Spatial Model Selection and Spatial Knowledge Spillovers: A Regional View of Germany

Knowledge and technological change are often assumed to be the driving forces for long run economic growth. Regions with a higher level of knowledge compared to other regions exhibit a higher per-capita income on average. Agglomeration effects can lead to a steady increase of income and widen the productivity gap between rich and poor regions. It follows that regions are spatially related. From this point of view, it is not surprising that neighbouring region's spillover affects the own economic performance in a positive or negative manner. Recent contributions show that particularly knowledge spillover leads to an enhancing of agglomeration effects: superior regions with respect to per capita income are attractive for company establishment, due to the fact of superior human capital endowment compared to other regions. These core regions play the role of so called knowledge generators because of the fact that they may benefit from well established research density. Densely located underperforming neighbouring regions may benefit from spillover created by core regions to boost their economic performance. Of course, absolute distance towards the core region affects underperforming regions probability of participating in knowledge spillover potential. The implication is, that regions are not only spatially related to each other but they are also spatially heterogeneous with respect to their knowledge potential, which is endogenously influenced by knowledge spillovers. One essential purpose of spatial econometrics is to uncover knowledge spillovers from regional data. Although several empirical contributions have devoted to the identification of spatial dependence in several contexts, the majority of them do not control for spatial heterogeneity in the data. The aim of this paper is to introduce a spatial model selection mechanism for cross sectional data, which controls for both spatial heterogeneity and spatial dependence. Furthermore this mechanism also considers the fact of spatial limited spatial effects. So far existing model selection criteria only tackle the problem of spatial dependence. Using regional data on German NUTS-2 regions, this paper investigates, whether own regional economic performance - measured by per capita productivity - is influenced by neighbouring regions. The selection mechanism which is established in this paper provides evidence of spatially bounded spillover effects, particularly caused by patenting activity of neighbouring regions, if we control for spatial heterogeneity of regions. If one neglects this heterogeneity, results and, eventually conclusions regarding the significance of determining factors of regional economic performance can be severely biased.

Klarl, Torben (2010), Spatial Model Selection and Spatial Knowledge Spillovers: A Regional View of Germany, ZEW Discussion Paper No. 10-005, Mannheim.