Towards robust house price index estimation: An empirical investigation of San Diego, California

I develop a two step method for modeling the spatial and temporal variation in house prices across the regional landscape. In the first step, I utilize the Goetzmann and Spiegel (1997) Distance Weighted Repeat Sales (DWRS) model to construct "neighborhood" level indices. The DWRS model is a hybrid model: it provides "hedonic controls" via the distance function within the repeat sales framework. In the second step, I model the abnormal appreciation in the neighborhood (defined as the appreciation of the neighborhood's house price index over and above the regional average in the period) within an Autoregressive-Generalized Autoregressive Conditional Heteroskedasticity in Mean (or AR-GARCH-M) framework to characterize its temporal variation. I construct a buffer of "neighbors" variable for each neighborhood and use this to capture the spatial pattern of abnormal house price appreciation.; The benefits from adopting this two-step procedure are at least three-fold, Firstly, adopting the DWRS model enables the construction of accurate neighborhood-level house price indices that would otherwise not be available for small geographic areas. Secondly, this procedure models neighborhood effects and adjacency effects within a repeat sales context (unlike previous modeling within a hedonic specification). Thirdly, the AR-GARCH-M framework models the impact of neighbors and lagged values explicitly. A picture of "neighbors" in "time and space" thus emerges.; I also provide an explicit derivation of the variance-covariance matrix with spatial components for the DWRS model and explore the various conceptual and measurement issues that arise in implementing the estimation procedure.; Finally, I implement the two step procedure for a data set of repeat sales of single family homes in San Diego County, CA between 1995 and 1998 geocoded to Census tracts (a smaller geographic area than is usually employed). This yields 408 quarterly tract-level house price indices that show a wide range of appreciation over the period. I find that the use of a single area-wide house price index (such as the standard Weighted Repeat Sales index) as a proxy for house price appreciation within a specific area or neighborhood is not justified. This work should be of particular interest to the mortgage and mortgage-backed-securities market participants.