A Semi Parametric Method for Estimating Local House Price Indices

Abstract

Spatial autoregressive hedonic models utilize house prices lagged in space and time to produce local house price indices: e.g., the STAR model might be used this way. This paper complements these models with a semiparametric approach, the Local Regression Model (LRM) designed to use large databases to produce neighborhood price indices and confidence intervals at each point on a spatial. The greater flexibility of the LRM may allow it to identify space-time asymmetries missed by other models. The LRM is fitted to 49,511 sales from 1972Q1 - 1991Q2 in Fairfax County, Virginia. The local price indices display plausible and significant variations over space and time. On average, LRM indices are the same as a standard hedonic price index for the entire county. The LRM price indices in selected neighborhoods are shown to differ significantly from those in some other neighborhoods. A new method for estimating standard errors addresses an overlooked problem common to all local price indices: How to evaluate the amount of noise in the estimates. Out-of-sample prediction errors demonstrate that LRM adds significant information to the hedonic model. Moran's I-statistics suggest that future research could usefully add a STAR or spatial autocorrelation model to the LRM.