Housing Sub-markets and Hedonic Price Analysis: A Bayesian Approach

Sub-market definition has been identified as an important step in hedonic price analysis, to improve price predictions as well as estimates of marginal prices. Early market segmentation procedures were based on various definitions of housing or neighbourhood type. With the advent of new technologies (e.g. GIS), and statistical techniques (e.g. spatial models, clustering techniques), more sophisticated forms of analysis have become possible. The objective of this paper is to contribute to the literature on market segmentation by investigating the use of Bayesian models with spatially varying coefficients to delineate sub-markets based on geographical and functional considerations. The case study is the housing market in the city of Toronto in Canada. A comparison to a geographically weighted regression (GWR) model is made in terms of error in estimating and predicting hedonic prices, as well as patterns of submarkets indicated by the spatially varying coefficient processes. Our results provide an interesting contrast to previous research that used moving window approaches to define so-called soft market segments. In addition, we describe how to introduce qualitative (i.e. expert) information obtained from the relevant property assessment corporation into hedonic price forecasting within the Bayesian framework.