 ==>> 点击下载文档 |
本文从商业银行个人信用评分系统构成的角度总结并梳理了国内外的相关
研究，在此基础上对相关概念进行了界定，明确了个人信用评分系统的优化目
标，运用博弈论分析了商业银行个人信贷业务中的博弈关系，进而从理论层面
明确了个人信用评分系统优化的关键问题，包括缺少宏观指标及指标冗余问题、
拒绝推论及样本容量问题、信用评分模型的选择问题、违约损失及信用等级问
题，针对这些关键问题，全面地优化了个人信用评分系统
首先，针对个人信用评分指标体系中存在的缺少宏观指标及指标冗余问题，
本文从形成有效且完善的信用评分指标体系的角度出发，分析了影响个人信用
的主要因素，提出了影响个人信用的宏观经济因素，并基于结构方程模型验证
了宏观经济因素对个人信用的影响，提取了有效的宏观经济指标；设计了
PSO-CFS 算法对个人信用指标体系进行属性约简，解决指标冗余问题，提高了
系统的评分精度，同时也有利于商业银行有针对性地进行信用调查和管理
其次，针对个人信用评分样本集中存在的拒绝推论及有效样本容量不足两
个关键问题，本文从构建具有代表性的个人信用评分样本集的角度出发，提出
采用案例推理方法引入拒绝样本来解决拒绝推论问题，并针对案例推理方法中
案例检索和案例重用假设的不合理性，对案例推理方法进行了优化，设计了优
化的案例推理模型，提高了案例推理对拒绝推论问题的解决能力；采用蒙特卡
洛模拟生成有效的信用样本，增加样本集中有效样本的数量，进而构建了更能
代表样本总体的个人信用评分样本集。哈尔滨工业大学管理学博士学位论文
- II -
再次，针对个人信用评分模型中存在的模型的选择和优化问题，本文从建
立精确的个人信用评分模型的角度出发，对现有的多种单一模型进行研究分析，
阐明各个单一信用评分模型的优点及局限性，并在此基础上建立基分类器池；
分别以精度、差异度和误判损失率为选择标准，考虑基分类器之间的互补性和
差异性，设计全局搜索算法输出最优分类器子集；通过比较，选择了行为知识
空间法作为基分类器的融合方法，最终根据商业银行信用风险管理需求实现最
优组合模型的输出
最后，本文针对个人信用评分系统应用中存在的忽略违约损失及缺少信用
等级细分这两个关键问题，对个人信用评分中的违约损失率进行了界定，明确
了个人信用评分中违约损失率的计算方法，以“信用等级越高，违约损失率越小”
为基本原则，构建了基于违约损失率的信用等级划分模型，并以我国商业银行
个人信用样本集中的样本进行了实证，建立了 9 级个人信用等级模型，对各个
级别的贷款用户特征进行了总结，简单阐明了商业银行对于不同等级客户所应
采取的主要信用风险管理策略，完成了对贷款申请者信用等级的有效细分，加
强了系统在商业银行信用风险管理中的应用
综上，本文对个人信用评分系统的指标体系、样本集、评分模型及应用四
个环节进行了全面优化，提高了系统对信用风险的识别能力。本文研究的理论
意义在于，分析了个人信用评分系统对金融市场信贷配置的重要作用，提出了
个人信用评分系统的优化目标；结合宏观经济因素对个人信用的影响，为解决
个人信用评分中的拒绝推论问题提出了较为有效的方法，同时构建了更具有适
用性的个人信用评分模型，以及个人信用等级划分的方法。本文研究的实际意
义在于从信贷资源优化配置及商业银行信贷决策的角度去优化个人信用评分系
统，使其能够更好地为商业银行信用风险管理而服务，提高商业银行的风险管
理能力，促进金融市场健康发展，同时也为信用时代下政府及其他金融机构开
展信用风险管理提供参考
关键词：个人信用评分系统；系统优化；属性约简；拒绝推论；分类器融合Abstract
- III -
Abstract
With the development of economic and technology, credit consumption is
growing all over the world. Credit individuals, credit economy and credit society
gradually formed and have a huge impact on the economic system. The consumer
credit situation has continued to heat up the situation. For commercial banks,
personal credit is an important part of the future development strategy. Personal
credit risk has decentralized universal, non-systematic and other characteristics
which have become the most difficult risk for commercial banks. The personal
credit scoring system can comprehensively investigate the subjective and objective
factors. The personal credit system explores the effective information from the
historical credit samples and makes a predictive judgment on the personal credit
risk. It is one of the indispensable tools to guide the personal credit risk
management of commercial banks. This paper will take the personal credit scoring
system as the research object. The effect is expressed in promoting the effective
allocation of credit resources. Furthermore, the paper puts forward the optimization
goal of personal credit scoring system and optimizes the personal credit system.
