Desenvolvimento de robô de investimento para day trade baseado em SVM One Class e RNA

Abstract

The financial market is the trading environment for various financial instruments, including stocks, bonds, currencies and derivatives. This market is of vital importance for the proper functioning of capitalist economies. An important segment of the financial market is the securities market, which enables speculation on futures contracts. In Brazil, the mini Ibovespa futures contract (WIN) is the financial asset most traded by individuals in the intraday trading modality (daytrade). Traders and investors who carry out purchase and sale operations in this market face numerous challenges that make their task difficult at the time of decision making. These challenges can affect both more experienced investors and beginners in this market. Most studies available in the literature employ conventional statistical and econometric approaches to try to predict the future price of a given financial asset through regression analysis. Therefore, there is a lack of research in the field of developing models dedicated to predicting the direction of price, that is, treating the problem as one of classification. In this context, this work proposes an artificial intelligence model based on SVMOne Class Artificial Neural Networks (ANN), which the proposal is to predict the direction of the price of the Bovespa Index (WIN) futures contract in a 5 (five) minute graph time. The main differential of this work compared to those available in the literature is the use of the SVMOne Class or the Savitzky-Golay filter. Another highlight is the validation of results via backtesting, using an automated trading system for the financial market called Expert Advisor (EA) developed on the free MetaTrader5 (MT5) platform. Backtesting allowed to obtain metrics in a simulation environment with real data from the financial market. The results obtained from the backtesting are more realistic and, therefore, differ from the results achieved only by analyzing the assertiveness of the AI model, which presented an average rate of 60.22% of assertiveness in the predictions. This analysis served to prove the importance of validating AI models by applying backtesting systems to analyze the results.