FORECASTING HOUSE PRICE INDEX USING ARTIFICIAL NEURAL NETWORK

Property forecasting is an important component in the decision-making process for investors and governments in supporting asset allocation, formulating property funding strategies, and determining suitable policies. Multiple regression analysis has been accepted as the most commonly used technique for property forecasting. However, this technique has a limited ability to effectively deal with relations among variables, nonlinearity, and multicolinearity. A number of alternative techniques have been applied to improve the reliability of forecasting information, one of which is artificial neural network. This study applies this technique to forecast residential property prices in Malaysia. Unemployment rate, population size, mortgage rate, and household income were chosen as the independent variables while Housing Price Index (HPI) was chosen as the dependent variable. Quarterly time-series data from 2000 to 2009 were used for training and testing the ANN model. Subsequently, data for 2010 and 2011 were used to validate the model. Model validation has resulted in the mean absolute percentage error (MAPE) of 8% thereby suggesting a high level accuracy of ANN in its ability to learn, generalize, and converge time series data efficiently and to produce reliable forecasting information.