Hedonic real estate price estimation with the spatiotemporal geostatistical model

Abstract

This study argues that the spatiotemporal geostatistical model for real estate prices, which accounts for and incorporates spatial autocorrelation, can be estimated successfully using the Bayesian Markov Chain Monte Carlo (MCMC) estimation. While this procedure often encounters difficulty in calculating probabilistic densities in the Metropolis–Hastings (MH) algorithm, this study introduces a feasible and practical estimation method, providing useful estimated parameters for the model. Using single-family house transaction data, we show that ordinary estimations of real estate prices, with respect to certain explanatory variables, may lead to the underestimation of standard errors of coefficients for explanatory variables with spatial effects unless spatial autocorrelation is controlled for. Our model also makes it possible to obtain accurate in-sample predictions and moderately improved out-of-sample predictions for real estate prices. This study further estimates a “decay rate:” a diminishing correlation between real estate prices and increasing distance, showing that geographical proximities are likely to have an important impact on real estate prices, especially at a range under 600 m.