Scholarlink

<p>China’s local government financing platforms (LGFPs) faced a debt burden of RMB 60 trillion by the end of 2023, intensifying credit risk uncertainty&#xa0;and threatening economic and social stability. To address this challenge, this paper proposes a credit risk assessment method for LGFPs based on a PSO-XGBoost-SHAP framework: particle swarm optimization (PSO) is employed for hyperparameter tuning, extreme gradient boosting (XGBoost) is applied for risk classification, and SHAPley Additive exPlanations (SHAP) is utilized to interpret model outputs. Within this framework, thirty-one risk indicators are constructed from local government financial risk, LGFP credit risk, and government support, thereby extending existing assessment systems. Furthermore, a total of 3876 LGFPs are identified in accordance with official guidelines, and multi-source information is integrated for modeling. The results show that: (1) the proposed model demonstrates strong stability and generalization capability, outperforming benchmark models such as support vector machine, logistic regression, and random forest in terms of the AUC metric; (2) the contributions to LGFP credit risk, in descending order of importance, are regional fiscal risk, LGFP credit risk, and government support; (3) high credit risk among LGFPs is concentrated in two types of regions: provinces with weak economies and high debt levels (Guizhou, Yunnan) and economically strong provinces with prominent implicit debt (Henan, Jiangsu), demonstrating significant regional heterogeneity. Policy recommendations are provided to help mitigate LGFP credit risk.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Multi-source information fusion for credit risk assessment of local government financing platforms using PSO-XGBoost and SHAP

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月19日

查看全文
  • Liang Xu,
  • Debin Fang

摘要

China’s local government financing platforms (LGFPs) faced a debt burden of RMB 60 trillion by the end of 2023, intensifying credit risk uncertainty and threatening economic and social stability. To address this challenge, this paper proposes a credit risk assessment method for LGFPs based on a PSO-XGBoost-SHAP framework: particle swarm optimization (PSO) is employed for hyperparameter tuning, extreme gradient boosting (XGBoost) is applied for risk classification, and SHAPley Additive exPlanations (SHAP) is utilized to interpret model outputs. Within this framework, thirty-one risk indicators are constructed from local government financial risk, LGFP credit risk, and government support, thereby extending existing assessment systems. Furthermore, a total of 3876 LGFPs are identified in accordance with official guidelines, and multi-source information is integrated for modeling. The results show that: (1) the proposed model demonstrates strong stability and generalization capability, outperforming benchmark models such as support vector machine, logistic regression, and random forest in terms of the AUC metric; (2) the contributions to LGFP credit risk, in descending order of importance, are regional fiscal risk, LGFP credit risk, and government support; (3) high credit risk among LGFPs is concentrated in two types of regions: provinces with weak economies and high debt levels (Guizhou, Yunnan) and economically strong provinces with prominent implicit debt (Henan, Jiangsu), demonstrating significant regional heterogeneity. Policy recommendations are provided to help mitigate LGFP credit risk.