Investigating and forecasting the impact of crop production shocks on global commodity prices

In this study, we investigate and forecast the impact of crop production shocks on the global prices of three major international agricultural commodities: maize, soybean, and cocoa. We perform a thorough assessment of the forecasting performances of five econometric and machine learning models using 60 years of data. First, we train the models on production and price data to forecast the monthly price variations for each crop separately considering different time horizons. Next, we implement a cross-validation procedure to identify the models with the most accurate forecasting ability for each crop. After choosing the best forecaster, we identify the most influential producing areas using several local and global model-agnostic interpretation tools. Our findings indicate significant differences among commodities in terms of prediction accuracy, with cocoa exhibiting a higher level of prediction error compared to less volatile markets like maize and soybean. Our results reveal a significant influence of Northern America’s maize and soybean production on the global prices of these commodities. The effects of production on prices are asymmetrical: small decreases in US production lead to substantial price increases, while small increases in production do not systematically decrease prices. In contrast, cocoa price variations are influenced by production coming from several regions, not from a single one.