Small Enterprise Credit Rating Model Based On Information Sensitivity

Small business is an important part of the national economic and social development. Credit rating results directly affect on their financing channels, possibility and costs of small enterprises. The results of the small businesses credit rating will affect even the signing of the purchase and sale contracts, participation in bidding and government procurement, etc. Especially, the difficulty and the highly interest for the small enterprises loans always have not been effectively solved. It is very difficult and complex to solve the problem. Now, the credit rating results have been incorporated into the credit approval process in the commercial bank and other financing institution. Obviously, scientific and reasonable small business credit rating will help to solve the problem of small business loan. Moreover, credit rating is also directly related to the safety of the national financial system. Therefore, research on the small business credit rating is very important.The essence of the small business credit rating is credit risk evaluation. In order to reflect vividly the relative size of the enterprise credit risk, credit risk grade needs to be divided. Credit rating index system and weight of credit rating index are the foundation of credit risk assessment. The credit rating index system is achieved by screening the credit rating indicators. That is eliminating the indices that the ability to identify the credit risk is weak and information overlapping is high. Credit rating index weight means determining the weight of credit rating indices. This thesis mainly focuses on the small business credit risk rating and its reasonableness test. The main work is shown as follows:(1) In order to select the indices that credit risk identification ability is big, an index screening method is proposed based on credit risk identification ability. Firstly, the value of the credit rating index is divided into credit default and non-default. And according to F test ideas to eliminate the indices that they have not a significant impact on the default state, it can ensure the retained every indicator have certain influence on the credit defaults state. F test ideas are shown as follows:the bigger relative gap between inter group variation and intra group variation of a credit rating index, the more influence on default state, so the index should be retained. Second, get the percentage of principal component to the credit rating indices and the partial derivative of the each retained principal component to the credit rating index, and multiply them. This thesis uses the summation of product terms to express the sensitive degree of the credit rating indices information to the change of index size. It is called as information sensibility. On this basis, according to thought of information sensibility to the indices that their ability to recognize the credit default comprehensive risk is weak. The ideas of information sensibility are shown as follows:the information sensibility of the index is bigger, the ability of the index's explaining credit risk information of the credit rating indices set is bigger, ability to recognize the credit default comprehensive risk is bigger, and the index should be retained. This method overcomes the insufficient that the existing research only depend on influence of the credit rating index to the default or not when eliminating the redundant indices, and overcomes the insufficient that it only depend on a single factor loading does not selecting reasonably indices in existing method of the principal component dimension reduction.(2) In order to reduce the degree of information overlap among the credit rating indices. a model of selecting credit rating indicators is proposed based on the deleting information overlapping twice. The information overlapping of the credit rating index system is reflected by its ill-condition index. The information overlapping degree among an indicator and the rest indices is determined by the decline extent of the indices group ill-condition index after deleting an index, and it is called information overlap of the index. The credit rating indices' information overlap is reduced rapidly by removing the indices with higher information. To avoid this situation, the information overlap among all indices is low, but the information overlap among some indices is high, further the index with smaller information sensitivity is removed in any two highly correlated indices. It overcomes the shortcomings that the existing research does not consider the degree of the information overlap among all credit rating indices.(3) In order to determine the weight of the credit rating index, a weighting method for credit rating index is proposed based on dual credit default risk identification.According to the following train of thought to determine weight of credit rating index for the first time:the larger amount of information (information gain) is, the bigger ability of recognizing default state for the index, the bigger weight of the index. Information sensitivity of a credit rating index can reflect the ability of recognizing credit default comprehensive risk. According to the following train of thought to determine weight of credit rating index for the second time:the bigger information sensitivity of the credit rating index is. the bigger ability of recognizing credit default comprehensive risk for the index, the bigger weight of the index. The credit risk value of default state is used in the first weight, but default state cannot depict the credit risk subtly. And the second weight reflect the ability of the credit rating index's recognizing the comprehensive risk of the credit default, but it does not utilizes the credit risk value of default state effectively. At last, it uses the multiplication integration normalization method to synthesize the two weighting methods, makes complement on each other, And it overcomes the shortcomings that the existing research only considering the default state factors, or no considering it when weighting the credit rating index. At the same time, the credit comprehensive score of the small enterprises is determined by screening and weighting credit rating indicators. And the credit score ranges corresponding to different credit risk rating grades are divided, the grades of the small enterprises credit risk are determined.(4) In order to test the rationality of the credit rating index system for small enterprises, a rationality test model of credit rating index system is proposed based on Bayesian Discriminant. The idea of this model is shown as follows:the accuracy of discriminating default state for an index system is higher, the ability of index system to identify the basic credit risk is greater. This model overcomes the lack that the existing studies do not consider the difference of units and dimensions among different indices, and the multiple coefficient of determination is used wrongly. In addition, in order to test weighting rationality of small enterprise credit rating indices, a weighting rationality test model of small enterprise credit rating index is proposed based on rate of credit inconsistency. The idea of the test model is shown as follows:comparing inconformity between the credit scores and the actual default loss rate of all the samples enterprises. This method overcomes shortcoming that the existing studies lack of such test methods. And in the empirical study, it compares the small enterprise credit rating index system with the other two credit rating index systems of the existing studies in two different aspects of accuracy rate of discrimination default state and information overlapping. And it further verifies the rationality of the suggested credit rating indices system.