Previsão do preço do petróleo utilizando modelos lineares, não lineares e ensembles

• Ano de defesa: 2024
• Autor(a) principal: Matos, Camila
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Ponta Grossa; Brasil; Programa de Pós-Graduação em Engenharia de Produção; UTFPR
• Palavras-chave em Português: Petróleo - Derivados - Preços; Preços - Determinação; Modelos matemáticos; Petroleum products - Prices; Basing-point system; Mathematical models; CNPQ::ENGENHARIAS::ENGENHARIA DE PRODUCAO; Engenharia/Tecnologia/Gestão
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/35724

Crude oil is the primary material for fuel and chemical products and is considered a vital mineral for the economy, remaining dominant at least for the next decade. One type of crude oil used as a benchmark in determining oil prices is West Texas Intermediate (WTI), as it influences the movement of other crude oil prices. As the factors affecting international oil prices become increasingly nonlinear, predicting oil prices become challenging. This study emerged from identifying gaps in the existing literature, where the steps for forecasting oil prices were not clearly addressed, and there was uncertainty about the amount of data required and the effectiveness of handling large volumes of data due to computational demands. To address these gaps, the aim was to develop a methodology for forecasting crude oil prices, employing a methodological approach that included a systematic literature review and the application of Design Research Methodology (DRM) steps. To achieve the objectives, 13 forecasting models were tested on two data series: the full series and the broken series. Price data were extracted from the US Energy Information Administration database and analyzed using Python and R. The LSTM, AR, and Ensemble 4 models achieved the best results, with minimal differences between the metrics of the two tested series. The proposed methodology offers a detailed and adaptive approach, covering everything from data collection to the analysis of external variables and structural breaks. Key limitations include reliance on historical data and the high computational demands of the models, which may limit real-time application. Future research should incorporate additional external data, optimize the parameters of nonlinear models, and test the proposed methodology in different commodity markets to validate and adapt the models to new economic contexts.