Abstract: The prediction and early warning of corn futures prices can help guide the high-quality development of the agricultural economy. Since June 2020, the corn futures prices have fluctuated violently, and accurate and efficient corn futures price forecasting methods are urgently needed. Given the problem that the existing researches do not fully consider the pandemic situation, policy regulation, and potential forecast information in news texts, this research, based on both qualitative and quantitative data, proposes an effective forecast framework for corn futures price interpretability, which takes multiple factors into consideration, such as the supply and demand relationship of the corn market, policy adjustments, international market shocks, epidemic shocks, the impact of emergencies and other factors that lead to the fluctuation of corn prices. At the same time, aiming at the problem of insufficient interpretability of existing corn futures price prediction, a novel DE-TFT interpretable corn futures price prediction model is proposed. The differential evolution algorithm is used to efficiently optimize the parameters of the Temporal Fusion Transformers (TFT). TFT is a novel attention-based deep learning model that combines high-performance forecasting with temporal dynamic interpretable analysis, showing excellent performance in forecasting research. The TFT model can produce interpretable corn futures price prediction results, including attentional analysis of time steps and importance ranking of input variables. In the empirical study, the latent dirichlet allocation topic model is used to analyze the content and topics of corn news information and policy adjustments collected by “China Grain Network”, and the CNN classification model is used to extract the potential prediction information of news information. The interpretable experimental results show that the introduction of the Baidu index “pandemic”, which reflects the domestic epidemic situation and the quantified corn news text features, can further improve the accuracy of corn futures price prediction.

Key words: corn futures price, time series forecasting, interpretable neural networks, text mining, deep learning