Spatial scale of agglomeration and dispersion: Theoretical foundations and empirical implications

This paper revisits a wide variety of existing economic geography models in a many-region setup. It investigates the spatial scale of agglomeration and dispersion intrinsic to each model. In our unified analytical framework, these models reduce to two canonical classes: one with a global dispersion force and the other with a local dispersion force. Their formal distinction is that the former is dependent, whereas the latter is independent of the distance structure of the model. These classes exhibit two stark differences. The first difference concerns their response to transport costs: Global and local dispersion forces are triggered by higher and lower transport costs, respectively. Consequently, in a realistic model with both types of dispersion forces, a decrease in transport costs simultaneously causes both agglomeration at the global scale and dispersion at the local scale. The second difference concerns the agglomeration pattern: multiple agglomerations emerge and spread globally over the regions in the former, whereas agglomeration always takes the form of a unimodal regional distribution of mobile agents in the latter. Endogenous agglomeration mechanisms generally do not isolate the locations at which agglomerations grow or decline for a given change in transport costs. However, they offer predictions for the global spatial distribution of agglomerations as well as the local spatial extent of an individual agglomeration. This knowledge provides a consistent explanation for the set of seemingly unrelated empirical results from reduced-form regressions on regional agglomerations (e.g., Baum-Snow, 2007; Baum-Snow, Brandt, Henderson, Turner and Zhang, 2017; Duranton and Turner, 2012; Faber, 2014); it provides a new set of testable hypotheses. Moreover, our analytical framework provides formal predictions of treatment effects in the structural model-based approaches for regional agglomeration. Applications to the most standard formulations (e.g., Allen and Arkolakis, 2014; Redding and Sturm, 2008) are ...