maximum likelihood framework as well as on a general method of moments estimation approach. We compare the performance of these tests in a series of Monte Carlo simulation experiments against a wide range of alternatives of spatial autocorrelation, under a number of different error distributions.
