Testing for spatial autocorrelation in a fixed effects panel data model

The aim of this paper is to assess the relevance of spatial autocorrelation in a fixed effects panel data model and in the affirmative, to identify the most appropriate spatial specification as this appears to be a crucial point from the modeling perspective of interactive heterogeneity. Several LM test statistics as well as their LR counterparts, which allow discriminating between endogenous spatial lag versus spatially autocorrelated errors, are therefore proposed. Monte Carlo experiments show their good finite sample performance. Finally, an empirical application is provided in the framework of the well-known Feldstein-Horioka puzzle.