摘要
现有关于违约损失的研究通常关注均值预测,但是均值无法有效刻画违约损失率分布的非正态、极端偏斜和双峰特征。本文利用条件分位数回归模型,完整刻画了违约损失率的分布,从新的视角量化了协变量对违约损失率的影响。研究结果表明,相比各种均值回归模型,分位数回归模型在估计违约损失率时具有显著的优势,违约损失率建模方法可为商业银行等贷款机构估计违约损失率增加选择。
Existing most studies about default loss usually focus on the average forecast, but the mean cannot effectively depict the distribution of the default loss rate, especially, as for non-normal extreme deflection and bimodal characteristics. This paper depicts perfectlythe distribution of the default loss rate using conditional quantile regression model, which is a new perspective to quantify the impact on default loss rate model from covariate. The comparative test results among several models show that quantile regression model in estimating the default loss rate has a more significant advantage than all kinds of mean regression model. The modeling approach suggested in this paper can be an alternative method used to estimate the default loss rate for commercial banks and other lending institutions.
出处
《证券市场导报》
CSSCI
北大核心
2018年第8期55-65,共11页
Securities Market Herald
基金
国家社会科学基金项目"中国老年人长期护理与医疗保障体系改革研究"(15BJY183)
国家社会科学基金一般项目"新兴经济体金融脆弱性测度及协同治理研究"(17BJY184)
山东省自然科学基金项目(联合专项)"贝叶斯分层模型及其在信用风险量化中的应用"(ZR2018LG001)
山东省高等学校人文社科计划一般项目"基于非参数贝叶斯分层模型的信息不完全下信用风险量化研究"(J17RA103)
济南大学自然科学基金"债券利率期限结构的半参数贝叶斯分层模型及其应用研究"(XKY1636)
关键词
不良贷款
违约损失率
分位数回归
债务违约
non-performing loan
loss given default
quantile regression
default of debt
