Redes neurais convolucionais aplicadas à negociação de ativos no mercado financeiro
Carregando...
Arquivos
Citações na Scopus
Tipo de produção
Dissertação
Data
2021
Autores
Nascimento, D. G.
Orientador
Bianchi, Reinaldo Augusto da Costa
Periódico
Título da Revista
ISSN da Revista
Título de Volume
Citação
NASCIMENTO, D. G. Redes neurais convolucionais aplicas à negociação de ativos no mercado financeiro. 2021. 112 f. Dissertação (Mestrado em Engenharia Elétrica) - Centro Universitário FEI, São Bernardo do Campo, 2021. Disponível em: https://doi.org/10.31414/EE.2021.D.131316.
Texto completo (DOI)
Palavras-chave
Redes neurais (Computação),Bolsa de valores
Resumo
Previsão do mercado financeiro tem sido um desafio bastante popular nas pesquisas de Aprendizado de Máquina (AM). O desejo da maioria dos investidores é tomar decisões com base em critérios objetivos que venham a proporcionar maior retorno nas operações. Recentemente, estudos têm usado técnicas de Aprendizado Profundo (AP), como Redes Neurais Convolucionais (Convolutional Neural Networks - CNN), para realizar regressão nos preços ou classificação de sinal de negociação em ativos do mercado financeiro. Neste trabalho, é proposta uma arquitetura de sistema que utilizada uma CNN a fim de realizar a indicação da melhor operação para cada momento no mercado de ações, este sistema foi chamado CNN Trading Classifier (CNN-TC). Este sistema é composto por pré-processamento dos dados, classificação
pelo modelo CNN e tomada de decisão no mercado. O mesmo foi avaliado com base em dados das bolsas de valores brasileira e americana em três períodos diferentes, para isto foram feitas
avaliação estatística, utilizando as métricas de classificação acurácia, precisão, revocação e F1, e financeira com base nas classificações realizadas pelo modelo. Além disso foi realizado um teste em ambiente simulado utilizando o software MetaTrader a fim de atestar a eficácia desta abordagem. Os resultados mostram que o sistema teve resultados estatísticos e financeiros melhores na maioria das avaliações em comparação com o uso de outros modelos de AP e superou a estratégia Buy and Hold (BH) e retornos da renda fixa.
Stock market forecasting has been a quite popular challenge in machine learning researches. Most investors want to make decisions based on criteria that will provide greater returns in their operations. Recently, studies have been using Deep Learning techniques, such as Convolutional Neural Networks (CNN), to perform price regression or trade signal classification in financial market. In this work, a system architecture that uses a CNN model is proposed to perform the indication of the best operation for each moment in the stock market, this system was called CNN Trading Classifier (CNN-TC). This system consists of data pre-processing, classification by the model and decision making in the market. It was evaluated based on data from the Brazilian and American stock market in three different periods. For this, statistical evaluation was performed, using the metrics of accuracy, precision, recall and F1 classification, and financial based on the classifications performed by the model. In addition, a test on a simulated environment using the MetaTrader software was evaluated in order to attest to the effectiveness of this approach. The results show that the system had better statistical and financial results in most evaluations compared to the use of other Deep Learning models and overcame the strategy Buy and Hold (BH) and fixed income returns.
Stock market forecasting has been a quite popular challenge in machine learning researches. Most investors want to make decisions based on criteria that will provide greater returns in their operations. Recently, studies have been using Deep Learning techniques, such as Convolutional Neural Networks (CNN), to perform price regression or trade signal classification in financial market. In this work, a system architecture that uses a CNN model is proposed to perform the indication of the best operation for each moment in the stock market, this system was called CNN Trading Classifier (CNN-TC). This system consists of data pre-processing, classification by the model and decision making in the market. It was evaluated based on data from the Brazilian and American stock market in three different periods. For this, statistical evaluation was performed, using the metrics of accuracy, precision, recall and F1 classification, and financial based on the classifications performed by the model. In addition, a test on a simulated environment using the MetaTrader software was evaluated in order to attest to the effectiveness of this approach. The results show that the system had better statistical and financial results in most evaluations compared to the use of other Deep Learning models and overcame the strategy Buy and Hold (BH) and fixed income returns.