Automated valuation models (AVM) are increasingly often used by institutional mortgage lenders and real estate agents to determine the value of real estate. Usually, AVM use hedonic models as a basis for appraisal. However, the results are often criticised by real estate professionals, as they lack transparency and comparability to traditional appraisal methods.

Therefore, we provide an alternative approach for automated real estate valuation of self-used property in Austria: Instead of using pure hedonic models, we develop a statistically supported sales comparison approach. Therefore, in a first step we select comparable houses or flats in the surroundings of the valuation object; in a second step, we adjust the prices of these comparables using hedonic methods; finally, the market value is derived as a (inverse distance-weighted) mean of these comparables. The quality adjustment parameters are differentiated with respect to property characteristics, time and location, and thus provide a basis for structured valuation of the results.

For our research, we have an extensive data set for Austrian single family homes and flats at our disposal. We apply generalized additive models (GAM) that take into account nonlinearity as well as discrete and continuous spatial heterogeneity. The results are shown in an interactive appraisal dashboard hosted by DataScience Service GmbH.