Número de negócios e volume financeiro na previsibilidade do Ibovespa: uma aplicação de redes neurais artificiais

Abstract

This study focused on the use of information on the number of business and financial volume traded as explanatory variables that improve the performance in the predictability of Ibovespa return index. They are used daily information from the Ibovespa theoretical portfolio during the period from 23/01/2006 until 18/04/2008. Three models are applied to forecast based on Artificial Neural Networks with two layers of features and intermediate sigmoid activation. At training, the ANNs are adjusted automatically according to a criterion of error pre-defined based on a sample of 758 points. At the time of the forecast, the tracks ANNs are 50 steps ahead for verification of performance. The results show that the inclusion of information on number of business and volume of business improves the fit of ANNs in the phase of training, but the improvement was not maintained during the forecast.