中国商业银行信用风险内部评级法的要素建模和信息化实现研究

发布时间：2018-08-13 10:44

【摘要】：随着中国金融体系改革的进一步深化,巴塞尔资本协议的逐步实施,以及互联网金融的蓬勃发展都对信用风险量化管理提出了更高的要求,而内部评级法的要素建模是实现信用风险量化管理的基础。由于数据、模型复杂度等原因,内部评级法的要素建模及其后续应用都需要信息化的支撑。如何借鉴国内外的研究成果,实现中国商业银行内部评级法的要素建模及其信息化是一个非常值得研究的课题。本文的创新性工作和研究成果总结如下:1.PD预测模型分析了债务人、债项等对PD的影响,给出了变量筛选策略和样本配比方式;提出了命中率、覆盖率和准确率等模型判断标准。采用Logistic、SVM和DT建立了3类6个模型,并在每类中选出一个较好的模型作为最终模型,根据3类模型的最终模型利用MDA方法建立了高效的组合模型。2.LGD预测模型给出了LGD的分布特点,分析了LGD的影响因素;采用DT建立了全样本判别模型;针对非极端情况,利用Beta变换、Logit变换和WOE变换建立了LGD的点估计模型;采用广义Beta建立了LGD的分布模型。结合各模型的优点,建立了包含点估计和分布估计的组合预测模型。3.EAD预测模型给出了违约使用率的分布特征,分析了其影响因素。通过线性回归方法建立了使用率违约预测的初级模型;根据违约使用率分布的特殊性,建立了包含DT、SVM的组合模型用于判别全使用的情况;针对非全使用率的样本,运用WOE变换、广义Beta给出了其点估计和分布估计预测模型。4.M和监管资本计量模拟在准确计量M的基础上,分析了债务人、债项对M的影响;利用线性回归法建立了M的预测模型。采用标准法、混同和分池方式三种方案计量监管资本,对比了三种方式的差异及M对监管资本的影响。5.内部评级法系统原型设计给出了“三统一”的建设思路实现内部评级法的系统建设;提出了内部评级法系统的作业调度优化算法和数据处理优化方案。本文结合中国的实际情况,研究了内部评级法的数据建设、要素建模和信息化,对中国商业银行实施内部评级法及其信息化建设具有重要的指导意义,同时也可为国外相关的建设提供指导。
[Abstract]:With the further reform of China's financial system, the gradual implementation of the Basel Capital Accord and the vigorous development of Internet finance have put forward higher requirements for the quantitative management of credit risk. The element modeling of internal rating method is the basis of quantitative management of credit risk. Because of the data and complexity of the model, the modeling of internal rating method and its subsequent application need the support of information technology. How to use the domestic and foreign research results for reference to realize the element modeling and informatization of the internal rating method of Chinese commercial banks is a subject worth studying. The innovative work and research results of this paper are summarized as follows: 1. The prediction model of PD analyzes the influence of debtors and debts on PD, gives the strategy of variable selection and the method of sample matching, and puts forward the judging criteria of hit ratio, coverage rate and accuracy rate. Three kinds of six models are established by using Logistic-SVM and DT, and a better model is selected as the final model in each class. According to the final model of the three kinds of models, an efficient combination model .2.LGD prediction model is established by using MDA method. The distribution characteristics of LGD are given. The influencing factors of LGD are analyzed, the full sample discriminant model is established by DT, the point estimation model of LGD is established by Beta transform logit transform and WOE transform, and the distribution model of LGD is established by generalized Beta. Combined with the advantages of each model, a combined forecasting model including point estimation and distribution estimation is established. 3. EAD forecasting model gives the distribution characteristics of default utilization rate, and analyzes its influencing factors. The primary model of default prediction is established by linear regression method. According to the particularity of default usage distribution, the combined model including DTS-SVM is established to judge the situation of full use. The WOE transform is applied to the samples with incomplete usage rate. Generalized Beta gives its point estimation and distribution estimation prediction model. 4. M and supervisory capital metrology simulation. On the basis of accurately measuring M, the influence of debtor and debt on M is analyzed, and the prediction model of M is established by using linear regression method. Adopting standard method, mixing and dividing pool method to measure supervision capital, this paper compares the difference of three ways and the influence of M on supervision capital. The prototype design of internal rating method system gives the idea of "three unified" construction to realize the system construction of internal rating method, and puts forward the job scheduling optimization algorithm and data processing optimization scheme of internal rating method system. Based on the actual situation in China, this paper studies the data construction, element modeling and informatization of internal rating method, which has important guiding significance for the implementation of internal rating method and its informatization construction in Chinese commercial banks. At the same time can also provide guidance for foreign related construction.
【学位授予单位】：天津大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：F832.33

【相似文献】