When Does Online Learning Help? Forecast Horizon Effects in Agricultural Commodity Price Forecasting
EGE 15th INTERNATIONAL CONFERENCE ON APPLIED SCIENCES, İzmir, Türkiye, 12 - 14 Haziran 2026, cilt.1, ss.416-426, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Cilt numarası: 1
- Basıldığı Şehir: İzmir
- Basıldığı Ülke: Türkiye
- Sayfa Sayıları: ss.416-426
- Açık Arşiv Koleksiyonu: AVESİS Açık Erişim Koleksiyonu
- Çukurova Üniversitesi Adresli: Evet
Özet
Accurate forecasting of agricultural commodity prices is essential for production planning, market stability, and food security. Although online learning methods have gained attention for their ability to adapt to changing market conditions, their effectiveness across different forecasting horizons remains unclear. This study examines the impact of forecast horizon on the performance of classical and online learning approaches in agricultural commodity price forecasting. Daily price data from 2013 to 2021 were obtained from a dataset containing 132 agricultural commodities. The 20 commodities with the highest number of observations were selected for analysis. Three forecasting methods were compared: Ordinary Least Squares (OLS), Random Forest (RF), and Sliding-Window Recursive Least Squares (SW-RLS). Models were built using lagged price variables and evaluated for 1-day, 7-day, and 30-day forecasting horizons. Performance was assessed using the coefficient of determination (R²), Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE). The results show that forecast horizon strongly influences model performance. OLS achieved the highest predictive accuracy for all commodities in both 1-day and 7-day forecasts. However, for the 30-day horizon, SW-RLS outperformed competing methods for several commodities, particularly onion and cauliflower, where local price dynamics changed more rapidly over time. These findings indicate that online learning methods do not universally outperform traditional models; instead, their benefits depend on the forecasting horizon and the characteristics of the underlying price series.