Um modelo preditivo de cotação de ações de empresas estatais brasileiras utilizando redes neurais artificiais no ambiente MATLAB

• Ano de defesa: 2019
• Autor(a) principal: Alves, João Victor da Silva
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; Brasil; Programa de Pós-graduação em Gestão Organizacional (Mestrado Profissional)
• Palavras-chave em Português: Redes Neurais Artificiais; LSTM; Modelo preditivo; Previsão de preços de açõe; Artificial Neural Networks; Predictive model; Stock Price Forecast; CNPQ::CIENCIAS SOCIAIS APLICADAS::ADMINISTRACAO::ADMINISTRACAO PUBLICA::ORGANIZACOES PUBLICAS; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::TEORIA DA COMPUTACAO::COMPUTABILIDADE E MODELOS DE COMPUTACAO
• Link de acesso: https://repositorio.ufu.br/handle/123456789/28218; http://doi.org/10.14393/ufu.di.2019.2475

Properly predicting financial market behavior is a goal pursued by investors, managers and researchers. Correct and reliable predictive models aim to assist decision making, seeking to minimize losses and generate gains in financial transactions intrinsic to stock market activities. In this context, the use of models based on artificial neural networks (ANN) has increased significantly in the last decade. Since ANN are capable of learning and storing experimental knowledge, their application in finance can predict stock market behavior. Thus, the aim of this dissertation is to develop a model that uses ANN to forecast the maximum and minimum market share prices of some of the currently selected large state-owned brazilian companies, from 2013 to 2017. To achieve this objective, historical stock data were traded on the official Brazilian stock exchange (currently called B3), which became input variables for the RNAs used in the construction of the proposed predictive model. The architectures chosen to build the predictive model were MLP (Multilayer Perceptron) and LSTM (Long Short-Term Memory). The results presented by the MLP RNA-based model have been compared with LSTM RNA-based solutions, which have been widely used in scientific research aimed at developing predictive models for finance and several others. In the comparative analysis, LSTM RNA results were more accurate and reliable when compared to MLP RNA solutions. Thus, it was possible to conclude that RNA with LSTM architecture and its current optimizations are possible options for the construction of models to predict stock market behavior. Notably, the technological product built from this research provides an RNA-based predictive model, enabling the choice of MLP or recurrent LSTM architecture.