PROFTS: a multi-agent automated trading system.

• Ano de defesa: 2019
• Autor(a) principal: Reis, Éverton Rodrigues
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Aprendizado computacional; Automated trading system; Autonomous portfolio management; Fundamental analysis; Machine learning; Multi-agent systems; Portfólios (Gerenciamento); Sistemas multiagentes; Technical analysis
• Link de acesso: http://www.teses.usp.br/teses/disponiveis/3/3141/tde-12022020-101943/

Portfolio management is a challenging and complex problem. The use of automated trading systems (ATS) is becoming common nowadays. However, most of them focus on maximizing return, without considering the risk, while few consider the relation between risk and return and investor\'s preferences. Moreover, most ATS use technical analysis/data, very few apply fundamental analysis, and quite none apply both of them, which is the way how most human analysts deal with this problem. In this work, it was proposed an architecture for an automated trading system (ATS) that manages an active stock portfolio, combining both fundamental and technical analysis, for different types of investor\'s risk profile. The architecture, called PROFTS, was built using a multi-agent approach and machine learning techniques, and validated through simulations of the Brazilian stock market. The results were compared with the performance of the IBrX 100 using a buy and hold strategy. A financial distress prediction model was also utilized, in order to filter out companies that were bankrupting from those that were undervalued. From the results, considering trading costs, the PROFTS was profitable. Portfolios that used technical analysis combined with fundamental analysis presented statistically better results than those that just used fundamental analysis. Moreover, portfolios that used the financial distress prediction model also presented statistically significant average lower risk when compared with those that did not use them.