Redes Neurais LSTM e otimização de portfólio para auxílio a tomada de decisão na Bolsa de valores

Detalhes bibliográficos
Ano de defesa: 2022
Autor(a) principal: Vinícius Teodoro de Castro Pires
Orientador(a): Não Informado pela instituição
Banca de defesa: Não Informado pela instituição
Tipo de documento: Dissertação
Tipo de acesso: Acesso aberto
Idioma: por
Instituição de defesa: Universidade Federal de Minas Gerais
Brasil
ICEX - INSTITUTO DE CIÊNCIAS EXATAS
Programa de Pós-Graduação em Ciência da Computação
UFMG
Programa de Pós-Graduação: Não Informado pela instituição
Departamento: Não Informado pela instituição
País: Não Informado pela instituição
Palavras-chave em Português:
Link de acesso: http://hdl.handle.net/1843/48518
Resumo: In an economic scenario where the profitability of traditional investments is often lower than the inflation of the same period, investing in the stock market has become an attractive option. The focus of the research carried out was to build investment strategies that rely on already known algorithms for portfolio optimization, neural networks and data processing, combining them in search of a bold strategy. The Long Short-Term Memory recurrent neural network was used to build models capable of predicting the expected return and evaluating actions for daily operations. The strategies built brought combinations of different optimization objectives, such as Sharpe maximization, quadratic utility maximization and variance minimization with different ways of building the expected return parameter of the optimizers. In addition, the performance of daily purchase and sale operations was evaluated based on the value predicted by the model for the assets within the optimized portfolios of each strategie. As a benchmark of models created, the IBOVESPA and the IPCA were chosen, which in the evaluation period between January 2021 and April 2022 resulted in -9% and 14.78%, respectively. The strategies with the best results of the project were in optimized portfolios that used the expected return predicted by LSTM as an input, where Sharpe had a return of 35.79% and Quadratic Utility 66.93%. Despite the best result, the portfolio that optimizes Quadratic Utility has low diversity and the portfolio that maximizes Sharpe has greater diversity and may be more suitable for investors who do not want a high degree of risk. Considering the diversity of stocks in the portfolio composition, the minimization of variance is an interesting proposal as it had a return of 11.33% but presented less volatile results and a greater number of different assets monthly. Operations in optimal portfolios did not obtain good results due to the change in the composition of the portfolio in each operation, which does not guarantee an optimized portfolio during the period of operation.