A comprehensive analysis of Brent Crude oil forecasting methods combining machine learning and quantitative models with application

Abstract

This study utilizes time series analysis and machine learning techniques to forecast Brent Crude oil prices. The forecasting models include an ARIMAX model, three machine learning models (GRU, LSTM, CNN) and an ARIMAX model augmented with LASSO regularization.

The findings indicate that the ARIMAX model exhibits the best forecasting performance; however, it is prone to overfitting. To address this issue, LASSO regularization is applied to the ARIMAX model to penalize complexity. Surprisingly, incorporating LASSO regularization results in reduced forecasting performance compared to the initial ARIMAX model.

Among the basic machine learning models, the GRU model demonstrates highest predictive accuracy, followed by the LSTM model, while the CNN model exhibits lower predictive accuracy. When adding a dropout term, we find that the ranking order changes, and the CNN exhibits highest predictive accuracy. Further, when generalizing the models using cross-validation, we find that LSTM exhibits the best overall forecasting performance among the machine learning models.

As an extension of our main research, we seek to use machine learning models to predict out-of-sample Brent Crude oil price and evaluate its impact on the valuation of Aker BP and Vår Energi. When applying the forecasted Brent Crude oil prices obtained from the Prophet model, the findings reveal that both companies are undervalued relative to their current market values.

These results underscore the significance of accurate forecasting models in informing investment decisions and highlight the potential undervaluation of the companies analyzed.