基于信号噪音差的个人信用评估模型研究

发布时间：2018-05-28 02:16

本文选题：信号噪音差 + 信用风险　；参考：《浙江工商大学》2016年硕士论文

【摘要】：随着中国经济的迅猛发展,居民消费观念的改变,以及信贷政策的积极支持,我国商业银行个人信贷业务不断发展,规模日益庞大。在这快速发展的过程中,信用交易的扩大化导致银行面临的信用风险越来越复杂,商业银行的信用风险管理成为风险管理体系中关键的一个部分。为了能够增强商业银行抵御风险的能力、降低其潜在的经营损失,关于商业银行个人信用风险评估和预测的话题越来越受到关注。本文在此背景之下,选择商业银行个人信用评估模型进行研究,旨在探索商业银行对个人信用风险的预测和评估方法。科学合理的评估方法不仅可以有效地降低商业银行信贷评估的成本,减少潜在损失,而且对中国间接融资的规模扩张和金融市场的健康持续发展都将产生积极影响,在理论和实际上都具有较大的意义。在对个人信用风险评估时多采用信用评分模型,即通过筛选有信息含量的个人特征指标,借助一定的分析方法进行建模。其中,如何在客户所提供的诸多个人特征变量中筛选出最有价值、信息含量最高的变量,如何对所选取的变量进行科学合理的编码分组,这些都成为商业银行在个人信用风险评估中所遇到的关键问题。基于对此话题的浓厚兴趣及其所体现的应用价值,本文选取德国某银行和澳大利亚某银行的客户数据作为样本,借助SAS统计软件,对不同方法下的个人信用风险评估和预测进行研究。本文引入信号噪音差方法,将其与传统T检验方法做比较,通过实证进行检验,从而体现信号噪音差方法在变量筛选方面的优越性。此外,在对特征变量进行编码分组时,本文提供了传统的由多到少逐步合并和基于信号噪音差方法的由少到多逐步分组两种不同的编码方法,结合SAS程序实证检验两种方法的有效性,基于信号噪音差的编码方式为商业银行在信用风险评估变量编码问题上提供了又一种值得借鉴的方法。在信用评估模型的构建方面,本文采用基于Logistic回归分析的信用评分模型和基于修正的朴素贝叶斯分类法的违约概率预测模型。通过实证检验出对于德国某银行和澳大利亚某银行两个样本,在运用Logistic回归分析所构建的模型中,基于信号噪音差的编码方式与基于传统编码方式所构建的模型具有基本相同的预测精度。并且总体来看,运用Logistic回归分析所构建的模型比运用修正的朴素贝叶斯分类法构建的模型具有更高的预测准确率。这说明信号噪音差方法在变量的筛选和编码方面具有很高的参考和运用价值,而且运用Logistic回归分析方法构建信用评分模型是一个较优的选择,同时修正的朴素贝叶斯分类法由于简单方便等优点也是一种可以借鉴的方法。
[Abstract]:With the rapid development of Chinese economy, the change of residents' consumption concept and the active support of credit policy, the personal credit business of Chinese commercial banks is developing continuously and the scale is increasing. In the process of rapid development, the expansion of credit transaction leads to more and more complex credit risk faced by banks. The credit risk management of commercial banks becomes a key part of the risk management system. In order to enhance the ability of commercial banks to resist risks and reduce their potential business losses, the topic of personal credit risk assessment and prediction of commercial banks has been paid more and more attention. Under this background, this paper chooses the commercial bank personal credit evaluation model to carry on the research, in order to explore the commercial bank to the personal credit risk forecast and the appraisal method. Scientific and reasonable evaluation methods can not only effectively reduce the cost of credit assessment of commercial banks and reduce potential losses, but also have a positive impact on the scale expansion of indirect financing in China and the healthy and sustainable development of financial markets. Both in theory and in practice have great significance. In the assessment of personal credit risk, credit scoring model is often used, that is, by screening the individual characteristic index with information content, and by means of certain analytical method, the model is established. Among them, how to select the most valuable and most informative variables among the many personal characteristic variables provided by customers, and how to code the selected variables scientifically and rationally. These are the key problems that commercial banks encounter in personal credit risk assessment. Based on the strong interest in this topic and its application value, this paper selects the customer data of a bank in Germany and a bank in Australia as a sample, with the help of SAS statistical software. The evaluation and prediction of personal credit risk under different methods are studied. In this paper, the signal noise difference method is introduced and compared with the traditional T test method, and then the signal noise difference method is proved to be superior to the traditional T test method in the selection of variables. In addition, in the coding block of feature variables, this paper provides two different coding methods, one is from more to more progressive grouping, the other is based on signal noise difference method. Combining with the SAS program to test the validity of the two methods, the coding method based on the signal noise difference provides another method for commercial banks to use for reference in the problem of credit risk assessment variable coding. In the construction of credit evaluation model, the credit scoring model based on Logistic regression analysis and the prediction model of default probability based on modified naive Bayes classification are adopted in this paper. Through the empirical test, two samples of a bank in Germany and a bank in Australia are tested in the model constructed by using Logistic regression analysis. The coding method based on signal noise difference has the same prediction precision as the model based on traditional coding method. In general, the model constructed by Logistic regression analysis has higher prediction accuracy than that constructed by modified naive Bayes classification. This shows that the method of signal noise difference has a high reference and application value in the selection and coding of variables, and it is a better choice to construct credit rating model by using Logistic regression analysis method. At the same time, the modified naive Bayesian classification is also a useful method for reference because of its simplicity and convenience.
【学位授予单位】：浙江工商大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：F832.4

【相似文献】