Distance in Geographic and Characteristics Space for Real Estate Price Prediction

Abstract

The common approach to predict the price of residential property is the hedonic price model and its extension to the case of spatial autoregression. The hedonic approach models the dependence between the price and internal characteristics of an apartment, house characteristics and external characteristics. To account for the unobserved quality of the surrounding environment price model includes factors of spatial price correlation, where the distance is usually measured as the distance in geographic space. Determining the price the seller focuses not only on the observed and unobserved factors of the apartment, house and its environment but also on the prices of similar marketed objects which can be selected both by geographic proximity and by characteristics similarity. In this paper, we use ensemble clustering approach to measure objects proximity and test that the proximity of objects in the characteristics space along with spatial correlation explains the significant variation in prices that in turn leads to an improvement of predictive ability of the model.