Empirical growth models with spatial effects

Abstract: Recent contributions to the regional science literature have considered spatial effects in empirical growth specifications. In the case of spatial dependence, following theoretical arguments from the new economic geography and the endogenous growth models, this phenomenon has been associated to the existence of externalities that cross regional borders. However, despite the general consensus that interactions or externalities are likely to be the major source of spatial dependence, they have been modelled in a rather ad hoc manner in most of existing empirical studies. More surprisingly, the empirical evidence on the preferred spatial specification is mixed and seems to depend on the set of regions, time period and specification. The assumption in this paper is that externalities across regions in long-run growth is mostly a substantive phenomenon caused by technological diffusion and pecuniary externalities, while regional transmission of random shocks should only play a minor role in the process of growth in the long-run. Accordingly, spatial dependence in empirical growth models should be of the substantive type (spatial lag and/or spatial cross-regressive) rather than nuisance error dependence (the so-called spatial error specification). However in a large number of studies, the standard spatial econometrics tools fail to detect the true externality mechanisms, especially when the growth model is underspecified. In contrast to the ad hoc method applied in most of the literature so far, we propose to base the analysis on structural growth models including externalities across economies and to apply the appropriate spatial econometrics tools to test for their presence and to estimate their magnitude in the real world. 1 1