Based On BP Neural Network Of A Bank Of SME Credit Risk System Evaluation

In the context of the ?new normal? economy, many small and medium-sized enterprises will become the engine of the great transformation of China's economic and the new growth point of current and future profit of commercial banks. However, the financial information of the small and medium-sized enterprises is not perfect currently and the management of enterprises is opacity, which trouble commercial banks and other financial institutions to carry out the loan to the enterprises seriously. As a new entrants of Hunan market, the Changsha branch of A bank need to develop small and medium-sized enterprises as its customer continuously in order to improve its competitiveness in the Hunan market.First of all, this article sorts out the relevant theori es of credit assessment of small and medium-sized enterprises. Especially, it analyzes the financing mechanism of small and medium-sized enterprises from the perspective of the asymmetric information theory and transaction cost theory. Futhermore, this pap er analyzes the small and medium-sized enterprise credit evaluation method from a technical point, providing theoretical and technical basis for the research of this paper.Secondly, this paper builds a small and medium-sized enterprises credit evaluation system for the Changsha branch of A bank from the dimensions of index system and neural network model. Firstly, the small and medium-sized enterprise credit evaluation index system is based on the relevant information of the small and medium-sized enterprises collected by the Changsha branch of A bank. And this system should follow four principles: the balance of significance and comprehensiveness, the balance of stability and dynamics, the combination of feasibility and effectiveness and the combination of quantitative index and qualitative index. The index system which suit the credit risk of small and medium-sized enterprises should be constructed under these four principles. Then this paper builds a BP neural network model. After reviewing the key techno logies and methods of the construction of neural network(especially the BP neural network) in detail, this paper puts forward a BP neural network model using Matlab7.0. This model can fit the credit conditions of small and medium-sized enterprises and the character of the data sample collected by the bank.Thirdly, basing on the data of 1021 samples collected by the Changsha branch of A bank, this paper constructs a credit index system of small and medium-sized enterprises and does an detailed assignment. Then this paper tests and explains questions concerning the level and structure of the BP neural network model while the program is running. Then the data sample of the credit conditions of small and medium-sized enterprises is divided into two parts, the 80% of the training sample and the 20% of test data, and the BP neural network is trained and tested. The result turns out to be satisfied; When the output is credit score index, the accuracy of training sample is 90.352%, the accuracy of the test sample is 90.009%, the standard deviation of the test sample is 5.098, and the average error is 0.052; When the output is indicators for loan or not, the accuracy of training sample is 83.958%, the accuracy of the test sample is 82.534%, the standard deviation and average error are 0.546 and 0.004, respectively. Afterwards, this paper changes the division method of training sample and test sample to test the robustness of the accuracy of the BP neural network. The BP neural network is proved by these two plans to be robust in the prediction of credit conditions of small and medium-sized enterprises.Finally, on the basis of the difficulties in the process of building credit evaluation system of small and medium-sized enterprises for the Changsha branch of A bank, this paper puts forward four relevant suggestions: enrich the sample database of small and medium-sized enterprises, strengthen the dynamic management of the index system, balance the third party credit rating and the internal rating and construct a group of multiple internal credit rating model in the bank.