Forecasting Corn Futures Prices Using the LSTM Model

The central government's No. 1 document introduced new requirements for “maintaining the stability of agricultural product prices.” However, recent years have seen significant fluctuations in corn prices in China, highlighting the importance of exploring the trends of corn futures prices for stabilizing spot corn prices. Given the increasingly complex financial market environment, traditional linear econometric methods and machine learning approaches face inherent limitations in price forecasting. This study utilizes extensive data and applies the Long-Short Term Memory (LSTM) deep learning model to forecast corn futures prices. During the model training process, historical corn futures data were used as hidden input feature variables, along with additional indicators such as spot corn prices, real-time quotes from corn deep processing enterprises, downstream product prices in the corn industry chain, prices of similar substitutes, and the Baidu Search Index for “corn prices.” The model was trained and fine-tuned with hyperparameters to assess its predictive accuracy under different time windows, learning rates, and iteration counts. Experimental results indicate that incorporating external influencing factors significantly enhances the model's predictive accuracy compared to using only historical price data. The model with a learning rate of 0.01, 50 iterations, a time window of 5, and 4 hidden layers achieved the highest accuracy. The findings demonstrate that integrating external factors and optimizing hyperparameters substantially improve the LSTM model's predictive precision, outperforming models that rely solely on historical data. The study provides policy recommendations on promoting advanced technology, enhancing data collection and information sharing, and improving market transparency and fairness. Additionally, the conclusion discusses the extensive application potential and promising development prospects of the LSTM model in corn futures price forecasting. This model enables market participants to obtain more accurate market trend forecasts, facilitating more informed investment decisions. The research also offers new insights and methodological references for scholars in financial forecasting, contributing to the further advancement of this field.Funding: This study was supported by the 2024 Key Research Project of Higher Education Institutions of the Henan Provincial Department of Education Project Number 24B790006, the Henan Provincial Department of Education Project Number 2021GGJS120, and the Henan Provincial Department of Education Project Number 2021SJGLX478. Additionally, this study received support from the Xinxiang City Federation of Social Sciences Project Number SKL-2024-157.