Data-driven techniques for mass appraisals. Applications to the residential market of the city of Bari (Italy)

The need for evaluation models capable of returning ‘slender’ and reliable mass appraisals of properties belonging to different market segments has been made mandatory by the events that are covering the global real estate finance, because of the emergence of non-performing loans in the banks’ balance sheets. In Italy, the non-performing loans have been estimated by the Italian Banking Association equal to about 300 billion euro in 2014. In the present paper, three approaches of data-driven techniques (hedonic price model, artificial neural networks and evolutionary polynomial regression) have been applied to a sample of residential apartments recently sold in a district of the city of Bari (Italy), in order to test the respective performance for mass appraisals. The models obtained by the implementation of the three procedures have been compared in terms of statistical accuracy, empirical compliance of the results and complexity of the functional relationships.