Forecasting Prices of Strategic Food Products: Time Series Models in the Context of Post-Crisis Market Disruption

In a post-crisis context, anticipating variations in raw material prices is essential for smart cities. At the beginning of this decade, two major global crises disrupted various market sectors: the COVID-19 pandemic and the Russian-Ukrainian war. Wheat is an essential product that contributes to food security in many countries. The price of this cereal fluctuated considerably during the two crises. Russia and Ukraine account for up to 28% of the wheat export market. Today, the wheat market seems to be gradually stabilizing from these disturbances. This paper explores reliable forecasting models to estimate wheat prices and support decision-making in the post-crisis period. To this end, two time-series forecasting techniques are compared: the Autoregressive Integrated Moving Average Model (ARIMA), known for its reliability in financial forecasting, and the Long Short-Term Memory Network (LSTM). The results reveal that the LSTM model achieves an accuracy of 97.49%, outperforming the ARIMA model, which registers an accuracy of 77%. These results underline the effectiveness of deep learning models for forecasting during periods of high price volatility, compared with conventional models.