In recent years, rare events such as the COVID-19 pandemic and the Russia–Ukraine war have caused disruptions in global agricultural commodity supply chains, especially through supply and demand imbalances. As a result, unusual price fluctuations have been observed in commodity markets. In such periods, accurate modeling and forecasting of agricultural commodity prices is important to keep the agricultural sector sustainable for both national economies and individual investors. In this study, the performance of several methods used to model financial time series with rare and low-probability events is compared. These methods include the Generalized Autoregressive Conditional Heteroskedasticity (GARCH) model, the State Space Model (SSM), Long Short-Term Memory (LSTM) neural networks, and hybrid GARCH-LSTM models. The analysis uses weekly global agricultural commodity prices for wheat, sugar, cocoa, coffee, cotton, corn, soybean, and oats from March 2001 to December 2025. The results show that the State Space Model demonstrates significantly superior performance in both price modeling and future forecasting.
Neslihanoglu et al. (Wed,) studied this question.