Alocação de carteiras de ações utilizando aprendizado de máquina e regras fuzzy

Abstract

Long-term investments through stock portfolios attract the attention of investors, who are looking for simple and effective ways to select stocks to compose a portfolio, as well as find a series of assets that will have greater appreciation in the future. Over the years, several techniques have been developed, and this work aimed to use Machine Learning (AM) to predict future stock values. To maximize their gains, investors look for information or techniques that are able to help them. In view of this, AM models have become frequently used to discover stocks that have a high potential for appreciation. Unlike short- and medium-term applications that make use of time series, long-term applications require simpler methods. In this work, a strategy for allocating portfolios for long-term investments (1 year) with AM models was elaborated and evaluated, namely: Linear Regression (RL), Regression Tree (AR), Random Forest (RF), K- Nearest Neighbors (KNN), Bayesian Ridge (BR) and Fuzzy Inference Systems (SIF). In addition to this main objective, a qualitative analysis of the interpretability of the rules of a model based on fuzzy rules was also carried out, in order to seek ways to understand the relationship between input and output attributes based on linguistic terms. The result obtained in this work showed that the approach by using AM models to predict the performance of stocks over a year and subsequent stock selection is efficient for investors to obtain returns higher than the S&P500 market indices and by manual methods.