In this work the principal index of the Brazilian stock market
was studied through artificial neural networks and also by using
the adaptive exponential smoothing method. It is shown that the
neural networks have a superior performance to predict the correct
sign of the index return. The relative number of times the correct
movement of the market was predicted by neural networks was
about 0.60 which is similar to other reports performed onto more
developed markets by using ARN. The windowing technique used
to provide the training data and technical details of ARN did not
modify the forecasting results of the experience. The AES method
did not contribute to predict the correct sign of the return, in spite
that both methods, ARN and AES, produced almost the same RMSE
in the prediction of the return values. Since the profitable strategies
are related with the predictable character of the market movement
our study shows the possibility to develop support decision
systems for the Brazilian market based into the predictive possibilities
of the neural networks.