Fitting spatial econometric models through the unilateral approximation

Maximum likelihood estimation of spatial models based on weight matrices typically requires a sizeable computational capacity, even in relatively small samples. The unilateral approximation approach to spatial models estimation has been suggested in Besag (1974) as a viable alternative to MLE for conditionally specified processes. In this paper we revisit the method, extend it to simultaneous spatial processes and study the finite-sample properties of the resulting estimators by means of Monte Carlo simulations, using several Conditional Autoregressive Models. According to the results, the performance of the unilateral estimators is very good, both in terms of statistical properties (accuracy and precision) and in terms of computing time.