Highly Disaggregated Land Unavailability

Abstract

We use new large-scale data techniques and comprehensive high resolution satellite imagery for the contiguous United States to construct novel datasets that capture the geographic determinants of house prices and housing supply: The percentage of undevelopable land – Land Unavailability – and its complement, buildable land. Our Land Unavailability measure expands on the popular proxy from Saiz (2010) by:

(1) using higher resolution satellite imagery from the USGS;

(2) more accurate geographic boundaries; and

(3) multiple levels of disaggregation.

Using highly disaggregated data we show that Land Unavailability is uncorrelated with housing demand proxies, validating Land Unavailability as an instrument for house prices; that the geographic components of Land Unavailability, especially in combination with modern machine learning techniques, provide substantial incremental predictive power for house prices; and previous studies that employed limited land unavailability datasets underestimated the impact of house prices on unemployment during the Great Recession by 30%. With our buildable land dataset we then test the supply side speculation theory that aims to explain the previously puzzling large house price growth in traditionally elastic housing markets. In line with theory, results document that housing markets with intermediate amounts of buildable land, those that are elastic in the short run but plausibly inelastic in the long run, experienced abnormally large house price growth during the 2000s.