PREDICTING SINGAPORE'S PRIVATE RESIDENTIAL HOUSING PRICE: COMPARING HEDONIC REGRESSION AND MACHINE LEARNING METHODS

Abstract

ABSTRACT Purpose – This study aims to predict Singapore’s non-landed private residential housing prices based on all information available at the time of evaluation. Zestimate proposed by Zillow in America and X-Value created by SRX in Singapore are popular online free tools utilized by local people that provide reliable estimations on property values. This study intends to construct models that approach or outperform X-Value, and compare the performances of the trained models.

Methodology – One linear method Hedonic Price Model, two machine learning models Random Forests and XGBoost are built on a training set and evaluated on a test set. The two tree-based methods mirror a professional valuer’s appraisal process with the Direct Comparison approach. To evaluate a property by Direct Comparison, a variable called WAUP is computed via a specific algorithm to summarize the mean of unit price of the property’s all comparable (nearby and recent) transactions. Model performances on the test set are measured by Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE)

Findings – Test MAPEs returned by the hedonic model, Random Forests and XGBoost are 10.67%, 4.71%, 4.65% respectively. WAUP, Distance to CBD (Raffles Place MRT Station), Property Age are important variables in both machine learning methods. Without consideration of test sets, X-Value performs slightly better than Random Forests.