Predicting commodity prices amidst Russia-Ukraine war: Some evidences from artificial intelligence models

• Forecasts align with real prices, indicating existence of different distinguishable phases in both cycles. • AI models capture nonlinear trends of the commodity prices cycle during crisis and uncertainty. • Political instability leads to bearish phases, well predicted by AI models. • AI models help assess and mitigate risks associated with geopolitical uncertainty. • DBN outperforms GRU and LSTM in forecasting commodity prices. This study aims to predict commodity prices amidst the Russia-Ukraine conflict using Artificial Intelligence (AI) models. Specifically, deep learning model Gated Recurrent Unit (GRU) and machine learning model Dynamic Bayesian Networks (DBN) are employed to forecast prices of crude oil, natural gas, wheat, and gold. Historical data during the period 16/06/2018–15/09/2023 is divided into training and testing sets. Learning is carried out over the period before the war whereas a testing is carried out over the war period to find out the most effective model in predicting the price of commodities during crises. The predictability accuracy of the two models is compared using different performance criteria. The empirical results show that the forecasted values obtained from the DBN model seem to more successfully mimic the commodity price dynamics and therefore markedly closer to the actual evolution of commodity market. The DBN model outperforms the GRU model in forecasting commodity prices. These findings underscore the practical utility of AI forecasting models in navigating volatile commodity markets during geopolitical turmoil, providing valuable insights for investors and policymakers alike.