A flexible method of housing price index construction using repeat-sales aggregates

The major issue which we address in this article is the one-size-fits-all nature of the typical city-level housing price index. In this vein, we make two contributions. First, we develop a new algorithm to ensure feasible estimation of geographically granular repeat-sales price indices in cases of low transactions counts. This facilitates the estimation of a balanced panel of 63,084 U.S. Census tract-level indices (2010 definitions) at an annual frequency between 1989 and 2021, which we release alongside this article. Second, we use these indices to estimate city-level price indices that are robust to heterogeneous submarket appreciation and nonrandom sampling, two issues that confound classic approaches. Different index targets require alternative weighting schemes, and these formulations can result in index differences that can widen over time horizons. However, in some cases, sample-based indices are quite similar to more strictly defined index targets; for instance, in the early COVID-19 period, standard sample-based indices are actually quite similar to a unit-representative house price index for large cities.