Highly Disaggregated Land Unavailability

Abstract

We combine high-resolution satellite imagery with modern machine-learning methods to construct novel datasets that capture the geographic determinants of U.S. housing supply. This Land Unavailability (LU) measure is a markedly more accurate house price predictor than the popular proxy of Saiz (2010). We apply LU to fundamental housing finance problems to provide substantially more precise housing wealth elasticity estimates, new evidence on the relationship between house prices and entrepreneurship during COVID-19, and novel empirical tests of the supply-side speculation theory. Finally, our LU instrument is broadly uncorrelated with housing demand proxies, supporting its use as an instrument for house prices.

Keywords: Land Unavailability, Buildable Land, Real Estate, Housing Market

JEL Classification: R30, R31, R20