Enhanced futures price-spread forecasting based on an attention-driven optimized LSTM network: integrating an...

Area of Science:

• Computational Finance
• Machine Learning
• Financial Econometrics

Background:

• Financial market prediction is challenging due to complex temporal dependencies and heterogeneous data in futures price-spreads.
• Traditional machine learning and standard Long Short-Term Memory (LSTM) models exhibit limitations in pattern mining and hyperparameter optimization.

Purpose of the Study:

• To propose an Improved Grey Wolf Optimizer with Multi-headed Self-attention and LSTM (IGML) model for enhanced futures price-spread prediction.
• To improve feature interaction and automate hyperparameter tuning for financial time-series forecasting.

Main Methods:

• Integration of a multi-head self-attention mechanism to improve feature interaction within the LSTM framework.
• Development of an improved grey wolf optimizer (IGWO) with four enhancements for automated hyperparameter selection.
• Validation of IGWO's convergence efficiency on benchmark optimization problems.

Main Results:

• The IGWO algorithm demonstrates superior convergence efficiency in optimization tasks.
• The IGML model significantly reduces prediction errors on real futures price-spread datasets.
• Achieved reductions in Mean Squared Error (RMSE) by up to 88% and Mean Absolute Error (MAE) by up to 85% compared to baseline models.

Conclusions:

• The proposed IGML model effectively captures intricate financial market dynamics.
• IGML offers a significant advancement over traditional methods for futures price-spread prediction.
• The enhanced optimization and attention mechanisms contribute to superior forecasting performance.