Spatial filtering, model uncertainty and the speed of income convergence in Europe

In this paper we put forward a Bayesian Model Averaging method aimed at performing
inference under model uncertainty in the presence of potential spatial autocorrelation.
The method uses spatial filtering in order to account for uncertainty in
spatial linkages. Our procedure is applied to a dataset of income per capita growth and
50 potential determinants for 255 NUTS-2 European regions. We show that ignoring
uncertainty in the type of spatial weight matrix can have an important effect on the
estimates of the parameters attached to the model covariates. After integrating out
the uncertainty implied by the choice of regressors and spatial links, human capital
investments and transitional dynamics related to income convergence appear as the
most robust determinants of growth at the regional level in Europe. Our results imply
that a quantitatively important part of the income convergence process in Europe is
influenced by spatially correlated growth spillovers.