Incorporating Geospatial Data in House Price Indexes: A Hedonic Imputation Approach with Splines

The increasing availability of geospatial data (i.e., exact longitudes and latitudes for each house) has the potential to improve the quality of house price indexes. It is not clear though how best to use this information. We show how geospatial data can be included as a nonparametric spline surface in a hedonic model. The hedonic model itself is estimated separately for each period. Price indexes are then computed by inserting the imputed prices of houses obtained from the hedonic model into the Fisher price index formula. Using a data set consisting of 454507 observations for Sydney, Australia over the period 2001-2011 we demonstrate the superiority of a geospatial spline over postcode dummies as a way of controlling for locational effects. While the difference in the resulting price indexes is not that large – since the postcodes in Sydney are quite narrowly defined – we nevertheless find evidence of a slight bias in the postcode based indexes. This can be attributed to systematic changes over time within each postcode in the locational quality of houses sold.