Spatial Prediction Models for Real Estate Market Analysis

The econometric modeling of real estate prices is an important step in their valuation. As shown in the theory and practice of valuation, the most important determinant of these prices is location. Therefore, models comprising the spatial components give better estimates than a-spatial models. The purpose of this paper is to compare the quality of prediction for several models: a classical linear model estimated with OLS, linear OLS model including geographical coordinates, Spatial Expansion model, spatial lag and spatial error models, and geographically weighted regression. The evaluation will be based on the calibrated models for the real estate market data in Wroclaw in 2011. The study confirms that the inclusion of the spatial aspect of the analysis may result in improvement in the quality of models. Best fit to the data among the presented methods has proved a geographically weighted regression.