Spatial SUR models: Specification, Testing and Selection

Abstract

Our paper focuses on the case of SUR models with spatial effects. Specifically, the problem that we pose is testing for the presence of spatial effects in these multivariate models, considering that one of our main interests is to deal with the question of selecting the most adequate specification for the data. In order to do this, we obtain several tests that check for the fundamental hypothesis of the model, including the assumption of time stability of the spatial dependence mechanisms. We solve the discussion in a maximum-likelihood framework because this facilitates the obtaining of appropriate tests. The second part of the paper contains a Monte Carlo experiment in order to study the behaviour of the two most popular model selection strategies, Specific-to- General or General-to-Specific, using the collection of tests that have been proposed.