This paper summarizes the relevant research at home and abroad from the
perspective of personal credit scoring system. On this basis, the definition of
personal credit scoring system is defined. This paper abstracts the optimization goal
of personal credit scoring system as the important theoretical basis of this paper.
Based on the optimization goal of personal credit scoring system, this paper makes
a comprehensive analysis of the personal credit scoring system, and puts forward
the key problems of personal credit scoring system optimization, including the lack
of macroscopic indicator and indicator redundancy problem, refusal of inference
and sample capacity, the selection of credit scoring model, default loss and credit
rating issues. In response to these key problems, the paper comprehensively
optimizes the personal credit scoring system.
First, according to the lack of macroscopic indicators and indicators
redundancy in the indicator system of personal credit scoring, this paper aims to
form an effective credit rating index system. Based on the influencing factors of
personal credit, this paper focuses on the macroscopic impact of personal credit and
the effect of macroeconomic factors on personal credit which is verified by
structural equation model. Later effective macroeconomic indicators are extracted.
The PSO-CFS algorithm is designed to reduce indicator redundancy of personal
credit index system. It is proved that after the reduction the indicator system has哈尔滨工业大学管理学博士学位论文
- IV -
improved the accuracy of the personal credit score model, which is also beneficial
to commercial banks to carry out credit investigation and management.
Besides, according to the two key problems of the refusal of inference and
sample capacity in the personal credit score database, the paper treats constructing a
representative personal credit score sample library as the goal and proposes a
case-based reasoning method to introduce rejected samples to solve the problem of
refusal. In view of the irrationality of case retrieval and case reuse hypothesis in
case-based reasoning, the paper has optimized the case-based reasoning method to
design an optimized case-based reasoning model to improve the ability of
case-based reasoning to reject the inference problem. The paper uses Monte Carlo
simulation method to generate effective credit samples, increases the number of
valid samples in the sample library, and builds a better representative personal
credit score sample library on behalf of the total samples.
Additionally, according to the selection and optimization of the personal credit
score model, this paper aims to establish an accurate personal credit scoring model.
It studies a variety of existing single models .Then the paper clarifies the
advantages and limitations of the existing single credit score model. Based on this,
the base classifier pool is proposed. By establishing a variety of different
measurement methods, a classifier selection criterion with accuracy, difference and
misjudgment rate is established. Considering the complementarity and difference
between the base classifiers, the paper designed a base classifier which optimizes
the global search algorithm based on the precision, the difference degree and the
misjudgment loss rate as the selection criteria respectively. Through the comparison,
the paper chooses the behavior knowledge space method as the fusion method of the
base classifier, and finally, achieves the output of the optimal combination model
according to the credit risk management needs of commercial banks.
Finally, according to the two key problems of neglecting default loss and lack
of credit rating in the application of personal credit scoring system, this paper
defines the default loss rate in personal credit score and clarifies the calculation
method of default loss rate. Based on the the higher credit rating and the less loss
rate of default, this paper constructs the credit rating model based on default loss
rate, makes a demonstration of the sampl
。。。以上简介无排版格式，详细内容请下载查看
