Previsão de fundos de investimentos com o uso de machine learning

Abstract

The financial market is broad and provides to the investor with several investment options, such as fixed income products, stocks and investment funds, for example. For the case of Brazilians, the security of its application and its profitability are taken into account. However, for investment funds it is possible to estimate the future profitability of the application, using only historical data as a reference for its behavior. In this context, the present study aimed to analyze the Machine Learning techniques to forecast future variations of shares in three fund types: conservative, moderate and aggressive. For this purpose, Neural Network and Linear Regression were used, with different training sets, i.e., a total of nine data sets of historical series of quota and destination variation were separated by three different metrics: mean absolute error, mean square error and root of the mean square error. Using the software Weka (Waikato Environment for Knowledge Analysis), with the Forecast extension, the algorithms were trained and tested, the Neural Networks technique stands out, which had a better performance for the different funds and training sets. However, the Linear Regression obtained interesting results when forecasting the quota variation with the training sets using one year of historical data.