Spatial model selection strategies in a SUR framework. The case of regional productivity in EU

The purpose of the paper was to compare two well-known model selection strategies, the so-called Specific-to-General, Stge, and General-to-Specific, Gets, in a context of spatial SUR models. The two strategies use a battery of misspecification tests obtained in a maximum likelihood framework. The robust tests to local misspecification errors in the alternative hypothesis and the common factor test have been developed with this purpose. The paper includes a Monte Carlo experiment to compare their performance in a situation of small sample sizes. The results are mixed: Both alternatives work well under ideal conditions, but their efficiency deteriorates for different departures such as non-normality or endogeneity. All in all, Stge appears to be slightly preferable although our impression is that the two are complementary and can be used in common. The paper finishes with an application to the case of productivity for a large set of European regions.