Comparação entre modelos Box & Jenkins, suavização e ensembles: um estudo para previsão do preço do petróleo bruto WTI

Abstract

This work investigated the use of a range of models aimed at forecasting WTI crude oil prices, emphasizing their importance to financial markets and the global economy. The main objective was to develop predictive models using time series analysis techniques, such as Autoregressive (AR), Autoregressive Moving Average (ARMA), and Autore- gressive Integrated Moving Average (ARIMA), in addition to ARMA variants adjusted by Genetic Algorithms (ARMA-GA) and Particle Swarm Optimization (ARMA-PSO). Expo- nential smoothing techniques, including SES, Holt, and Holt-Winters, in both additive and multiplicative forms, were also addressed. The models were integrated using en- semble techniques, through mean, median, Moore-Penrose pseudo-inverse, as well as weighted averages with GA and PSO. The adopted methodology included preprocessing that applied techniques to ensure data stationarity, essential for reliable modeling. The results indicated that for one-step-ahead forecasts, the weighted average ensemble with PSO outperformed traditional models in terms of MSE, MAE, and MAPE. For multi- step forecasts (3, 6, 9, and 12), the ensemble with the Moore-Penrose pseudo-inverse showed better results. This study demonstrated the effectiveness of combining predictive models to forecast future WTI crude oil prices, providing a useful tool for analysis and applications. Thus, it becomes evident that it is possible to expand the application idea of linear nature models. It is recommended to explore these techniques in other variables to assess the robustness of ensemble methods and to introduce nonlinear models for evaluation.