Artificial neural networks for predicting real estate prices

Econometric models, in the estimation of real estate prices, are a useful and realistic approach for buyers and for local and fiscal authorities. From the classical hedonic models to more data driven procedures, based on Artificial Neural Networks (ANN), many papers have appeared in economic literature trying to compare the results attained with both approaches. We insist on the use of ANN, when there is enough statistical information, and will detail some comparisons to hedonic modeling, in a medium size city in the South of Spain, with an extensive set of data spanning over several years, collected before the actual downturn of the market. Exogenous variables include each dwelling's external and internal data (both numerical and qualitative), and data from the building in which it is located and its surroundings. Alternative models are estimated for several time intervals, and enabling the comparison of the effects of the rising prices during the bull market over the last decade.