Predicting house prices for Singapore condominium resale market : A comparison of two models

Abstract

The geostatistical model is employed to solve the spatial autocorrelation problem in the hedonic price model and to obtain the accurate predicted house prices in the multiunit housing market. The empirical results based on the transaction data of Singapore condominium resale market from July of 1993 to June of 2003 show that the geostatistical models can effectively capture the building related spatial correlation. Among three geostatistical models, the negative exponential correlogram is the most appropriate method to predict the house price for the Singapore condominium resale market.Moreover, a comparison of model performance in estimation and house price prediction is carried out between the geostatistical model and the two order spatio-temporal autoregressive (2STAR) model. We conclude that the geostatistical model can more effectively correct the inefficient OLS estimation resulting from spatial autocorrelation than the 2STAR model does. Regarding to the prediction ability, the geostatistical model outperforms the 2STAR model.