个人信用评估模型比较研究

【作者】 崔媛媛；

【导师】 丁娟娟；

【作者基本信息】 北方工业大学 ， 数量经济学， 2006， 硕士

【摘要】 随着我国商业银行消费信贷业务的展开,个人信用得到了空前的重视。在我国个人信用体系缺失的情况之下,如何在商业银行内部发展一种合理有效的个人信用评估方法,是商业银行实现贷前风险控制、进一步促进消费信贷发展的关键。 本文从定性分析和定量分析两方面入手,借鉴已有的理论方法和实证结果,结合我国的具体情况,经过比较分析,给出了可供我国商业银行参照的实证研究结果。 从定性分析角度,首先,运用经济学理论对个人信用行为进行分析。根据经济学理论中对人的行为的三个假设,即:利益最大化假设、有限理性假设和机会主义假设,对消费信贷中人的行为进行分析,研究影响个人信用的主要因素;其次,对信用的“5C”参数逐个讨论,深入分析它们与个人信用的内在联系;最后,对国内外的个人信用模型进行比较研究,讨论其异同。通过以上三个方面的研究,为建立个人信用评估指标体系奠定了理论基础。 从定量分析角度,依据定性分析的结果,建立个人信用等级评价指标体系,选用的指标包括:性别、年龄、收入、学历等;通过对各种信用评估模型优缺点的分析,兼顾样本数据的可获得性,最终选择了判别分析、Logistic回归、分类树和神经网络四种方法建立本文的个人信用评估模型,并就指标重要性分析、分类错误率、模型验证和模型有效性进行比较研究。判别分析、Logistic回归的优点在于模型容易解释,但是错分率较高;而分类树和神经网络的预测准确度要好些,但存在着易于训练过度和没有解释能力的缺点。就本文的数据而言,最终得出了神经网络模型最好,分类树模型次之,Logistic回归模型最差的结论。更多还原

【Abstract】 With the development of consumer credit in domestic commercial banks, personal credit is attached more and more value in our country. Under the situation that there is no personal credit system, how to develop reasonable and effective methods to evaluate customers’ credit in the internal of commercial banks is critical to realize risk control before granting and to further consumer credit.Based on former studies and conclusions, this thesis makes a qualitative analysis and quantitative analysis. Combining the special conditions in our country and using analytical comparison, the dissertation establishes the models, and presents the new conclusion that should be considered in domestic banks’ credit rating.From the point of a qualitative analysis view, first it uses three hypotheses in economics theory to analyze personal credit action and analyze the effect factor of personal credit. The hypotheses is most profit, finite senses and opportunism;secondly, it goes deep into analysis with "5C ";Finally, it compares foreign credit model to our country’s, and discusses these similarities and differences.From the point of a quantitative analysis view, according to the point of a qualitative analysis view, the thesis establishes a set of index system of personal credit evaluation model, including: sex, age, income, education and so on. Based on a set of car loan sample of Chinese commercial bank, a systemically comparative study of various statistical credit rating models was made in China. The comparative study indicated that every model has its own strength and weakness.The strengths of linear discriminant analysis, and logistic regression are that these models are explainable and their outputs can be a linear score card (so can be easily implemented). But these models have higher misclassification rate comparing with others. Neural network and classification tree models have a higher predict accuracy, but may be over fitted, and their outputs are hard to be explained. As for the data set of this dissertation, we finally come to the conclusion that Neural network model is relatively the best of the four models, classification tree model ranks the second, and Logistic regression method is not fit for the data we have used.更多还原