Geographic Propensity Scores for Endogenous Boundaries: An Application to HOLC Redlining

Abstract

Geographic policies often come in two parts: first, geographic boundaries are determined, and second, a treatment is applied based on the boundaries. Separating the effect of the treatment from the boundary drawing process complicates causal inference. A key example of this scenario is Home Owners' Loan Corporation (HOLC) redlining -- cities were divided into relatively homogeneous neighborhoods, and then assigned grades. A growing body of literature has connected these maps to negative long and short run outcomes. I develop tools to make causal assessments that rely on explicitly simulating counterfactual maps. To do so, I build on tools designed to sample from a specified probability distribution of maps. Using the simulated method of moments, I infer the underlying structure of the distribution from which the HOLC redlining maps were drawn. I then construct 1000 counterfactual redlining maps in 13 cities in order to study the short run impacts of HOLC redlining in 1940. I find negative impacts on home values and home ownership rates from $C$ and $D$ grades, and improvements on both dimensions from $A$ grades.