HEDONIC PRICE MODELS FOR HDB RESALE FLATS

Abstract

This study attempts to determine the proper specification of the hedonic price model for the HDB resale flats market through the exploration of various functional forms. Out-of-sample testing is used to assess the predictive performance of the various MRA models. R-Square, F-statistic, Mean Absolute Percentage Error (MAPE) and Absolute Percentage Error (PE) are used as the evaluation criteria. The linear model is found to be the worst performing model while the log-log (linear) model is the most appropriate model for the HDB resale flats market. Age, floor level, MRT, bus-interchange, flat type and models and HDB towns are found to be significant variables in affecting resale flat prices. In addition, this study also tries to find out whether there is any economic significance in employing the widely applied linear model to the log-log (linear) model chosen by this study. More than 46 per cent of the comparisons between the predicted prices by the 2 models resulted in a discrepancy of more than $5,000, with 20.49 per cent resulting in a discrepancy of more than $10,000 and 10.53 per cent resulting in more than $15,000. Furthermore, the difference in the treatment of the variables, age and floor, by the two models, is found to affect their predictive power. The logarithmic form of the age and floor variables enhances the predictive power of the models.