APPLICATION OF MACHINE LEARNING AND THE GEOPOLITICAL RISK INDEX IN FORECASTING THE VOLATILITY OF AGRICULTURAL COMMODITY FUTURES PRICES

Abstract

This study examines the influence of the Geopolitical Risk Index (GPR) on agricultural futures price volatility and compares the forecasting accuracy of statistical and machine learning models. Monthly data on corn and soybean futures prices were collected from the Chicago Mercantile Exchange (CME), Brasil Bolsa Balcão (B3), and Dalian Commodity Exchange (DCE) from June 2015 to May 2025. The models evaluated include SARIMAX, GARCHX, and LSTM neural networks. Exogenous variables incorporated are GPR, Geopolitical Threats (GPRT), Geopolitical Acts (GPRA), Country-Specific GPR (GPRC), and trading volume. Model performance is assessed using mean absolute percentage error (MAPE). Results indicate that LSTM models with exogenous variables are most effective at forecasting agricultural futures price volatility. For corn futures, LSTM (lookback = 12) provides the most accurate forecasts. The most effective exogenous variable is GPRC BRAZIL for CME, GPRC USA for B3, and GPRC RUSSIA for DCE. For soybean futures, LSTM (lookback = 6) exhibits the highest accuracy, with GPRA performing best for B3 and GPRC BRAZIL for DCE. Overall, the results demonstrate that integrating geopolitical factors with machine learning techniques effectively improves forecast accuracy of agricultural futures price volatility. This approach can be applied to risk management and hedging strategies in agricultural futures markets.