Purpose This study aims to develop advanced artificial intelligence methods for predicting agricultural prices by incorporating a diverse range of risk, green and behavioral factors such as the Bloomberg Agriculture Index and the S&P 500, the Geopolitical Risk Index, the Financial Stress Index, the S&P Green Index and the Twitter-based uncertainty index. Design/methodology/approach Using daily data spanning from July 2, 2014, to March 25, 2024, we forecast the prices of ten key agricultural commodities significantly impacted by recent global crises. To achieve this, we propose a forecasting model based on Long Short-Term Memory networks, Gated Recurrent Units (GRUs) and Recurrent Neural Networks (RNNs), which are optimized using a genetic algorithm. Findings The results of studying the importance of predictive factors using SHapley Additive exPlanations (SHAP) analysis highlight key factors such as the S&P 500 and Bloomberg Agriculture indices, along with emerging variables like green bonds and the Ukrainian PFTS Stock Index (PFTS). Moreover, empirical findings underscore the relevance of deep learning (DL) models, particularly GRU, in forecasting agricultural commodity prices. The proposed DL models demonstrate superior performance compared to traditional approaches, including bootstrapping techniques. Originality/value The findings provide a more comprehensive understanding of agricultural price dynamics, highlighting key factors that influence the prices of commodities such as canola, cotton, dairy, maize, sugar, cocoa, coffee, rice, wheat and soybean. This paper makes a significant contribution to the field of agricultural commodity forecasting by addressing the challenges of complex price trajectories and achieving exceptionally low error rates, offering critical insights for managing risks and optimizing financial strategies in volatile markets. Empirical results provide valuable tools for guiding the restructuring of financial policies in agricultural markets to address the challenges of a dynamic global environment.
Drira et al. (Sun,) studied this question.