摘要
进行内部评级的商业银行在设计主标尺过程中,由于债务人评级分布固有的离散属性,池化违约概率不可避免地会产生模型风险。适度地细分主标尺能够提高校准精度,但也会大大增加计算的复杂性。本文延续均值聚类优化的建模思路,同时引入差分进化算法改善运算速度,满足更多风险分栏、更小间隔尺寸的求解需要,借助较大规模样本数据支持可以显著提高金融机构的风险管理质量。
In designing the internal rating grading scales of the banks,the pooled default probability inevitably leads to model risk due to the discrete nature of the distribution of debtor rating.Moderately subdividing the grading scales can improve the calibration accuracy,but it also greatly increases the computational complexity.This paper extends the modeling idea of means clustering optimization,and introduces the differential evolution algorithm to improve the operational speed,which can meet the needs of more risk buckets and smaller interval size.With the larger sample and data support,it can significantly improve the quality of the risk management of financial institutions.
作者
刘久彪
LIU Jiu-biao(School of Finance,Tianjin University of Finance&Economics,Tianjin 300222,China)
出处
《天津商业大学学报》
2021年第4期60-65,F0003,共7页
Journal of Tianjin University of Commerce
基金
国家社会科学基金年度项目“我国银行系统流动性风险的动态度量与管理研究”(16BJY162)。
关键词
内部评级
主标尺
均值聚类优化
差分进化算法
internal rating systems
grading scales
means clustering optimization
differential evolution algorithm
作者简介
刘久彪(1979-),男,辽宁黑山人,副教授,硕士生导师,主要从事金融风险管理研究。
