STOCK MARKET PREDICTION USING NEURAL NETWORKS

Resumo

Trading on the stock exchange is a fundamental part of a country’s economic development, resulting
in profits directly or indirectly from market transactions. However, the stock exchange is highly dynamic, stocks
rise and fall as a result of changes in various parameters. Some techniques have already been used to try to predict
these changes in the market, but there is still a need for improvements to ensure greater accuracy in the market’s
bets, as most of them are high risk. Predicting an asset’s closing value is a difficult task because the asset’s value
is constantly changing and has a variety of parameters that influence its increase and decrease. In this work, an
algorithm based on Multilayer Perceptron Neural Network (MLP) was implemented to estimate the price of assets,
predicting the closing value of the stock market shares. The proposed network architecture was modeled with two
hidden layers, containing six neurons in the first layer and four neurons in the second layer and one neuron in
the output. In this case, MLP considerably closed the closing of all the proposed shares (EQTL, PETR4, IBM,
ABEV3, BBAS3, CIEL3, COGN3, LAME4, OIBR3 and VALE3) in the period of 10 consecutive years, forecasting
an average of 93.60% of the value of the actions in the year following the training of the network.