东篱科研大数据发现系统（DRDS）

位置：成果数据库 > 期刊 > 期刊详情页

诉讼处置不良贷款违约损失率估计的模型簇

ISSN号：1001-4098
期刊名称：系统工程
时间：2015.8
页码：123-132
分类：F830[经济管理—金融学]
作者机构：[1]天津大学管理与经济学部,天津300072, [2]中国人民大学财政金融学院,北京100080, [3]中央财经大学管理科学与工程学院,北京100081, [4]中国科学院数学与系统科学研究院,北京100190
相关基金：国家自然基金重点资助项目（70933003）;国家自然基金资助项目（70871109）; 国家自然科学基金青年项目（71203247）; 教育部人文社科青年基金资助项目（11YJC790015）
相关项目：商业银行跨周期动态信用风险模型构建与调整

关键词：不良贷款, 诉讼处置, 违约损失率LGD, Logit变换, Beta-正态逆变换, WOE变换, 广义Beta回归, 模型簇, Non-performing Loans（NPLS）, Disposal of Litigation, Loss Given Default（LGD）, Logit Transformation, Beta Transformation, WOE Transformation, Generalized Beta Regression, Collective Models

中文摘要：

违约损失率（LGD）是内部评级高级法要求的重要参数之一,已成为商业银行风险管理的重要手段。由于受数据等多方面的限制,国内外尚无对我国大型商业银行诉讼处置不良贷款违约损失率估计的研究。本文在对我国某大型商业银行诉讼处置不良贷款违约损失率全面统计分析的基础上,找出回收率的影响因素,采用决策树模型,判别出极端回收和非极端回收。针对非极端回收的情况综合运用Logit变换、Beta-正态逆变换、WOE变换等方法,建立点估计模型;随后利用广义Beta回归给出了LGD的分布模型;在对四个模型进行相关性分析的基础上用最小误差平方和的方法建立了组合模型,由此形成由判别模型与组合模型构成的模型簇。实证结果表明,极端回收的判别准确率高达77.6%,组合模型的均方误差低于5%;模型簇在极端回收和非极端回收两类表现出很好的一致性。

英文摘要：

Loss given default（LGD）is one of the important parameters of Advanced Internal Rating Method,which has become an important means of commercial banks risk management.With the continuous progress of China＇s legal system construction process,the litigation disposal of NPLs is becoming more and more important,but the litigation disposal is influenced by the politics,law,and culture,at the same time due to the lack of data and many other restrictions,there is no study about LGD estimation of litigation disposal of NPLs of China＇s large commercial banks.Based on the data about litigation disposal of a large commercial bank,we first utilize statistical analysis to identify the influencing factors of the Recovery Rate,and then construct a discriminant model using decision tree method to classify extreme recovery（RR=0or RR=1）.For the case of Non-extreme recovery（0RR1）,Logit transformation,Beta transformation,and WOE transformation are used to build point estimate model,and general Beta regression is applied to construct distribution estimate model.Four models＇ correlation is analyzed,and combination model is established based on minimizing the sum of square forecasting errors.Finally,collective models are built by discriminant model and combination model.The empirical results show that the accuracy rate of discriminant model is 77.6percent;the MRSE of combination model is lower than5 percent.The collective models have consistency both extreme recovery and Non-extreme recovery.

同期刊论文项目