Price determinants of public housing in Singapore : an econometric and machine learning approach

• Main Authors: Tang, Zachary Jia Ying, Peh, Benedict, Lim, Daren Jun Hao
• Other Authors: Wang Wenjie
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Social sciences::Economic theory
• Online Access: https://hdl.handle.net/10356/148246
• Institution: Nanyang Technological University

Understanding the price determinants of real estate and the ability to predict prices can be beneficial for homebuyers, homeowners and policymakers. This paper seeks to examine the price determinants in Singapore’s public housing market using both econometrics and machine learning approaches. The paper also studies feature importance within machine learning models to understand which variables are the most significant in improving predictive performance. Using transactional data from 2018-2020, we estimated a hedonic pricing model which quantifies the price effects of important housing attributes with a focus on a set of 18 locational amenities. Results showed a large and significant price premium for the proximity to a MRT (Mass Rapid Transport) station. In particular, an increase in proximity by 1 kilometre to a MRT station will increase flat prices by S$36,843 on average. The price effect is larger compared to bus interchanges and bus stops, suggesting that consumers value MRTs over bus transportation. In terms of education, we find that the proximity to a top-ranking primary and secondary school had a price premium of S$2,532 and S$13,999 on average respectively, significantly outweighing the price decreases due to noise and congestion from the schools. Results also indicate a S$15,601 price discount on average for every kilometre increase in proximity to nursing homes, quantifying the negative perceptions that consumers have of nursing homes. For machine learning, seven algorithms (LASSO, RIDGE, K-Nearest Neighbours, Artificial Neural Network, Regression Tree, Random Forest, Boosted Trees) were trained and evaluated based on the out-of-sample prediction accuracy. Our findings showed that nonlinear models outperformed linear models, with the best prediction accuracy produced by Artificial Neural Networks. In terms of feature importance, we find that the proximity to MRT stations and government hawkers were important considerations for homebuyers. This finding is robust across different machine learning methods. For other amenities, the findings were less conclusive due to their low relative importance.