Research On Intelligent Early Warning For Extreme Risk Of Chinese Financial Market Based On The Improved SVM

With the further development of economic globalization, the relation between Chinese financial market and international financial markets is becoming increasingly closer. At the same time of enhancing the level of opening, Chinese financial market also suffers the threats and challenges from abroad extremely financial risk, which becomes severer due to the weak risk control and prevention capability of Chinese financial market. It becomes an urgent problem for financial risk management and investors that how to build an effective early-warning model to accurately predict Chinese extremely financial risk in order to develop effective measures to deal with the risk.Taking CSI300 as the objects of research, this paper selects 16 internal and external risk characteristics indicators. And independent samples t test and k-s test are used to extract three internal characteristics including of opening price, closing price and volume that can depict extremely financial risk of Chinese financial market significantly. Meanwhile, Clayton-Copula is also used to calculate the risk transmission effect between Chinese financial market and abroad financial market and extract three external characteristic indicators including Hang Seng Index daily return, South Korean Stock Index daily return and Taiwan Weighted Index daily return that can depict extremely financial risk of Chinese financial market significantly. In addition, in order to accurately define the extremely and non-extremely financial risk samples, this paper improves and combines the method based on financial crisis time and EVT to accurately define the extremely and non-extremely financial risk samples and the condition indicators are defined successfully. And then the extracted internal and external characteristic indicators and the defined condition indicators are integrated to construct the early-warning indictor system of the extreme risk of Chinese financial market. On this basis, considering the complex nonlinear characteristic of Chinese financial market, an artificial intelligence technology named support vector machine(SVM) derived from the computer science is introduced into early-warning research. Meanwhile, for the challenge of building SVM early-warning model from the unbalanced sample problems in the financial market, this paper also introduces the unbalanced sampling processing methods derived from the computer science, especially the mixed sampling methods, to overcome the unbalanced sample problems of SVM, in order to construct an improved SVM model also named ODR-ADASYN-SVM model suitable for the early-warning of extremely financial risk of Chinese financial market. Thereby some tests are constructed to prove the superior performance of SVM model. The main contents are as follows:1. A research on the selection of kernel functions of the improved SVM. The kernel functions play a vital role in constructing SVM. But there is not yet a unified conclusion that SVM based on which kernel function has best prediction performance. In this paper, based on four kernel functions including linear, polynomial, RBF and Sigmoid kernel functions, the improved SVMs are compared. The result shows that the improved SVM based on linear kernel function has best prediction effect. But the difference of prediction effect between the improved SVM based on linear kernel function and those based on other kernel functions isn’t significant, which shows that there is no significant effect on the prediction performance of the improved SVM model with the change of the kernel functions and further proves that the improved SVM model has stable prediction performance.2. A research on the unbalanced sample processing methods. The unbalanced characteristic of financial market samples makes the research on SVM early-warning face great difficulties. It is crucial to solve the unbalanced sample problem of SVM for improving the prediction performance of SVM. Based on this, this paper designs a hybrid sampling method named ODR-ADASYN, which combines with SVM to create an improved SVM model named ODR-ADASYN-SVM. And then the improved SVM is compared with SVM, SMOTE-SVM, ODR-SVM and ADASYN-SVM. The result shows that the unbalanced sampling processing method including of SMOTE, ODR, ADASYN and ODR-ADASYN can effectively improve the prediction performance of SVM. Meanwhile, the hybrid sampling method of ODR-ADASYN can solve the unbalanced sample problem most effectively and overcome the blindness of SMOTE.3. A research on different early-warning models. At present, there are numerous early-warning models. Does SVM have more excellent prediction performance than other early-warning models? Based on this, after balancing the samples with ODR-ADASYN, this paper choses the artificial intelligence models represented by BP Neural Network, Decision Tree and the Statistical models represented by Logit, Probit, Fisher discriminant analysis, Bayesian discriminant analysis, distance discriminant analysis, which are compared with SVM. The result shows that for the prediction of extremely financial risk, SVM is superior to other early-warning model, but the advantage of SVM isn’t obvious. However, for the prediction of non-extremely financial risk, SVM is not only superior to but also significantly superior to other early-warning models, which proves that the proposed improved SVM model has best prediction performance.4. A research on the parameters of the improved SVM. There is a close relation between building model and setting parameters. This paper explores the impact of three parameters including the penalty parameter, the number of neighbor samples, the unbalanced level, on the prediction performance of the improved SVM model. The result shows that there isn’t obvious impact on the prediction performance of the improved SVM model with the change of the penalty parameter C, the number of neighbor samples k, the unbalanced level ?, which proves the improved SVM model has stable prediction performance. But when setting parameters, the parameter ? is prevented to set as 0 or 1. Meanwhile, when the penalty parameter C, the number of neighbor samples k, the unbalanced level C are set as 0.5, 5 and 0.2 respectively, the early-warning model of the improved SVM has the best prediction performance.Through above experiments, this paper believes that the improved SVM model is the optimal operation tool and method to deal with and prevent the extremely financial risk for financial risk managements and investors. Financial risk managements can use the improved SVM model to accurately predict the extreme risk of financial market in future a period time. And then develop and implement related macroeconomic policies to prevent financial risk crisis in order to build the “firewall” to prevent financial risks to maintain the stability of the financial market, and promote the sustained and healthy economic development. Simultaneously, the investors can use the improved SVM model to capture the risk crisis of financial market, and then to adjust the strategies of financial asset investment to optimize the investment portfolio of financial assets, thereby manage financial assets more effectively and preserve even increase the value of the financial assets.