Especificación de modelos econométricos utilizando minería de datos

Abstract

In this paper we present a new procedure for selecting econometric models. It is based on a heuristic approach, called genetic algorithms, which permits us to explore the universe of available models starting from a general model without restrictions. This search process for the optimal model is guided only by the Schwarz Information Criterion, which acts as the lost function of the genetic algorithm employed for selecting the optimum. This procedure shows good performance with respect to other methodologies. As examples of its utility three problems where the algorithm was successfully employed are presented: the selection of variables that explain economic growth, the prediction of failure of firms and the construction of a portfolio with few actives that follows the behaviour of a stock exchange index such as the Spanish IBEX35.