Antecipação dos movimentos de preços de ações através de redes neurais artificiais

Abstract

The context of the research is the increase in the number of people investing their savings to gain some profit or hedge their portfolio. Consequently, Artificial Neural Networks (ANNs) are used in financial markets and studied to predict the asset's price. As there are a lot of possibilities for the architecture of ANNs and a high number of assets, the research aims to verify the capacity of ANNs to predict the movement of the price of the stock VALE3 and measure their accuracy. To achieve this aim, two ANNs and one aggregator module are constructed, which form the ensemble. After 20 experiments changing the architecture of ANNs, the best ensemble has a global accuracy equal to 57,89%, higher than both ANNs. It is more accurate to predict down movements than up movements, and the situation is better to hedge.