深度学习与企业债券信用风险
Deep Learning and Corporate Bond Credit Risk
姜富伟 1柴百霖 2林奕皓2
作者信息
- 1. 厦门大学经济学院,厦门 361005;厦门大学王亚南经济研究院,厦门 361005
- 2. 中央财经大学金融学院,北京 102206
- 折叠
摘要
本文构建了基于深度学习的企业债券信用风险预测模型(CDL),并探究其背后经济机制.研究发现,相比于经典机器学习模型和普通神经网络模型,CDL深度学习模型能够更准确地预测企业债券信用风险.机制分析表明,对于具有低评级等特征的相对风险更高的债券,CDL深度学习模型表现出更强的非线性预测能力.估值与成长类、无形资产类指标是模型预测的重要企业特征.CDL 深度学习模型还能有效识别交易量小、融资约束高、内部控制质量低的经济特征,进而识别出高风险债券.本研究为债券信用风险预测提供了新的思路,有助于维护金融市场稳定,促进经济高质量发展.
Abstract
This paper constructs a credit risk prediction model based on deep learn-ing(CDL),and explores the economic mechanism behind it.Empirical result shows that CDL model can predict corporate bond credit risk more accurately compared with classical machine learning model and ordinary neural network model.Mechanism analysis shows that CDL model has stronger nonlinear prediction ability for bonds with higher relative risk.In terms of enterprise characteristics,valuation and growth indicators and intangible asset indicators are more important in the model predic-tion.In addition,CDL model identifies bonds with high risk by effectively identifying economic characteristics such as small trading volume,high financing constraints,and low internal control quality.This paper provides a new way to predict bond credit risk,which is helpful to maintain financial market stability and promote high-quality economic development.
关键词
债券市场/信用风险/深度学习Key words
bond market/credit risk/deep learning引用本文复制引用
出版年
2024
NETL
NSTL
万方数据