This study aims to examine how multifactorial sources of uncertainty – including geopolitical tensions, energy price volatility, and climate variability – jointly influence the dynamics of global maritime freight markets. It further evaluates the predictive and portfolio performance of two competing learning frameworks, Empirical Risk Minimization (ERM) and Invariant Risk Minimization (IRM), in modeling freight rate behavior across the Baltic Exchange indices (BAID, BADI, BAIT).
Using daily data from 2015 to 2025, the study integrates financial, geopolitical, and climate variables to construct regime-dependent environments. A hybrid regime-switching framework is implemented, where ERM and IRM models are alternately applied under calm and stress conditions. Predictive accuracy is assessed via MAE, RMSE, and Rˆ2, while portfolio robustness is evaluated through Sharpe, Sortino, and Maximum Drawdown metrics. SHAP (SHapley Additive exPlanations) analysis provides model interpretability and identifies key drivers of freight rate forecasts.
Results reveal that ERM outperforms IRM in calm environments characterized by market stability, while IRM demonstrates superior robustness and generalization during geopolitical or climate-induced stress periods. SHAP analysis highlights near-surface temperature (T2M) and relative humidity (RH2M) as dominant predictors across all freight segments, underscoring the growing importance of climate factors in maritime risk modeling. The proposed hybrid regime-switching strategy combining ERM and IRM yields improved inter-regime stability and risk-adjusted performance.
The study contributes to the literature on robust financial forecasting by introducing invariant learning into maritime and energy economics. It provides an empirical stress test of predictive stability under compound shocks, highlighting the relevance of regime-aware learning for modeling structural instability in freight markets.
Findings offer valuable insights for investors, policymakers, and risk managers seeking to enhance portfolio resilience in the face of climate and geopolitical disruptions. The regime-switching framework can be operationalized in stress-testing exercises, dynamic asset allocation, and risk management decisions by aligning model choice with prevailing uncertainty regimes.
This paper is among the first to apply the Invariant Risk Minimization framework to maritime and energy finance. By jointly incorporating climate, geopolitical, and financial risks within an interpretable machine learning setting, it advances understanding of how invariant causal features can improve forecasting robustness and sustainability-oriented decision-making in global shipping markets.
