Abstract: The operational risk of foreign trade enterprises is increasing under the context of external demand contraction and intensive protectionism, leading to a greater risk of financial crisis. In response to the challenge of low accuracy in predicting financial crises for foreign trade enterprises, the early warning indicator system is optimized, and a combined model based on SMOTE-XGBoost is proposed. Firstly, a financial crisis early warning indicator system is established by integrating financial indicators and macro foreign trade indicators. Secondly, a combined model integrating synthetic minority over-sampling technique (SMOTE) and extreme gradient boosting algorithm (XGBoost) is constructed to analyze data from foreign trade listed enterprises in China. The results show that this combined model can achieve more accurate prediction and better overall stability than other models, with superior ACC, recall, F1-score, and AUC. This model can be used to assist foreign trade enterprises in proactively identifying potential financial risks and avoiding falling into financial crises.

Key words: financial crisis early warning, foreign trade listed companies, imbalanced data, XGBoost

摘要： 外需萎缩和保护主义加剧大大提高了外贸企业的经营风险，外贸企业陷入财务危机的风险加大。针对外贸企业财务危机预警准确率不高的难题，优化预警指标体系，并提出基于SMOTE-XGBoost的组合模型。建立融合财务指标和宏观外贸指标的财务危机预警指标体系；构建合成少数类过采样技术（SMOTE）和极限梯度提升算法（XGBoost）的组合模型，对我国外贸企业数据进行分析。研究发现SMOTE-XGBoost组合模型能够有效提高预测准确率，ACC、recall、F1-score、AUC值均优于其他模型，且具有良好的稳定性。该模型能够帮助外贸企业提前发现可能的财务风险，避免陷入财务危机。

关键词: 财务危机预警, 外贸企业, 不平衡数据, XGBoost