Accuracy and robustness of house price index methods

Abstract

We evaluate the statistical properties of five different house price index methods withthe objective of identifying one that is most accurate and robust when estimated at frequent timeintervals and for distinctly local markets. We adopt a split-sample technique to establish a consistentbasis for comparison of the different price index methods. Our results demonstrate that if suitabledata is available for estimation of price indexes, the arbitrary aggregation of data across time andgeography is not warranted. One model, the ‘hedonic imputation’, outperforms alternative modelson all measures of accuracy and robustness. Differences in levels of accuracy between differentmodels are found to be statistically significant.