Research On Credit Risk Measurement Of Commercial Banks Based On LOGIT-SVM

Credit risk is the main and basic risk of commercial bank.The steadydevelopment of commercial banks also need to rely on the effective management ofcredit risk.With the process of the marketization of interest rateaccelerate and thecontinuous financial innovation development,Commercial banks are facing more andmore fierce competition environment.Regarded as the most important risk forcommercial bank, credit risk has once again become the focus of financialindustry.Under this background, putting forward new quantitative models to measure,control and manage the credit risk of commercial bank,to avoid unnecessary losses, toreduce the overall bad debt ratio,is a pressing matter of the moment whichcommercial bank faces.However, the traditional credit risk assessment is throughstatic historical financial ratio analysis, not building a system of risk identificationmodel to predict and control the credit risk.Therefore, proposing a system modelwhich can effectively measure credit risk which the bank is facing is very important.SVM algorithm described in this paper is a kind of machine learning algorithm,which belongs to the convex optimization problem. When the number of financialdata variable is larger, this algorithm can usekernel function to mapping the highdimensional sample points into a low dimensional space in which they arecalculated,and we even does not need to understand the mapping form of kernelfunction.Thus it can be very convenient to solve the classification problem of highdimension sample points.At the same time, the SVM algorithm can overcome thedefect of traditional methods that sample data must meet certain distribution, SVMalgorithm has no requirement for the initial data distribution, so it can quickly classifythe small and nonlinear sample set, and the prediction accuracy is higher.In this paper, firstly the related theory of credit risk management is summarized,and introduces the traditional methods used in the process of commercial banks’ creditrisk management. Secondly, explaining the theoretical basis of application of SVMalgorithm from statistical learning theory perspective,and proposing the application ofan improved grid search method in SVM algorithm. It can improve the accuracy andefficiency of the traditional grid search method.And then the improved SVMalgorithm and traditional logistic model are integrated in a combined model, which overcomes the defectof logistic model that classificationaccuracy is lowerin theinterval near the critical point. Thus the combined model’s accuracy ofclassificationhas effectively improved.Finally, selecting representative financial data as the samplesfrom the A stock market. On the use of traditional logistic model, the traditional SVMalgorithm, the SVM algorithm based on an improved grid search method andLOGIT-SVM combination model for empirical comparative study. Empirical resultsshow that the combined model has the highest classification accuracycompared toother models.