Local, global and hierarchical forecasting of monthly maritime CO₂ emissions for 140 economies
140 ekonomi için aylık denizcilik CO₂ emisyonlarının yerel, küresel ve hiyerarşik tahmini
This paper is under review at a peer-reviewed journal. The journal name and full text will be added after acceptance.
Abstract
Carbon pricing of shipping requires timely national estimates of maritime emissions, yet no study has evaluated how to forecast the new monthly, AIS-based statistics that allocate ship emissions to economies. We compare local statistical models, global machine-learning models, their combinations and five reconciliation methods on 148 hierarchically organised series, using 13 rolling origins and horizons of 1–12 months. Automatic exponential smoothing and an equal-weight combination of exponential smoothing, ARIMA and Theta were the most accurate and best calibrated; global gradient boosting and pooled autoregression were significantly less accurate and their conformal intervals covered only 68–69% of outcomes at a nominal 95%. Reconciliation methods that ignore scale heterogeneity tripled bottom-level errors, whereas top-down proportions and shrinkage-based trace minimisation preserved accuracy. Simple local models remain the safe default for short, heterogeneous panels.
Keywords
Cross-learning; Forecast reconciliation; Exponential smoothing; Conformal prediction; Shipping decarbonisation; Gradient boosting; Rolling-origin evaluation
Open code and data
Loading code…