Development and implementation of a predictive method for the stock market analysis, using the long short-term memory machine learning method

Abstract

In this work, the development and implementation of a computational tool based on the Long Short-Term Memory method were showed. The code was written in python, and consist of a recurrent neural network used in the field of deep learning. In the code, we implement artificial intelligence, which uses linear and logistic regression to make a predictive analysis based on historical data of each foreign exchange and the stock prices, with the target of predicting the next point of the future price (the price of closing of the futures trading candlestick). Crossvalidation between linear and logistic regression is also performed to see which of the two has the highest success rate, that is, the accuracy of the method is evaluated using two validation alternatives. In addition, we make a matrix of pairs of different foreign exchanges to identify which are the most correlated or inverse, so that the program can open its range of operations (simultaneously with different foreign exchange), and with this, a greater number of operations can be made per time established (in the foreign exchange case it is from one minute to five minutes, operations strategy is called scalping). Finally, we present the results obtained, based on the behavior of foreign exchange and stock prices, using a statistical predictive to assess the accuracy of said sample statistical model taken in this study.