Predicting Housing Price Changes:An Evaluation of Alternative Hedonic Models

Accurate measurement of residential property price movements is important for government policy (e.g. one of the indicators used by central banks, computing affordability ratios). This thesis focuses on an evaluation of the accuracy of alternative econometric based price indices known as Hedonic Imputed price indices. These indices depend on the econometric model specification and estimation approach used to construct predictions of observed price changes. Four model specifications and two estimation approaches are studied in the thesis. The price changes implied by the constructed indices are compared (using mean squared error) to the median observed price changes based on properties that have sold repeatedly. The models differ in the specification of location; the two estimation approaches considered are the year-by-year estimation, popular in the price index literature , to a recently proposed smoothing method. Tornqvist and Jevon price indices are computed. They differ in the way transactions are weighted to construct the indices. The results indicate Hedonic Imputed price indices based on estimating the model by smoothing methods are more robust to model specification, and provide measures of price changes that are much closer to the observed price changes measured through the repeat sales.