Introducing the Synergy Based Approach for Forecasting the Crude Oil Prices with Traditional and Machine Learning Econometric Models

Crude oil plays a pivotal role in the global economy, influencing inflation, trade balances, and energy security. Accurate forecasting of crude oil prices is therefore essential for policymakers and market participants. This study proposes a hybrid forecasting framework that synergizes conventional econometric methods with machine learning (ML) techniques. First, the time series is decomposed using Ensemble Empirical Mode Decomposition (EEMD) to isolate intrinsic mode functions (IMFs). These components are then classified into deterministic and stochastic elements via spectral analysis. Second, traditional models such as ARIMA and GARCH are applied to the relevant IMFs, while advanced ML models (LSTM and XGBoost) are fitted to both original and residual series. Finally, a synergy model combines econometric and ML outputs, with Bayesian optimization applied for hyperparameter tuning. Model performance is assessed using key error metrics such as Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE). The findings suggest that hybrid models integrating conventional econometric methods with machine learning approaches, optimized through Bayesian techniques, achieve superior forecasting accuracy compared to standalone models. Additionally, the Diebold-Mariano (DM) test confirms that these synergy-based models offer the most reliable predictions for crude oil prices.