Financial Market Shares Prediction Using LSTM Networks

Resumo

The financial market provides economic growth, as it is very efficient in capturing savings and other
resources for more productive activities. Stock investment is seen as a challenge even for specialist shareholders
with many years of experience, in which the results for this type of investment are quite variable, as they are based
on different macroeconomic factors with highlights in political events, bank rates, expectations investors, financial
conditions in general, among others. Therefore, accurately predicting the variation of these prices is quite
challenging, thus making it of great interest to investors to minimize this problem. Thus, artificial intelligence
techniques are quite favorable to predict stocks in the financial market. To predict stocks on the stock exchange,
we propose a methodology based on a recurring artificial neural network, the Long Short Term Memory (LSTM).
This type of neural network, designed for time series, takes time and sequence into account, giving a feedback
loop connected to your previous decisions, being quite appropriate for predicting data such as the stock exchange.
The input data of the LSTM network were shares on the Petrobras stock exchange, in the period from January
2014 to December 2017 to make a projection of a future month which is the month of January 2018 based on the
previous data, the operations were simulated in this action, obtaining the forecast of the stock's share with a
difference of 0.03 and 0.23 cents.