| Since the collapse of the Bretton Woods system, the collapse of the financialinstitutions of developed countries and the case of losses, especially since2008, the globalfinancial crisis caused by America’s subprime crisis, the market risk becomes one of themost important risks faced by financial institutions. At present, the financial market riskmeasuring methods include sensitivity analysis,volatility, VaR, pressure test and extremevalue theory. VaR is the mainstream of financial market risk measurement model.Traditional risk measurement methods to the characteristics estimated on the tail of theincome distribution, such as VaR,has certain defects, and it is also sick to the measurementof the risks of extremes. So scholars introduced the extreme value theory to analyzefinancial market risk and estimating the losses under the extreme market conditions fromanother angle. Also, the age of big data information storm has been changing our thinking,life and work,and also puts forward higher requirement to the data processing methods,statistical indicators and the choice of distributions. How to measure the financial marketrisks based on high-frequency data more correctly becomes the new question for study.In this paper, the dynamic VaR were estimated,based on the high frequency financialtime series and conditional extreme value distribution. First of all, this paper established theBSP-HAR-RV model to select the optimal frequency. China’s stock markethigh-frequency-data’s frequency is per minute.Select the optimal trading frequency through drawing the average double power variation realized volatility scatter plot, and at the sametime considering the effects of measurement errors and market microstructure. The modelchooses the double power variation realized volatility as index, effectively predicts the jumppoint that can be possibly, range frequency from0to240, draws the average doublepower variation realized volatility scatter plot and select the optimal frequency. This articlehas also established BSP-RM-RV-POT model to estimate the dynamic VaR of highfrequency financial time series. In order to calculate the under quantile of the incomedistribution, get the logarithm yield of sample first, then take a minus sign on the logarithmyield sequence, and bulid the model on the basis of the negative yield distribution series. Byrealized mean and realized volatility estimation of autoregressive model, get the residualsequence. Estimate the sequence of the disturbance through POT of conditional EVT.The research results show that under the optimal frequency, the parameters of themodel fitted are all passed the test of significance and the fitting results can predictvolatility trends. With the comparison between5min and10min resluts, the frequency of f=10min has the best significance level and significance level of the variables are0.01, andcompared with the optimal frequency f=11min, the result of the weeks of realizedvolatility significance level is0.10. But from the perspective of the overall situation of themodel, the residual error of the results of the optimal frequency model is0.0004895, lessthan the model of residual standard error by f=5min and f=10min and the result iscoming to be better. Therefore, from the overall, the choice of the optimal frequency canimprove the accuracy of the model. BSP-RM-RV-POT estimates the dynamic VaR of highfrequency financial time series, by "realized mean" and "realized volatility" estimation ofautoregressive model to get the residual sequence, get the fitting figure GDP plot bygeneralized pareto distribution (GDP) being carried out on the residual error sequencefitting. From the four picture of the GDP model diagnosis, the estimated results of themodel is appropriate, is that,in full consideration of the influence of time on the average sequences, the model can still be applied to dynamic VaR estimation. |
PDF Full Text Request