| Credit is a lending business on the condition of debt service.The nature of credit risk is the default risk.Therefore,the essence of credit risk rating is the identification of default risk,revealing the possibility of debt repayment of different grades’ customers and the rate of default loss.In the credit risk rating,if the default feature extraction error,will bring a significant impact to the whole society and the entire economic system,such as the global financial crisis in 2008 was due to the misjudgment of default feature and the excessive issuance of subprime loans.On the contrary,if identifying the default characteristics accurately and screening the default customers effectively,will reduce the loss of the bank greatly.For example,the scale of loans of Chinese commercial banks was 66.6 trillion yuan in 2014,if the loss rate of loan customers was reduced by 10%,which would reduce the loss of 7 trillion nearly for commercial banks.The research of credit risk rating of small enterprises includes the following four aspects:the establishment of credit risk evaluation index system,the establishment of evaluation equation,the division of credit rating and the identification of key characteristics for customers that default risk is greatest.This paper was studied based on the above four issues respectively:One is the construction of the credit risk evaluation index system,selected the indexes through three steps to ensure that each indicator in the index system has the ability of default identification.Two is the construction of credit risk evaluation equation,select one that the closeness degree is largest,the ability to identify is greatest from five empowerment methods,to ensure that the credit scoring of the default customer is low and the non-default customers is high.Three is the division of credit rating,which develops a nonlinear programming model to assign credit ratings according to the LGD pyramid principle and credit score clustering standard,we demonstrate that credit grade classification must meet the default pyramid principle,that is,the higher the credit rating,the lower the LGD.Additionally,the classification must ensure that customers with a similar credit status are more likely to be assigned to the same credit level.Four is to the mining of the key indicators and features of small enterprises which affect the loss rate of loan default.This study includes seven chapters.The first chapter is the introduction,which introduced the topic,the existing researches,and the methods and so on.The second chapter summarized the theoretical basis of credit risk rating.The third chapter interprets the establishment of the index system of small enterprises.The fourth chapter constructed the credit risk evaluation model.The fifth chapter discusses the credit rating model based on the LGD pyramid principle and credit score clustering standard.The sixth chapter mined the key indicators and features of small enterprises which affect the loss rate of loan default.The seventh chapter presented the conclusions and outlooks.The major works of this study are as follows:(1)Established the credit risk evaluation index system of small enterprises.Using a Chinese regional commercial bank’s 3045 small enterprises loan for empirical study,involved nearly 20 years’ data since 1994.And this study establishes the credit risk evaluation index system of small enterprises including 16 indices,such as speeding ratio,enterprise credit situation and average disposable income of urban residents in the past three years.(2)Constructed the credit risk evaluation equation of small enterprises.By comparing the objective weighting method of entropy method,variation coefficient method and variance method,and the statistic value χ2 of Wilks’ Lambda,the AUC value of the ROC curve to reflect the ability of default identification,according to the distance between the customers’credit score and the positive and negative ideal points,to construct the closeness degree C which could used to identify the optimal weighting method.The empirical results show that the Wilks’ Lambda weighting method has the biggest default identification capacity and were the optimal weighting method in those five methods.(3)Established the credit rating model based on default Pyramid and credit score clustering of small enterprises.Constructed a nonlinear programming model to divide the credit rating with the objective function in which the deviation of credit score is minimized within the groups,a constraint of LGD increasing with credit rating from high to low.Divided the 3045 small enterprises and dig out the default loss rate of different credit rating.(4)Mined the key indicators and features of small enterprises which affect the loss rate of loan default.Firstly,through establishing the order of Logit between credit rating and evaluation indicators,identifying the key indicators such as "Super quick ratio","Enterprise credit in three years".Secondly,by analyzing the correlation between the different characteristics of the key indicators and the default loss rate,it is necessary to test the default loss rate of which features’ customer is the biggest under the same key indicators.For example,the interval of "Super quick ratio" is S3:[0,0.5);the classification of "Enterprise credit in three years" is "have default record but clean up".The innovation of this thesis has the following three aspects:(1)The innovation of credit rating:By establishing a nonlinear programming model to divide the credit rating of loan customers and study the default loss rate of loan customers in different grades.The minimum deviation within the group of the customers’ credit score in each grade is the objective function to ensure that the customers with similar credit score are divided into the same level.The default loss rate of the following grade is greater than the default loss rate of the previous grade increased strictly is the constraint condition to establish the non-linear programming model of the credit rating,so that the credit rating results in case credit situation of different customers in the same level is similar,to ensure that the credit rating results to meet the default pyramid principle that the lower credit rating,the higher the default loss rate,to avoid the credit rating is not low,but the default loss rate is very high absurd phenomenon.(2)The innovation of key feature screening:identifying key features that the default loss rate is the largest by the LSD test.In the case of a key indicator corresponding to different characteristics,such as the key index of "enterprise credit situation" corresponds to "credit records,and no default,no loans","no credit record ","credit records,and no default,loans","default records,it has been settled","default records,outstanding" five characteristics.The discriminant scale LSD is constructed based on the variance within the group of the customer’s default loss rate within the different characteristics,and it is determined whether there is a significant difference in the loss rate of the different characteristics.In the different characteristics that have the significant difference,by comparing the size of the default loss rate to determine the customer’s default risk of which credit characteristics is the largest in the key indicators of"corporate credit situation".If the default loss rate of a certain characteristic is significantly larger than other characteristics’ default loss rate,the feature is the key feature,to seize the key to credit risk management,open up the new idea of credit risk rating theory,change the drawbacks only based on the customer order and ignoring the deep mining and exploration of the key features of credit risk management in the existing research fundamentally.(3)The innovation of determine the weight of the indicators:according to the idea that the greater the default identification ability of the credit evaluation results,that is," the higher the credit score of non-default customers,the lower the credit score of default customers",the better the corresponding empowerment method,to choose an optimal from different empowerment methods.According to the credit evaluation equation Sj=w1x1j+…w2x2j…+wnxnj to determine the customer’s credit score,through the distance between the customers’ credit score and positive and negative ideal point to construct the closeness degree C that can reflect the default identification ability of the credit evaluation results,if the credit score of the default customer is closer to the worst value 0,the credit score of the non-default customer is closer to the best value 1,then the closeness degree C is greater,the corresponding to weighting method can distinct between non-default and default customers,to the greatest extent,and then to select one that the closeness degree is largest,the ability to identify is greatest from different empowerment methods,to ensure that the credit scoring of the default customer is low and the non-default customers is high,not only to avoid the shortcoming that evaluation results can not distinguish default and non-default customers effectively to lead to a large number of overlapping;but also to avoid the weakness that empowerment method of random subjective choice is no connection with the evaluation purpose in the existing study. |
