Comparing econometric analyses with machine learning approaches: a study on Singapore private property market

Abstract

We aim to compare econometric analyses with machine learning approaches in the context of Singapore private property market using transaction data covering the period of 1995-2018. A hedonic model is employed to quantify the premiums of important attributes and amenities, with a focus on the premium of distance to nearest Mass Rapid Transit (MRT) stations. In the meantime, an investigation using machine learning algorithms under three categories-LASSO, random forest and artificial neural networks is conducted in the same context with deeper insights on importance of determinants of property prices. The results suggest that the MRT distance premium is significant and moving 100m closer from the mean distance point to the nearest MRT station would increase the overall transacted price by about 15,000 Singapore dollars (SGD). Machine learning approaches generally achieve higher prediction accuracy and heterogeneous property age premium is suggested by LASSO. Using random forest algorithm, we find that property prices are mostly affected by key macroeconomic factors, such as the time of sale, as well as the size and floor level of property. Finally, an appraisal on different approaches is provided for researchers to utilize additional data sources and data-driven approaches to exploit potential causal effects in economic studies.