Sistema automático de negociação para a bolsa de valores utilizando redes neurais multilayer perceptron e regressão linear

Abstract

Predicting the behavior of the financial market has always attracted the interest of investors. Manual trading brings a number of difficulties for investors, so it is increasingly common to use automated trading systems. This work used MLP neural networks to try to predict the financial market. Two forms of training were used by the neural networks: one based on the Linear Regression Slope indicator and the other based on the index closing price the next day. A trading system with stop loss, take profit mechanisms and money management was built using the Percent Volatility strategy. The BOVA11 index was used for trading in demo accounts of a financial broker. The system was based on predictions of neural networks trained and selected periodically for trading. The trading period applied was from 07/01/2014 to 06/30/2018. Neural networks trained by closing price of the next day outperformed the one based on linear regression slope and the buy and hold strategy in the analyzed period.