Using data-driven prediction methods in a hedonic regression problem

The traditional studies about hedonic prices apply simple functional forms such as linear or linearity transformable structures. Nowadays, it’s known in the literature the importance of introducing non-linearity to improve the models’ explanatory capacity. In this work we apply data-driven methods to carry out the hedonic regression. These methods don’t impose any a priori assumption about the functional form. We use the nearest neigbors technique as non-parametric method and neural networks and genetic algorithms both as semi-parametric methods. Neural Networks have already been employed to the specific hedonic regression problem but, to the authors’ knowledge, this is the first time that a genetic algorithm is employed. The empirical results that we have obtained demonstrate the usefulness of applying data driven models in the estimation of hedonic price functions. They can improve traditional parametric models in terms of out-of-sample R2.