A Bayesian Approach to Residential Property Valuation Based on Built Environment and House Characteristics

Abstract

Housing market receives a broad attention from society. Understanding how built environment and house characteristics are valued in housing market is critical to investment and city development. However, this problem is challenging because of the existence of submarket resulted from the heterogeneous urban form and physical barriers. Traditionally, residential property valuation is carried out by Hedonic Price Model(HPM). However, traditional HPM based on linear model has limitation in valuation accuracy and suffers from submarket effect. In this paper, we propose a Bayesian approach to residential property valuation based on built environment and house characteristics. Specifically, we introduce a latent variable representing housing submarket and model corresponding factors and HPM into a Bayesian network. Utilizing the dependencies modeled in the Bayesian network, our model is able to capture the characteristics of submarket in location proximity, house attribute similarity and substitutability. Meanwhile, our model leverages the mutual enhancement of clustering and regression to build HPMs for each submarket. We conduct empirical evaluations quantitatively and qualitatively in housing market of Nanjing, China. The result shows that our method outperforms all baseline methods in residential property valuation accuracy. Besides, using our model, we are able to interpret the submarkets in Nanjing and quantify the effect of house features on housing price in each submarket.