Scholarlink

Objective <p>To solve the complex risk identification and nonlinear transmission path in supply chain finance, this study aims to construct a deep learning model that can accurately quantify and dynamically predict supply chain financial risks.</p> Methods <p>Firstly, the scenario optimization reduction technique is used to improve the conditional risk value to accurately measure the tail risk. Secondly, the deep combination features are extracted from the multi-dimensional risk indicators through CNN. Finally, the quantified CVaR risk value is fused with the depth features and input into the LSTM to capture the dynamic evolution of risk over time. A hybrid risk identification model COSR-CVaR-CNN-LSTM is constructed.</p> Results <p>The study model performed best in the risk identification task, with an AUC value of 0.967 and an F1 score of 0.933. In terms of risk quantification accuracy, the root mean square error was only 0.124, which was significantly reduced by 38% compared with the suboptimal model. It can effectively improve the accuracy of investment risk, financing risk, operational risk and tax policy risk prediction.</p> Conclusion <p>The research model can provide financial institutions with more reliable risk assessment tools, and has important practical value and feasible technical solutions in implementing real-time credit approval, dynamic risk pricing and intelligent regulatory applications.</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

Supply chain financial risk identification based on improved CVaR measurement model

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月21日

查看全文
  • Daiyou Xiao,
  • Yaping Wang

摘要

Objective

To solve the complex risk identification and nonlinear transmission path in supply chain finance, this study aims to construct a deep learning model that can accurately quantify and dynamically predict supply chain financial risks.

Methods

Firstly, the scenario optimization reduction technique is used to improve the conditional risk value to accurately measure the tail risk. Secondly, the deep combination features are extracted from the multi-dimensional risk indicators through CNN. Finally, the quantified CVaR risk value is fused with the depth features and input into the LSTM to capture the dynamic evolution of risk over time. A hybrid risk identification model COSR-CVaR-CNN-LSTM is constructed.

Results

The study model performed best in the risk identification task, with an AUC value of 0.967 and an F1 score of 0.933. In terms of risk quantification accuracy, the root mean square error was only 0.124, which was significantly reduced by 38% compared with the suboptimal model. It can effectively improve the accuracy of investment risk, financing risk, operational risk and tax policy risk prediction.

Conclusion

The research model can provide financial institutions with more reliable risk assessment tools, and has important practical value and feasible technical solutions in implementing real-time credit approval, dynamic risk pricing and intelligent regulatory applications.