A carregar...
Publicação
Previsão de séries temporais financeiras: uma abordagem com Long Short-term Memory Deep Neural Networks
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 3.91 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
O intuito desta dissertação é identificar o método que proporciona resultados mais fidedignos na previsão de séries temporais financeiras, de entre os modelos tradicionais e as novas metodologias de Machine Learning. A fim de alcançar esse objetivo, foi considerada uma base de dados com 4842 valores diários do fecho do índice bolsista Standard & Poor’s 500 (SP500), no intervalo temporal compreendido entre 3 de janeiro de 2000 e 1 de abril de 2019, excluindo os fins- de-semana e feriados. Primeiramente, os dados foram trabalhados no programa Eviews, de forma a obter o modelo econométrico ARIMA mais adequado e através da metodologia de Box-Jenkins procedeu-se para a previsão da série financeira em estudo. Contudo, para obtenção de melhores resultados, a série foi reduzida, iniciando no dia 2 de março de 2009, devido à crise financeira de 2007-2008. Para comparação de modelos, foi realizada a previsão dos dados da bolsa SP500, através de um modelo Deep Neural Network – Long Short-Term Memory (código programado em Python). Conclui-se que os melhores resultados de previsão foram obtidos com os modelos de redes neuronais, tanto para curto como para longo-prazo.
The aim of this dissertation is to identify the method that provides the most reliable results in forecasting financial time series, between the traditional models and the new Machine Learning methodologies. In order to achieve this objective, were considered 4842 daily closing values of the Standard & Poor’s 500 (SP500) stock index, since January 3, 2000 until April 1, 2019, excluding weekends and holidays. First, we use Eviews software, in order to obtain the right ARIMA econometric model, and by using the Box-Jenkins methodology, we forecast the SP500 financial time series. However, for best results, the series was split, starting on March 2, 2009, due to the 2007-2008 financial crisis. For model comparison, was realized the prediction of the SP500 stock index data, using a Deep Neural Network - Long Short-Term Memory model (code programmed in Python). It is concluded that the best results were obtained with neural network models, for both short and long term forecast.
The aim of this dissertation is to identify the method that provides the most reliable results in forecasting financial time series, between the traditional models and the new Machine Learning methodologies. In order to achieve this objective, were considered 4842 daily closing values of the Standard & Poor’s 500 (SP500) stock index, since January 3, 2000 until April 1, 2019, excluding weekends and holidays. First, we use Eviews software, in order to obtain the right ARIMA econometric model, and by using the Box-Jenkins methodology, we forecast the SP500 financial time series. However, for best results, the series was split, starting on March 2, 2009, due to the 2007-2008 financial crisis. For model comparison, was realized the prediction of the SP500 stock index data, using a Deep Neural Network - Long Short-Term Memory model (code programmed in Python). It is concluded that the best results were obtained with neural network models, for both short and long term forecast.
Descrição
Tese de mestrado, Matemática Financeira, Universidade de Lisboa, Faculdade de Ciências, 2019
Palavras-chave
Série temporal financeira Mercados financeiros Standard & Poor’s 500 Machine Learning Deep Learning Long Short-Term Memory Modelos Autorregressivos de Médias Móveis Previsão Teses de mestrado - 2019