Regularidades no lineales en índices accionarioS: Una aproximación con redes neuronales

Abstract

The artificial neural networks (ANN) have turned into an important tool to shape and to predict the stock returns. Due to the fact that those models incorporate nonlinear variables (characteristic of the majority of the economic and financial series) they work better than the statistical traditional models such as linear regressions or Box-Jenkins' model. This study brings the attempt of finding regularities in the stock indexes of 27 countries by means an approximation of artificial neural networks and their contrast with linear regressive models finding evidence that reaches to the current discussion on the "Efficient Market Theory". Likewise dynamics out of sample predictions are realized sustained also by a nonparametric test confirming excellent results of the neural networks in contrast with the traditional autoregressive models.