Geography in Reduced Form

Geography models have introduced and estimated a set of competing explanations for the persistent relationships between firm and location characteristics, but cannot identify these forces. I introduce a solution method for models in arbitrary geographies that generates reduced-form predictions and tests to identify forces acting through geographic linkages. This theoretical approach creates a new strategy for spatial empirics. Using the correct observables, the model shows that geographic forces can be taken into account without being directly estimated; establishment and employment density emerge as sufficient statistics for all geographic forces. I present two applications. First, I show how the model stands as an alternative to the monocentric city framework, generating gradients relating density to firm and location qualities that are analogues to the distance gradients of the monocentric model. These new gradients are significantly more powerful than their traditional analogues, and because they impose no structure on the city, they are especially useful for cities with multiple centers. Second, I show how the model generates reduced-form tests that distinguish between spillovers and firm sorting and find evidence of both. ∗Dartmouth College and Michigan State University Contact: [email protected].