An Artificial Neural Network and Entropy Model for Residential Property Price Forecasting in Hong Kong

Abstract

Traditional approaches for housing price prediction fall short of accuracy, as it is difficult to identify a set of variables and account for their weightings when conducting forecasting. This study aims to explore an effective and efficient mathematical model for the housing price forecasting, so as to help developers, purchasers and financial institutes to obtain more reasonable property pricing through better decision‐making in the context of the fluctuant property market in Hong Kong. It began with a review of the macro and micro factors that affect the housing price and an entropy‐based rating and weighting model was presented with the aim of providing objectives and reasonable weighting to these variables. Then based on the key variables, the predictive ability of artificial neural networks (ANNs) was examined. In the empirical study, data were quantified and scaled with reasonable assumptions. Various networks were designed to examine the performance of ANN towards different parameters. Different sample sizes and different sets of input variables, together with different net structures and net types were undertaken to test the accuracy of ANN. From the comparison results of the R squared, as well as the mean absolute errors, the authors found that ANN performs well in forecasting with smaller sample size and standard net type. The overall results of this research demonstrated that the integration of Entropy and ANN can serve desirable function in the housing price forecasting progress.