Previsão do preço de ações brasileiras utilizando redes neurais artificiais

• Ano de defesa: 2021
• Autor(a) principal: Torres, Sérgio
• Orientador(a): Hadad Junior, Eli
• Banca de defesa: Basso, Leonardo Fernando Cruz , Mendonça, Diogo de Prince
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Presbiteriana Mackenzie
• Programa de Pós-Graduação: Administração de Empresas
• Departamento: Centro de Ciências Sociais e Aplicadas (CCSA)
• País: Brasil
• Palavras-chave em Português: GRU; LSTM; machine learning; MDA; redes neurais artificiais
• Área do conhecimento CNPq: CNPQ::CIENCIAS SOCIAIS APLICADAS::ADMINISTRACAO::ADMINISTRACAO DE EMPRESAS
• Link de acesso: https://dspace.mackenzie.br/handle/10899/28379

The stock market analysis is influenced by the upward and downward trend related to the asset prices in relation to short periods. Due to the feelings and the expectations of investors, which can affect the movement of shares. One of the possibilities to maximize the returns in stock forecast is to predict the directional asset movement instead of forecasting the prices. This paper analyses the use of artificial neural networks to preview the directional asset movement and compares the performances of two architectures: LSTM and GRU. Shares and Ibovespa index are analyzed in several time horizons aiming at the observations of specific patterns. The performance of the said architectures is very similar to each other: for short time horizons, the LSTM architecture presented a better accuracy, and for long time horizons, the GRU is more accurate than LSTM.