Real estate pricing models with spatial autocorrelation: a review

Abstract

In a first part, we review the literature about classical hedonic pricing models for real estate. Dubin (1988) introduced spatial autocorrelation in the real estate literature. Indeed house values depend on distance to various amenities so that neighboring houses should be affected similarly by locational attributes. ." It is believed that the improvement obtained by spatial models stems from incorporating the influence of omitted variables. Pace and LeSage (2004) find that OLS estimates overstate the sensitivity of house prices to racial composition, income, poverty, school quality and education levels, relative to estimates based on spatial models that take spatial dependence into account. We find many different approaches in the way the spatial dimension is introduced in hedonic modeling: models inspired from the geostatistics literature, models belonging to the family of simultaneous spatial autoregressive models, geographically weighted regression and moving window approaches, Bayesian models, etc. Finally we discuss the interest of the spatial quantile regression approach.