Forecasting the Istanbul Freight Index: A Comparative Analysis of SARIMA, XGBoost, and Facebook Prophet Across Multiple Time Horizons

Accurate freight rate forecasting is of critical importance for maritime stakeholders in making informed decisions amid volatile market conditions. This study evaluates the performance of three forecasting models, Seasonal Autoregressive Integrated Moving Average (SARIMA), Facebook Prophet, and Extreme Gradient Boosting (XGBoost), in predicting the Istanbul Freight Index (ISTFIX), a key benchmark for small coaster and mini-bulk carrier markets across Black Sea and Mediterranean routes. Weekly ISTFIX data covering the period from January 7, 2008 to March 17, 2025 are employed in the analysis. A rolling-window split strategy is applied to test forecast accuracy across four time horizons: 1-month, 3-month, 6-month, and 1-year. Model performance is evaluated using MAPE, MAE, and RMSE. The results demonstrate that XGBoost achieves superior short-term accuracy with a MAPE of 6.12% at the 1-month horizon, maintaining near-reasonable performance at longer horizons. SARIMA performs consistently across all horizons with MAPE values ranging from 7.16% to 26.63%. Prophet, by contrast, fails to capture the memory-dependent and volatile nature of ISTFIX, recording a MAPE of 50.09% even at the 1-month horizon. The findings demonstrate that while traditional trend-based models struggle with the extreme volatility of freight markets, tree-based machine learning algorithms excel at capturing complex, non-linear price movements.