| 基于Hellinger散度的企业违约预测模型 |
| Corporate default prediction model based on Hellinger divergence |
| 摘要点击 28 全文点击 0 投稿时间：2022-06-04 修订日期：2023-11-16 |
| 查看/发表评论 下载PDF阅读器 |
| 中文关键词 违约预测; Hellinger散度; 指标离散化; 最小角回归; Probit回归 |
| 英文关键词 default prediction; Hellinger divergence; indicator discretization; least angle regression; Probit regression |
| 基金项目 辽宁省社会科学规划基金 |
| 作者 | 单位 | 邮编 | | 沈隆 | 大连理工大学 | 116024 | | 周颖 | 大连理工大学 | 116024 |
|
| 中文摘要 |
| 以提升商业银行企业客户信用风险识别和管理水平为目的, 提出一种系统的违约预测方法. 首先, 采用Hellinger散度对指标进行离散化, 并对离散指标进行虚拟变量编码. 其次, 在反映信息重复的指标对中, 剔除Hellinger散度小的指标, 并采用添加了L1范数惩罚项的最小角回归方程, 遴选违约预测能力最强的变量组合. 最后, 以预测精度G-mean最大为目标, 反推Probit回归最优的违约预测临界点. 实证研究表明: 非财务因素和宏观因素对中小企业违约预测的影响不容忽视, 除``资产减值损失/营业利润"和``每股权益合计"等财务因素外, ``十大股东S指数"和``高管年薪披露方式"等非财务因素, 及``恩格尔系数"和``人均地区生产总值"等宏观因素具有关键影响. 该方法在准确性和稳健性方面优于对比模型, 且可以揭示影响企业信用风险的关键因素和关键阈值, 为商业银行授信审批和贷前审查工作提供依据. |
| 英文摘要 |
| A systematic default prediction method to improve the credit risk identification of corporate customers of commercial banks is proposed. First, Hellinger divergence is used to the indicators discretization and dummy variable coding. Second, the combination of indicators with the strongest default prediction ability are selected by using the least angle regression with an L1 penalty term. Finally, maximize the G-mean to invert the optimal threshold of the Probit regression. The results show that financial factors such as ``asset impairment losses/operating profit", and ``total equity per share", non-financial factors such as ``top ten shareholders S index", and ``disclosure method of executive compensation", as well as macroeconomic factors such as ``Engel coefficient", and ``per capita regional gross domestic product" have a critical impact. This method is more accurate and robust than comparative models and can reveal the key factors that affect credit risk, inspiring commercial banks' credit approval and pre-loan review work. |
| 关闭 |
|
|
|
|
|
