Spatial correlations exist for many economic phenomena. We also know such interactions are typically weaker across country borders than within countries, due to institutional, infrastructural or cultural factors. However, in spatial econometric analyses, the effect of borders is rarely taken into account, and all borders between regions are treated as equal. We distinghuish between different types of borders by splitting our weight matrix in two, and creating separate lags for within-country and cross-border effects.
We demonstrate the effectiveness of this method in an analysis of average productivity in a region for several manufacturing sectors. Without implying causality, we relate this productivity to average productivity in
surrounding regions as well as to agglomeration externalities both in its own region and in surrounding regions. We find border effects are indeed quite present: spatial lags within the country are sometimes statistically significant, those across borders never. However, the only agglomeration effect that is statistically significant across borders is unrelated variety.
Weight matrix
We split the weight matrix to make separate spatial weight matrices for within-country effects and between-country effects. We can then run the regression either with the normal, full, spatial effects, or with the two effects separately. Code is available upon request.
Keywords agglomeration externalities; spatial econometrics; border effects
Status Published in Letters in Spatial and Resource Sciences in 2017.