基于Stacking集成的上市公司债券违约预测

打开文本图片集

中图分类号：TP39；F832.5 文献标识码：A文章编号：2096-4706（2025）20-0171-07

Abstract： With theprosperityand developmentofthe bond market，creditrisk emerges，andbond defaults occur frequently. This papertakesthebonds issuedbylistedcompaniesfromJanuary2014toDecember2O23astheresearchobject，selects indicators from the financialand non-fianciallevels，selects the Stacking ensemble framework from seven Machine Leaming models，andestablishesabonddefault prediction modelTheresultsshowthat thefour modelsofRF，GBDT，XGBoostand LightGBM have beter prediction efectsonbond defaultsof listedcompanies.Among them，the predictionaccuracyof GBDT is 95.4% ，andtheclosertothe timeofdefault，thebeterthemodel fiting effect.Theprediction performanceofthe Stacking ensemble model is significantly improved，and the AUC value is increased by 0.7%～11.3% .Cash ratio，financial expense ratio， total issuance and coupon rate have a greater impact on bond default.

Keywords：bond default;MachineLearning; Stacking;ensemble algorithm; SHAP

0 引言

近年来，我国证券市场发展良好，债券市场是企业和政府融资的重要途径，在促进经济发展和提升资本市场活跃度方面发挥了重要作用。（剩余9249字）