Modeling Grain Futures Prices Through Uncertainty Indices and Mixed‐Frequency Fusion: An Interpretable Deep Learning Framework

This study innovatively develops an interpretable mixed‐frequency feature interaction deep learning network (IMF‐FIDNet) to improve high‐frequency grain futures price prediction via effective multi‐frequency data integration, with a focus on ensuring robustness amid market uncertainty. By refining advanced mixed‐frequency processing methods, proposing a new deep learning model, and integrating multiple modules, IMF‐FIDNet enhances feature interaction modeling between low‐frequency uncertainty indicators and high‐frequency grain prices. Experiments show it outperforms traditional models in accuracy and robustness, and effectively supports investment decisions; further, its interpretability quantifies uncertainty indices' contributions, confirming macro‐indicators' role in high‐frequency price forecasting.