The Study On Early-warning Of Credit Crisis For Merchant Bank In China On Neural Network

The Credit Crisis is world problem. The size of the Credit Crisis has a vital significan-ce for the bank's operating production and even the survival and development.For a long time, it is short of effective Credit Crisis monitoring and control instruments to the Merchant Bank in China.It is usually overabundance in post-processing, qualitative analysis and static analysis but less in advance prevention, quantitative analysis and dynamic analysis.With the emergence and gradual improvement of artificial intelligence techniques, especially the Neural Network which can solve non-linear problem prominently,these problem has been solved.With the center of improved and optimized three-layer BP artificial neural network model,the paper establish index system of Merchant Bank Credit Crisis, which including thirty second class index such as short term debt, long-term debt, profitability index,and so on. It selects 223 samples from Credit Management Information System of certain Merchant Bank in China. It uses the 3sigma evaluation rule to determine the initial customer's credit level and uses the Matlab to test the model.The result of the empirical analysis indicates that:(1) the indexes in the Merchant Bank don't follow the Nomal Distribution, thus we can't use the DMA model to establish Credit Crisis early-warning system in China and can't use T-test to select indexes, too.(2) Compared with traditional BP artificial neural network,the training result given by optimized BP artificial neural network is improved greatly, the accuracy up to 86.98% from 90.05%;so as the testing result,which improved to 90% from 83.33%.And the losses that misjudge high degree alarm to low for banks by the second neural network are less;(3) The paper creately uses the 3sigma evaluation rule in Credit Crisis rating,the last accuracy of results is good.