Modelos predictivos de redes neuronales en índices bursátiles

Abstract

This study analices the neuronal networks models' capacity to predict the sign of the weekly variations of CAC40, Hang Seng, KLSE, MMX, STI, Dow Jones Industry, S&P500, GDAX, Bovespa, Nikkei225 and FTSE100. The relative performance of the models was measured by the number of correct predictions of the index's variation sign, on the base of 51 joint out-samples, each one made up of 50 weekly observations. The results proved that the predictive capacity of the models change through the time, then it is necessary to reestimate the weight of the equation and reconstruct the model period after period. This suggest that an exclusive and unique explanatory model of the stock-exchange indice's evolution does not exist.