Yi Wang
Against the backdrop of expanding cross-border food trade and constrained inspection resources, identifying risk signals from massive customs records has become a core task for intelligent border supervision. This study constructs unbalanced panel data covering HS8 products, trade partners and monthly records using multi-source trade and detention data, and adopts PPML-HDFE estimation to separately examine driving factors for two categories of detention risks: procedural compliance failures and intrinsic safety defects. Empirical outcomes show obvious heterogeneity in influencing factors between the two risk types. Past detention records serve as the most stable risk indicator, with stronger persistence for safety-related violations. Abnormally low prices display a consistent positive correlation with procedural non-compliance yet offer limited explanatory power for safety risks. Cumulative long-term import volume mitigates detention probabilities, reflecting risk reduction brought by mature supply chains. The above conclusions remain robust after multiple robustness checks and endogenous bias treatments, and differ across food categories and economic levels of exporting economies. This research delivers quantitative evidence for classified risk warning and optimized allocation of inspection resources under intelligent customs frameworks.