Forecasting global trade networks under partial disruption: a network-attack framework and machine learning benchmark

The global economy observes shocks, such as oil price volatility, pandemics, and tariff disputes. This exposes the fragility of global trade and needs better forecast of disruptions in complex global trade network. This paper proposes a network-theoretic information processing framework that formalises unforeseen events as attack operators on a global trade network and categorises disruptions by the structural signatures they induce. Building on this representation, we introduce a new method (proposition and algorithm) that apply network evolution under shocks to predict (i) future bilateral trade (value and volume) and (ii) the dynamic status (network properties) of individual countries. The proposed method quantifies vulnerability and resilience over time using network analytics. It identified a distinct disruption profiles across major economies (e.g., Italy and Germany versus France, Spain, the U.S., and the U.K.). For predictive benchmarking, we compare the proposed network model against multiple machine-learning baselines; the network-based approach achieves more than 80% forecasting accuracy, while ANN and XGBoost achieve the strongest ML performance, with comparable results across periods. The study offers actionable, system-level insights for designing robust trade network structure and for evaluating policy interventions through an algorithmic lens, supporting decision-making at the intersection of computing and international economic networks.