PROPOSTA DE UM MODELO DE PREDIÇÃO DA BOLSA DE VALORES USANDO UMA ABORDAGEM HÍBRIDA

Abstract

The stock market is a highly complex market and an important way to raise funds for investors. Investors aim to achieve maximum profit. Thus, the purchase or sale of shares must be made on time. To achieve this goal, prediction techniques can be applied to the stock market in order to predict their behavior. There are several techniques that show promising results for prediction on the stock market, but each one has particular advantages and limitations. Thus, these techniques require a large number of variables and a complex architecture. This paper describes a proposal of a prediction approach model designed to raise the return's chances on investments in shares. It uses a hybrid approach based on the definition from a committee of learning machines which combines the advantages of three techniques (statistical, neural network and technical indicators) in order to overcome their respective limitations. The proposed approach is examined through a comparison with their individual forecasters according to the perspective of two metrics, which ones demonstrate significant gains in terms of profitability and precision when used with a committee prediction approach involving weights in the solution of individual forecasters.