Nonlinear Dynamics in Export Payment Risk: A Gradient Boosting and SHAP-Based Explainable AI Model

Abstract

Timely repatriation of export value after export is a global challenge. Some determinants such as the payment behavior of the importer, the liquidity and health of the foreign bank, and the country risk of the importer can provide early signals about whether the export value will be repatriated on time. In this study, we have classified how the export value is repatriated based on some historical data such as on-time, early, and delayed. Then, we have looked at the scores using machine learning algorithms such as Logistic Regression, Random Forest, Gradient Boosting Machine (GBM), and Artificial Neural Network (ANN). In the study, we found empirical superiority of GBM, which achieved a macro-averaged AUC-ROC of 0.915 and a Recall of 85.4%. According to the SHAP (SHapley Additive exPlanations) predictor, 55.3% of it depends on the payment behavior of the importer. The study also shows that there is a significant non-linear behavior pattern between non-compliant bank profiles and high country risk of the importer. By adding Explainable AI (XAI), this model can be made more transparent and interpretable supporting practical, risk-based decisions such as adjusting pre-shipment finance limits, determining post-shipment tenors, or initiating early export bill lien actions. As a result of this study, banks can be proactive in export bill liens, pre-shipment finance, post-shipment tenors and maintain compliance easily.