Is the European Border Puzzle Dead at Last? A Structural Gravity Approach using Regionalised EU27 Trade Data

Despite the methodological and theoretical advances made in structural gravity modelling, which largely ‘resolved’ the US-Canadian border puzzle, border effect estimates for the EU’s internal borders have remained puzzlingly large. This paper investigates and demonstrates that the reason why these border effect estimates remain so large in comparison to the US-Canadian border effect estimates, is inherently related to the difference in the geographical level at which trade data has been measured and used by EU border effect studies. Using both regional-level trade flows derived from a unique set of regionalised IO tables for the EU27 at the NUTS2 level over the period 2000-2010, as well as self-constructed adjacency and distance matrices at this level of geographical aggregation, this study estimates a set of structural gravity models with importer-time, exporter-time and dyadic fixed effects by resorting to Poisson Pseudo Maximum Likelihood. I document that the border effects are considerably lower as opposed to studies that only use national-level trade flow data. Those studies frequently find border effect estimates around the 10 in magnitude, but regionalised trade data shrinks these estimates to only 2 for trade in goods. This implies that regions trade on average, across the EU, only 2 times more with a region within the same nation as opposed to a similar region, in size (total regional production value) and distance, across the national border. Trade in services is however subject to higher border frictions, albeit it markedly higher in the EU15 than in the EU12. I additionally also investigate the extent to which the Euro has been able to reduce these border effects, but this study finds no statistical evidence that the introduction of the Euro fostered closer economic integration in the short-run. I conclude with a note of caution against studies that endeavour to assess border effects using exclusively national-level data, as estimates are likely overstated in the presence of regional border effects that could arise from geospatial clustering of firms. Unfortunately, this critique does not only apply to EU border effect studies, but it additionally invalidates most non US-Canadian border effect studies as well (id est African and Asian border effect studies).