RV-FELM: Futures commodity price forecasting based on RIME-VMD algorithm coupled with FA-ELM

Area of Science:

• Financial Engineering
• Computational Economics
• Machine Learning

Background:

• Commodity futures are vital for material trade hedging.
• Accurate price prediction enables informed production and consumption decisions for countries and firms.

Purpose of the Study:

• To introduce a novel machine learning ensemble method for commodity futures price prediction.
• To enhance forecasting accuracy by combining decomposition and physical optimization algorithms.

Main Methods:

• Variational Mode Decomposition (VMD) optimized by the Rime Optimization Algorithm (RIME).
• Prediction of trend and seasonal terms using Extreme Learning Machines (ELM) and Fourier Attention (FA) models.
• Synthesis of prediction results for final commodity futures price forecasting.

Main Results:

• Achieved low Mean Absolute Percentage Errors (MAPE) for crude oil prices: 0.48% (1-step), 0.66% (3-step), 0.75% (6-step).
• Achieved low MAPE for soybean prices: 0.22% (1-step), 0.27% (3-step), 0.37% (6-step).
• Outperformed benchmark models in horizontal and directional accuracy, demonstrating robustness.

Conclusions:

• The proposed ensemble model effectively captures time and frequency domain characteristics of commodity futures series.
• The method offers superior accuracy and robustness in predicting crude oil and soybean futures prices.
• This approach provides a valuable tool for hedging and decision-making in commodity markets.