Comparativo entre rede neural artificial de alimentação recorrente e para frente na previsão da cotação do contrato futuro do mini índice Bovespa

Abstract

Artificial neural networks that use forward and recurrent feeding approach in their architecturesare great tools to be used in analysis and prediction in time series, and therefore, in thismonograph it was investigated which one is the most suitable to be used in a stock exchange environment. A comparison was made between the forward and recurrent feeding approach inartificial neural networks, with predictions made by a Long Short Term Memory and Multilayer Perceptron network for the quotation of the future contract of the mini Bovespa Index in November 2019, to be later calculated the Pearson correlation coefficient and the determination coefficient R2of the set of its predictions with the set of the price negotiated in the month. Withthe results of the experiments, the recurrent feeding neural network obtained the best Pearsoncorrelation coefficient being better than the forward feeding network and also better than thepredictors chosen to be used as baseline.