fitting spatial regression to large datasets using unilateral approximations

Maximum likelihood estimation of spatial models typically requires a sizeable computational capacity, even in relatively small samples and becomes unfeasible in very large datasets. The unilateral approximation approach to spatial models estimation (suggested in Besag, 1974) provides a viable alternative to maximum likelihood estimation that reduces substantially computing time and the storage required. Originally proposed for conditionally specified processes, in this 20 paper we extend the method to simultaneous and to general bilateral spatial processes. We prove consistency of the estimators and we study their finite-sample properties via Monte Carlo simulations.