| dc.description.abstract | Two common methods in real estate analysis are hedonic regression and repeat sales. While research has pointed out the merits of combining these models constructing house price indices, no study to our knowledge has examined this invention for property valuation. This paper investigates potential benefits of the described combination in a price prediction context, constructing house value estimates by combining predictions from repeat sales and various hedonic regression specifications enhanced to account for spatial effects. Three of these enhancements regression kriging; mixed regressive, spatial autoregressive model and geographically weighted regression are acknowledged spatial econometric models. Further, the article proposes a fourth augmentation which addresses systematic residual patterns in regressions with district indicator variables and the presence of outliers in housing data. Running the models on a data set containing 16,417 transactions in Oslo, Norway, we find that the repeat sales combination reduces the median absolute percentage error of the hedonic models with 6.8 % to 9.5 %, where larger gains are observed for less accurate spatial enhancements. We attribute the improvements to both spatial and non-spatial information inherent in previous sales prices. While the former has limited utility for well specified spatial models, we believe the non-spatial information in previous sales prices is able to capture otherwise hardly observable phenomena, making its contribution potentially highly valuable in automated valuation models. | |
