Scholarlink

<p>Accurate financial distress prediction for listed companies is crucial for informed decision-making by investors and financial institutions. Recent advancements have highlighted the potential of graph models due to their ability to represent structures and capture relational data. This study leverages both textual and tabular (non-textual) data to construct association networks, applying Graph Sample and aggregate model (GraphSAGE) to integrate features for comprehensive prediction of financially distressed companies. An empirical analysis of Chinese listed companies shows that our model outperforms 10 others, including Random Forest and Logistic Regression, in metrics such as KS and G-mean. The inclusion of textual networks notably improves prediction accuracy, achieving a MK of 0.694 and a G-mean of 0.847. Additionally, compared to the tabular network, the textual network exhibits more closely linked nodes among related companies, highlighting its effectiveness in capturing relational complexities.</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-Perspective Fusion Graph Model for Financial Distress Prediction of Listed Companies

  • 期刊论文

  • 发表日期：

    2026年1月29日

查看全文
  • Zhipeng Zhang,
  • Jiaxin Pang,
  • Gang Li

摘要

Accurate financial distress prediction for listed companies is crucial for informed decision-making by investors and financial institutions. Recent advancements have highlighted the potential of graph models due to their ability to represent structures and capture relational data. This study leverages both textual and tabular (non-textual) data to construct association networks, applying Graph Sample and aggregate model (GraphSAGE) to integrate features for comprehensive prediction of financially distressed companies. An empirical analysis of Chinese listed companies shows that our model outperforms 10 others, including Random Forest and Logistic Regression, in metrics such as KS and G-mean. The inclusion of textual networks notably improves prediction accuracy, achieving a MK of 0.694 and a G-mean of 0.847. Additionally, compared to the tabular network, the textual network exhibits more closely linked nodes among related companies, highlighting its effectiveness in capturing relational complexities.