Explorando a Integração de Técnicas de Aprendizado de Máquina e Modelos Estatísticos na Previsão da Curva de Preços Futuros de Petróleo
N/D
Autores
Orientador
Co-orientadores
Citações na Scopus
Tipo de documento
Dissertação
Data
2024
Título da Revista
ISSN da Revista
Título do Volume
Resumo
Compreender as dinâmicas do mercado futuro de petróleo é de extrema importância para a
economia global, pois afeta produtores, investidores, formuladores de políticas econômicas e
acadêmicos. Neste contexto, o presente estudo abordou a modelagem da estrutura a termo dos
preços futuros do petróleo, questão vital para o mercado global de energia, para a
macroeconomia, para empresas e para investidores, concentrando-se no petróleo tipos Brent e
West Texas Intermediate (WTI). O objetivo deste estudo foi explorar a eficácia de uma
combinação de diferentes técnicas de modelagem na projeção da curva de preços do petróleo,
o que envolveu: (1) a decomposição da estrutura a termo em fatores distintos; e (2) a aplicação
de diversas metodologias na modelagem, desde as mais tradicionais até as técnicas de fronteira,
para estes fatores que são cruciais para determinar o nível e a forma da curva. O foco central
foi aplicar e comparar uma variedade de modelos em relação à sua eficácia da modelagem e à
projeção de preços na curva, buscando identificar a técnica mais eficiente, com base em sua
precisão e menor erro em previsões fora da amostra. As abordagens foram desde as técnicas
econométricas tradicionais, como Autoregressive Integrated Moving Average (ARIMA) e
Vector Autoregression (VAR), até métodos avançados de aprendizado de máquina e deep
learning, incluindo Least Absolute Shrinkage and Selection Operator (LASSO), Random
Forest, Extreme Gradient Boosting (XGBoost) e, especialmente, o modelo Long Short-Term
Memory (LSTM). O estudo foi conduzido, principalmente, com a metodologia de Dynamic
Nelson-Siegel (DNS) e, para decompor a curva de preços, seguimos o modelo apresentado por
Barunik e Malinska (2016). Os resultados ressaltaram a superioridade do LSTM em projeções
fora da amostra dos parâmetros obtidos pelo DNS, demonstrando sua eficácia na reconstrução
das curvas de preços dos petróleos tipos Brent e WTI. Esta capacidade do LSTM de capturar
complexidades temporais e dinâmicas nos dados financeiros é particularmente relevante no
mercado de petróleo, que é influenciado por uma variedade de fatores econômicos e
geopolíticos. Assim sendo, espera-se que este estudo contribua significativamente tanto com a
Academia quanto com o Mercado, no que tange à modelagem de preços de petróleo, marcando
um avanço na análise financeira, ao oferecer insights valiosos para investidores, analistas e
formuladores de políticas econômicas; aprimorando a compreensão das dinâmicas de preços do
petróleo; e combinando técnicas de fronteira na modelagem de preços futuros do petróleo.
Understanding the dynamics of the oil futures market is extremely important for the global economy, as it affects producers, investors, economic policy makers and academics. In this context, the present study addressed the modeling of the term structure of future oil prices, a vital issue for the global energy market, for the macroeconomy, for companies and for investors, focusing on Brent and West Texas Intermediate (WTI) oil types. The objective of this study was to explore the effectiveness of a combination of different modeling techniques in projecting the oil price curve, which involved: (1) the decomposition of the term structure into distinct factors; and (2) the application of various modeling methodologies, from the most traditional to frontier techniques, for these factors that are crucial for determining the level and shape of the curve. The central focus was to apply and compare a variety of models in relation to their modeling effectiveness and price projection on the curve, seeking to identify the most efficient technique, based on its accuracy and lowest error in out-of-sample forecasts. Approaches ranged from traditional econometric techniques, such as Autoregressive Integrated Moving Average (ARIMA) and Vector Autoregression (VAR), to advanced machine learning and deep learning methods, including Least Absolute Shrinkage and Selection Operator (LASSO), Random Forest, Extreme Gradient Boosting (XGBoost) and especially the Long Short-Term Memory (LSTM) model. The study was conducted mainly using the Dynamic Nelson-Siegel (DNS) methodology to decompose the price curve, following the model presented by Barunik and Malinska (2016). The results highlighted the superiority of LSTM in out-of-sample projections of parameters obtained by DNS, demonstrating its effectiveness in reconstructing Brent and WTI oil price curves. This ability of LSTM to capture temporal and dynamic complexities in financial data is particularly relevant in the oil market, which is influenced by a variety of economic and geopolitical factors. Therefore, it is expected that this study will contribute significantly to both the Academy and the Market, with regard to oil price modeling, marking an advancement in financial analysis, by offering valuable insights for investors, analysts and economic policymakers; improving understanding of oil price dynamics; and combining frontier techniques in modeling future oil prices.
Understanding the dynamics of the oil futures market is extremely important for the global economy, as it affects producers, investors, economic policy makers and academics. In this context, the present study addressed the modeling of the term structure of future oil prices, a vital issue for the global energy market, for the macroeconomy, for companies and for investors, focusing on Brent and West Texas Intermediate (WTI) oil types. The objective of this study was to explore the effectiveness of a combination of different modeling techniques in projecting the oil price curve, which involved: (1) the decomposition of the term structure into distinct factors; and (2) the application of various modeling methodologies, from the most traditional to frontier techniques, for these factors that are crucial for determining the level and shape of the curve. The central focus was to apply and compare a variety of models in relation to their modeling effectiveness and price projection on the curve, seeking to identify the most efficient technique, based on its accuracy and lowest error in out-of-sample forecasts. Approaches ranged from traditional econometric techniques, such as Autoregressive Integrated Moving Average (ARIMA) and Vector Autoregression (VAR), to advanced machine learning and deep learning methods, including Least Absolute Shrinkage and Selection Operator (LASSO), Random Forest, Extreme Gradient Boosting (XGBoost) and especially the Long Short-Term Memory (LSTM) model. The study was conducted mainly using the Dynamic Nelson-Siegel (DNS) methodology to decompose the price curve, following the model presented by Barunik and Malinska (2016). The results highlighted the superiority of LSTM in out-of-sample projections of parameters obtained by DNS, demonstrating its effectiveness in reconstructing Brent and WTI oil price curves. This ability of LSTM to capture temporal and dynamic complexities in financial data is particularly relevant in the oil market, which is influenced by a variety of economic and geopolitical factors. Therefore, it is expected that this study will contribute significantly to both the Academy and the Market, with regard to oil price modeling, marking an advancement in financial analysis, by offering valuable insights for investors, analysts and economic policymakers; improving understanding of oil price dynamics; and combining frontier techniques in modeling future oil prices.
Palavras-chave
Técnicas de Machine Learning; Modelos Estatísticos; Modelagem de preços futuros do petróleo; Petróleo tipos Brent e West Texas Intermediate; Long Short-Term Memory; Machine Learning Techniques; Statistical Models; Modeling future oil prices; Brent and West Texas Intermediate oil types; Long Short-Term Memory
Titulo de periódico
URL da fonte
Título de Livro
URL na Scopus
Idioma
Português
URL permanente
Notas
Membros da banca
Área do Conhecimento CNPQ
CIENCIAS SOCIAIS APLICADAS
CIENCIAS SOCIAIS APLICADAS::ECONOMIA
CIENCIAS SOCIAIS APLICADAS::ECONOMIA