A Spatial Model of Housing Returns and Neighborhood Substitutability

This paper provides a method for estimating housing indicesat the local level. It develops a "distance-weightedrepeat-sale" procedure to exploit the factor structure ofthe error-covariance matrix in the repeat-sales model. Adistance function defined in characteristic andgeographical space provides weights for the generalizedleast-squares model, and allows the use all of the repeated-sales in a metropolitan area to measure returns for thespecific neighborhood of interest. We use distance-weightedrepeat-sales to estimate return indices for all zip codes inthe San Francisco Bay area over the period 1980 through1994. When distance is defined in terms of socio-economiccharacteristics, we find that median household income is thesalient variable explaining covariance of neighborhoodhousing returns. Racial composition and educationalattainment, while significant, are much less influential.Zip-code level indices often deviate dramatically from thecity-wide index, depending upon income levels. This hasimplications for investors and lenders. Our resultsindicate that rates of return may vary considerably within ametropolitan area. Thus, simply using broad metropolitanarea indices as a proxy for capital appreciation within aspecific neighborhood may not be justified.