Scholarlink

<p>This study investigates the predictive power of implicit government guarantees (IGGs) in forecasting bond defaults among Chinese listed companies. We develop an innovative prediction framework that integrates IGG indicators with advanced machine learning algorithms (Random Forest and XGBoost) and compare their performance against traditional Logit models. Using a comprehensive dataset spanning 2011–2022, we incorporate macroeconomic variables, industry-specific factors, financial metrics, and governance characteristics. Our analysis reveals that (1) IGGs significantly reduce default probability for state-owned enterprises, with fiscal capacity serving as a critical boundary condition; (2) ensemble machine learning models substantially outperform Logit models, achieving 89.3% accuracy and 0.89 AUC; and (3) the interaction between ownership structure and IGG strength creates nonlinear risk patterns undetectable by conventional methods. We further demonstrate that incorporating IGG metrics improves prediction accuracy by 13.6% points compared to financial ratios alone. These findings provide actionable insights for regulators to reform bond pricing mechanisms and for investors to better assess China’s unique credit risk landscape. Our methodology offers a template for adapting default prediction models to institutional contexts where government intervention distorts market discipline. </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

Bond Default Prediction of Chinese Listed Companies: The Role of Implicit Government Guarantees and Machine Learning Models

  • 期刊论文

  • 发表日期：

    2026年3月13日

查看全文
  • Zhiyong Dong,
  • Xiuling Yuan,
  • Wenbing Li

摘要

This study investigates the predictive power of implicit government guarantees (IGGs) in forecasting bond defaults among Chinese listed companies. We develop an innovative prediction framework that integrates IGG indicators with advanced machine learning algorithms (Random Forest and XGBoost) and compare their performance against traditional Logit models. Using a comprehensive dataset spanning 2011–2022, we incorporate macroeconomic variables, industry-specific factors, financial metrics, and governance characteristics. Our analysis reveals that (1) IGGs significantly reduce default probability for state-owned enterprises, with fiscal capacity serving as a critical boundary condition; (2) ensemble machine learning models substantially outperform Logit models, achieving 89.3% accuracy and 0.89 AUC; and (3) the interaction between ownership structure and IGG strength creates nonlinear risk patterns undetectable by conventional methods. We further demonstrate that incorporating IGG metrics improves prediction accuracy by 13.6% points compared to financial ratios alone. These findings provide actionable insights for regulators to reform bond pricing mechanisms and for investors to better assess China’s unique credit risk landscape. Our methodology offers a template for adapting default prediction models to institutional contexts where government intervention distorts market discipline.