The Research Of Stock Volatility Model Based On ICA-NN-GARCH

With the growth of China's financial market,the position of the stock market in the financial market is more important.Fully understanding the existence of risks and preparing for risk prevention are conducive to the stability of the stock market.Stock volatility is an important indicator for measuring financial risk.Calculating stock volatility can predict the future trend of stocks,providing investors and managers with more precise choices and risk prevention.Scholars have made great progress in the study of stock volatility,but there is still room for improvement in prediction accuracy.Therefore,this paper uses the 10 stocks of the 5 sectors of the old and new kinetic energy to study the sample and establish a more accurate model to forecast stock volatility.Due to the complexity of high-dimensional financial time data,the use of a single generalized autoregressive conditional heteroscedasticity model(GARCH model)to deal with high-dimensional data will have problems such as low prediction accuracy and inaccurate accuracy,and the GARCH model also has too many parameters to be estimated.The calculation is not concise enough,so this paper uses a multivariate volatility model to predict stock volatility more accurately and effectively.Firstly,this paper selects the independent component analysis method(ICA)to extract high-dimensional information.The ICA method can extract the independent components in the data quickly and efficiently,thus playing the role of dimensionality reduction,and the ICA method has the characteristics of simple calculation and low memory consumption.Secondly,the GARCH model is used to eliminate the heteroscedasticity caused by time series.The GARCH model has a great advantage in dealing with the linear part of financial data.When fitting nonlinear data,it can be combined with neural network(NN).Based on this,this paper builds the ICA-NN-GARCH model.Finally,the ICA-NN-GARCH model is used to statistically describe the volatility of stocks of 10 listed companies in China.The experimental results show that the stock yield series exhibits characteristics such as peak thickness,skewness and kurtosis,as well as stability and The ARCH effect satisfies the necessary conditions for modeling,and the stock return series shows the characteristics of volatility clustering.Using the one-step prediction method based on rolling window to predict the value of risk(VaR),and using the maximum likelihood statistic LR of the Kupiec failure rate test to test the prediction accuracy of VaR,the test results show that although The GARCH model,the ICA-GARCH model and the ICA-NN-GARCH model can predict the VaR value,but the ICA-NN-GARCH model has higher prediction accuracy than the GARCH and ICA-GARCH models.Therefore,the ICA-NN-GARCH model can More accurate prediction of risk values,which in turn can provide a reference for a large number of stock investors.