ISBIS - Residential Real Estate Market Dynamics in Bengaluru

View the recording

Recent research on real estate markets has emphasised the need to characterise property price dynamics across both space and time. This study contributes to this literature by examining spatial and temporal dependencies in housing prices and by assessing whether alternative residual distributions materially improve model performance. Because housing price data can exhibit heavy tails, the standard Gaussian error assumption may be questioned. We propose a Bayesian spatio-temporal model that captures dependence using a separable spatio-temporal process with an additive mean structure and a flexible error specification. We apply the proposed approach to housing transaction data from Bengaluru, India, covering 62 monthly periods and 76 micromarkets from February 2015 to March 2020. The results provide strong evidence of spatial and temporal dependence in log housing prices. However, the choice of residual distribution has minimal impact on model fit or predictive accuracy across all evaluation settings. In contrast, explicitly modelling spatio-temporal dependence yields substantial and consistent improvements relative to a hedonic benchmark. Overall, the findings highlight the importance of modeling spatio-temporal structure in housing markets, while suggesting that Gaussian errors remain a robust and practical default even when the data display mild departures from normality.