Leveraging Automatically Optimized Forecasters and Large Language Model for Predicting Vietnamese Rice Export Price

This study evaluates the reliability of Vietnamese rice export price forecasting through an end-to-end framework, emphasizing systematic ablation studies across statistical and machine learning models. Input combinations range from univariate price data to macroeconomic indicators, sentiment signals derived via Large Language Models, and their integration. Results demonstrate that gradient boosting methods, particularly CatBoost, consistently outperform statistical models. Notably, sentiment-based models achieve performance comparable to those using full macroeconomic data. The integration of sentiment and macroeconomic features yields the most reliable forecasts, validating the role of sentiment signals in enhancing economic forecasting.