Real-Time Trade Data Improves Supply Chain Disruption Forecasting

IntroductionMaritime-dependent economies face a supply chain monitoring problem that has become a first-order policy concern under geopolitical uncertainty as customs administrations and supply chain authorities must allocate limited analytical capacity across thousands of imported product lines before disruptions surface. The current screening practice ranks products on origin concentration; however, the predictive validity of this heuristic has rarely been evaluated out of sample.MethodsThis paper reframes vulnerability assessment as a prediction problem using a large-scale panel of 5,925 Harmonized System (HS) six-digit products constructed from Korea Customs Service open statistics and observed monthly over 2010–2025 across Korea's 20 largest import partners, yielding roughly 161,000 unique out-of-sample product–month observations and 1.3 million model and benchmark scores. Proxy-defined disruption onsets were identified by a persistent quantity shortfall accompanied by rising unit values, each standardized by the product's own history, an operational signature designed to distinguish supply-side distress from demand contractions that yields 440 events. Gradient boosted ensembles and a logistic benchmark were evaluated under walk-forward validation with a confirmation-lag embargo against concentration rankings of the kind used in first-stage vulnerability screening.ResultsWithin the top 5% of monthly watchlists, the machine learning rankings attained precision approximately 1.7 times the 12-month target base rate and captured 31% of evaluable onset events, with a median first-alert lead of 9 months. Their watchlist precision advantage over the concentration rankings was statistically significant in paired annual comparisons and under moving-block bootstrap inference, while the concentration ranking itself performed below the base rate at the same budget, with the 12-month target rates showing no clear monotonic gradient across the concentration distribution.DiscussionConcentration is better interpreted as structural exposure than as a stand-alone ranking of short-horizon disruption risk, and customs-based dynamic indicators offer an early warning layer that can complement port and shipping information in the governance of maritime supply chains.