Recognition Models Of The Credit Risk Of Our Country's Commercial Bank Based On Support Vector Machines

Credit risk in financial markets is the oldest and most important form of risk. It has become the current market environment facing the economies of the most important financial risks. The terms of the actual situation in China, whether from the micro or macro perspective, Chinese financial system has accumulated a lot of risks and financial risks, mainly for credit risk. Credit risk is defined as the borrower can not arranging loans to the debt service and caused the risk of loss. However, with the risk of environment change and risk management technology development, this definition has not fully reflected the modern credit risk management and the nature and characteristics. Modern sense of credit risk should include default by the counterparty direct breach of contract and counterparties to the possibility of changes in the investment portfolio of the risk of loss.The first of the paper is the exordium, which introduced the background of credit risk of commercial bank, status of research and credit risk management in our country's commercial bank. The second part, introduced the credit risk on the concept and features. From the existing research, the commercial banks want to measure their credit risks which they are facing, they must solve the problem is how to determine the loan default probability, or determine whether it will default first. The third part introduced the methods of credit risk measurement under the New Basel Capital Accord, and had the brief commentary on the pros and cons of various methods. As the characteristics of credit risk are endogenous, non-linear, the probability distribution of asymmetry and data collection difficult and so on, the traditional measure of credit risk is difficult to accurately quantitatively analyze credit risk. Therefore, the fourth part apply a pattern recognition method of Support Vector Machines, which grew and developed from the statistical learning theory, to provides a new approach and new tools for the credit risk analysis and assessment. As SVM does not assume the distribution of samples, and the non-linear classification problem can be mapped to the high-dimensional characteristics space becoming the linear classification problem by introducing the Kernel function, which solved the problem that traditional methods took the assumptions of distribution are not same with the fact properly and traditional methods are helpless to the non-linear problem. The fifth part of the paper calculated the credit risk of the commercial bank based on SVM model using the pivoting algorism invented by Professor Zhongzhen Zhang. Although this algorism needs more compute ram to store the Hessian matrix, the speed of convergence of pivoting algorism is faster than the SMO algorism. In the sixth part, the paper selected the A-share listed company's financial data in 2007 as samples, using SVM model, and solving the convex quadratic programming by pivoting algorithm. Then discriminated the default probability of the sample, and take the probability of default compare with the traditional method of Multiple Liner Discrimination Analysis. From the comparison of the experiment, we can easily discover that the misjudgment rate on SVM was significantly lower than the misjudgment rate on Multiple liner Discrimination Analysis. The result verified that SVM does not assume the distribution of the sample and the feature of SVM which solving the non-linear problem of classification to deal with credit risk identification is definitely effective, and made a number of issues which will be solved by using SVM on the recognition of credit risk in the future. The last part summarized the whole paper, and took some suggestions about the study on credit risk of commercial bank in China.