Research On Extreme Risk Warning For Financial Market Based On SVM With Unbalanced Data Sets

In recent years, several outbreaks of the global financial crisis not only make the economy of different countries suffer a heavy blow, but also make financial risk management face more severe challenges. As for China's financial markets, the time of its establishment is very short, and it lacks mature experience to deal with the financial risks. However, the close relation between Chinese financial market and foreign financial markets further increases the likelihood of outbreak of extreme financial risk, which makes it face test of more challenging risk. From this view, to explore practical extreme risk warning model of China's financial markets, and to predict and prevent extreme financial risks are vital.Based on the analysis above, this paper uses the CSI 300 index as the object of the study, and first builds the warning indicator system of the model. Through a combination of the facts of financial crisis and EVT, it appraises extreme financial risk sample with the non-extreme financial risk sample and determines the status indicators. According to the internal characteristics of Chinese financial markets, and the extreme financial risks linkage effects, based on the initially selected the indicators of internal and external risk characteristics successfully extracted evoked internal indicators including opening price, closing price and trading volume of Chinese financial markets extreme risk by using the T test and K-S test. Next using Clayton-Copula function analysis of China's financial market risk conduction effect with the rest of the international financial markets, and thus extract external characteristic indicators of significantly inducing the outbreak of Chinese extreme financial risk, including HSI, KOSPI and TWII. After using state index and the extracted features internal and external risk indicators constitute early warning indicator system, builds SVM artificial intelligence technology to do extreme financial risk early warning research. For SVM under unbalanced data, ineffective forecast problem, combine the introduction of the unbalanced sample processing method and SVM to strive to enhance the performance of SVM model predicting unbalanced samples. It combines the Borderline-SMOTE and EasyEnsemble, the two sampling methods to construct an improved SVM China extreme financial risk early intelligent warning model Borderline-SMOTE-EasyEnsemble-SVM, and does further experimental design, proving the model's superior in extreme financial risk forecast performance. Next, the main contents of the experiment will be introduced:1. Compare the improved SVM in predicting the performance under different unbalanced sample data sets. Choose Balance-Scale dataset, Contraceptive dataset, Haberman dataset, Hepatitis datasets and Pima dataset in the international UCI standard database in machine learning to test the improved SVM performance. Borderline-SMOTE-EasyEnsemble-SVM has achieved excellent results forecast for each group of data sets. It fully illustrates the stable performance of the forecasting model and for the five groups from other areas and different levels of non-equilibrium data, it has achieved good prediction effect, which shows the strong feasibility of being applied to the extreme risk warning for the Chinese financial market.2. Determine the optimal parameters of the model. The selection of the parameters will have a key influence on the effect of established Borderline-SMOTE-EasyEnsemble-SVM warning model, so we need to find the optimization of all the model parameters. In this paper, it uses a grid search method to find the SVM optimization of parameters, and determines the Borderline-SMOTE Nearest Neighbor algorithm parameters k by trial and error, synthetic extreme risk parameter ? controls the number of samples, EasyEnsemble algorithm independent from the majority class samples random sampling T and si the number of iterations on Adaboost algorithm. The results showed that the use of synthetic methods Borderline-SMOTE extreme risk of changes in the number of samples will model prediction accuracy have a greater impact, and because the model is more stable, changes in other parameters of the model to predict the results less affected. The artificial synthetic samples numbers determined by ?= 0.8 make the best results of model predictions.3. Compare the improved SVM prediction performances of the extreme financial risk. The defined extreme and non-extreme financial risks constitute a typical unbalanced sample sets, and SVM has poor prediction effect of the unbalanced samples, so based on these, study the prediction performances of Borderline-SMOTE-EasyEnsemble-SVM, Borderline-SMOTE-SVM, EasyEnsemble-SVM, SVM for the extreme financial risks. Results show that, Borderline-SMOTE-EasyEnsemble-SVM has good stability, and is well suited for extreme risk warning Chinese financial market, and the combined of Borderline-SMOTE and EasyEnsemble will improve SVM prediction performance at the greatest degree in unbalanced data.Based on these empirical results, this paper thinks that the improved SVM model, Borderline-SMOTE-EasyEnsemble-SVM is capable of accurate prediction of extreme risks of financial markets and has high practical value. For financial and economic management, it is able to use the model to accurately predict whether extreme financial risks occur in the future, so as to formulate appropriate economic policies to resist the impact of the financial risks and maintain financial order and stability and ensure the safe operation of the financial markets, and ultimately promote macroeconomic stable and healthy development. For investors, it is possible to use the model to predict the financial products extreme risk to predict and make more effective risk management based on the forecast results. They can avoid possible extreme financial risk, and reduce losses as much as possible and increase profitability in finance and investment process through the timely adjustment of financial assets, investment strategies.