信用评分模型的开发及probit回归在模型中的应用

发布时间：2018-04-18 10:38

本文选题：信用评分 + probit回归　；参考：《山东大学》2012年硕士论文

【摘要】：信用风险作为最主要的金融风险类型,是当前金融界的最大课题。我国信贷业正进入飞速发展时期,许多银行相继提出建设零售银行的宏伟蓝图,这使得各银行信贷业务量日益巨大。传统的人工授信已经无法适应这种需求,信用评分模型这一在国外银行业和信贷业逐步兴起的技术,也必将在中国得到广泛运用。在银行业竞争日益激烈的情况下,信用评分的研究和模型的选取成为银行面临的一项极富有挑战性的管理问题。probit回归是与logistic回归十分类似的广义线性模型,用于解决因变量的二分类问题。在建立信用评分模型时,logistic回归是一种十分常用的统计方法,而probit回归在这方面却极少论及。本文利用probit回归建立申请信用评分模型,计算每个客户的违约概率,进而将客户分为两类,并对模型的分类效果进行了检验。本文首先从监管要求和银行内在需求两个方面阐述了大力发展信用评分模型的必要性,并概括总结了信用评分模型在应用中表现出来的巨大优势。本文的第二章介绍了国内外信用评分模型的发展现状,并对现有的建立信用评分模型的常用方法进行了描述和讨论,并比较了这些方法的优劣,最后又对变量指标的选取和数据处理的常用方法进行了介绍。本文的第三章是重点内容,首先介绍了建立信用评分模型的开发流程和所要面对的问题,这些问题包括模型分类、风险因素变量清单选取、坏样本数据定义及识别、建模数据来源等。然后,重点论述了建立模型的probit回归及检验模型分类情况的ROC曲线及CAP曲线及相关量化指标。本文的第四章是一个完整的建模过程。本文先从数据集中分别按照好坏样本比为1：1,2：1,3：1抽取建模数据,利用这些数据的原始值和woe值进行建模,并且用剩余的数据进行了样本外检验。最后,对中各种情况下得到的结果进行了比较,得出了一组最好的分类结果。本文的第五章是结论部分,对第四部分得到的结果进行了总结,指出了模型的不足,并展望了违约概率模型今后的发展。
[Abstract]:Credit risk, as the most important type of financial risk, is the biggest subject in the current financial circle.China's credit industry is entering a period of rapid development, many banks have put forward a grand blueprint for the construction of retail banks, which makes the amount of credit business of banks increasingly huge.Traditional artificial credit has been unable to meet this demand, and credit scoring model, which is gradually rising in foreign banking and credit industry, will be widely used in China.Under the increasingly fierce competition in the banking industry, the study of credit rating and the selection of models become a challenging management problem for banks. Probit regression is a generalized linear model similar to logistic regression.It is used to solve the problem of two classification of dependent variables.Logistic regression is a very common statistical method in establishing credit scoring model, but probit regression is rarely discussed in this respect.In this paper, the application credit rating model is established by probit regression, and the default probability of each customer is calculated, and then the customers are divided into two categories, and the classification effect of the model is tested.Firstly, this paper expounds the necessity of developing credit scoring model from two aspects of supervision requirements and internal requirements of banks, and summarizes the great advantages of credit scoring model in application.The second chapter introduces the development of credit scoring models at home and abroad, describes and discusses the common methods of establishing credit scoring models, and compares the advantages and disadvantages of these methods.Finally, the selection of variable indexes and common methods of data processing are introduced.The third chapter of this paper is the key content, first introduced the establishment of credit scoring model development process and the problems to be faced, including model classification, risk factors variable list selection, bad sample data definition and identification,Modeling data sources and so on.Then, the probit regression of the model and the ROC curve and CAP curve of the model classification are discussed in detail.The fourth chapter of this paper is a complete modeling process.This paper first extracts the modeling data from the data set according to the ratio of good and bad samples 1: 1: 2: 1: 3: 1, and uses the original value and woe value of these data to model the model, and tests the remaining data out of the sample.Finally, the results obtained in various cases are compared and a group of best classification results are obtained.The fifth chapter of this paper is the conclusion part, summarizes the results obtained in the fourth part, points out the shortcomings of the model, and looks forward to the future development of the probability of default model.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：F224;F830.3

【参考文献】