Spatial Econometric Models: The Pursuit of an Accurate Spatial Structure with an Application to Labor Market

In most applications of spatial econometrics, spatial prior information is used to distinguish close units (interacting, spatially dependent) from distant units (mutually independent). However, the true spatial setup is not actually known in most cases. Fortunately, even imperfect yet mostly valid spatial structures may lead to reasonably accurate model estimates. Still, the general validity of any used neighborhood structure should be consistently assessed as deviations from true setups cannot be easily avoided. Such assessments are typically based on statistical measures, which in turn facilitate and motivate the quest of improving spatial structures used for estimation. Enhanced spatial information may lead to higher efficiency of marginal effect estimation and better prediction accuracy. This article presents a heuristic algorithm that uses sectoral macroeconomic to generate enhanced neighborhood definitions. Suitable methods for statistical inference and efficiency verification are also provided. The proposed method is tailored for short panels and maximum likelihood estimators, yet its principles are generally applicable. An empirical demonstration of the approach is provided, based on NUTS2-level data for 10 contiguous EU member states.