Ensaios sobre o uso de redes neurais na previsão de taxa de câmbio

• Ano de defesa: 2020
• Autor(a) principal: Costa, Marisa Gomes da
• Orientador(a): Basso, Leonardo Fernando Cruz
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Presbiteriana Mackenzie
• Palavras-chave em Português: confidence set model; aprendizagem de máquina; LSTM; redes neurais recorrentes; temas emergentes
• Área do conhecimento CNPq: CNPQ::CIENCIAS SOCIAIS APLICADAS::ADMINISTRACAO::ADMINISTRACAO DE EMPRESAS
• Link de acesso: http://dspace.mackenzie.br/handle/10899/26467

This study aims to compare and evaluate the predictive power of artificial neural network models on exchange rates. Initially, a bibliometric study and literature review is carried out in order to identify the current research status in the area. Then, an empirical study is propesed to forecast various Exchange rates using data of opening, closing,high and low in daily frequency. The data sample includes exchange rates (BRL / USD, EUR / USD and GBP / USD) from January 2014 to December 2019. Forecasts are made for a period ahead. Different architectures of the LSTM recurrent neural network model were tested. To rank the models in terms of predictive power, the results of the predictions are compared to the prediction of the random walk model, using it as a benchmark, as well as ARIMA. The selection of models is made by the model confidence set (MCS). Lunde and Nason. The results indicated that the LSTM model is superior to the random walk and ARIMA for all analyzed currencies.