Research On VaR Measurement Of Financial Risk Based On Quantile Regression Method

With the development of the financial market,financial risks are also increasing,and the impacts for a country's economy are increasingly difficult to ignore.How to accurately measure financial risks is particularly important.Among many risk measurement models,VaR(value-at-risk)is the most popular risk measurement tool at this time.How to accurately measure financial risk is of great significance to the healthy development of the country's financial market.Currently,VaR measurement methods are mainly divided into the parameteic,non-parametrik and semi-parametric methods,such as the Generalize AutoRegressive Conditional Heteroscedasticity model and MonteCarlo simulation method.Many scholars use the parametric method to measure the risk of VaR,however,as financial data generally have the characteristics of peak and thick tail,the correct estimation of parameters becomes more difficult to grasp.Based on this,the paper chooses the Quantile regression model and neural network to study the risk measurement of stock market VaR.Firstly,by Guizhou Bailing stock and Kweichow Moutai stock as an example,Under 95% confidence level,the VaR of these stocks were measured based on the TGARCH model,quantile regression(QR model)and quantile regression model of neural network(QRNN model)were established to carry out measurement research.the Empirical results show that,by improving the sample partition algorithm,for the three VaR risk measurement models of Guizhou Bailing stock and Kweichow Moutai stock,the traditional TGARCH model has the worst measurement accuracy of VaR on the 95% confidence interval,and the effect of the model is not as expected.The results of the quantile regression model ranked second,which was an improvement compared with the effect and accuracy of the traditional parametric model.However,the risk measure of VaR by neural network quantile regression method is the most expected.Secondly,in order to further explore the VaR risk measure,this article has selected two sets of data: the Shanghai composite index and the Shenzhen component index.Using the methods of Quantile regression and neural network Quantile regression to measure under the 95% confidence level of the VaR study,and at the same time considering the extreme VaR condition of 99% confidence level for measure research,the general Quantile regression and neural network Quantile regression methods in the same conditions are compared.The results found that the Quantile regression of neural network is more ideal than the general Quantile regression for both the VaR measure under 95% confidence level and the extreme VaR measure under 99% confidence level.Research shows that the Quantile regression of neural network model is superior to the traditional TGARCH model and the general Quantile regression model in measuring financial risk VaR.