| With the rapid development of Internet finance,the measurement of its risk becomes an urgent and important need.At present,the study on the measurement of internet financial risk is still in the exploring phase.This thesis mainly studies the characteristics of the return rate of Fortune Treasure,which is an representative product of internet financial.Two models are established after analyzing the statistical characteristics of the return rate of Fortune Treasure in recent 4 years.One is to a bayesian POT risk measurement model to measure the tail risk of the return rate,the other is to a bayesian quantile autoregressive model to measure the overall risk of the return rate.The study of this thesis mainly includes the following three parts:In the first part,the basic statistical characteristics of the return rate of Fortune Treasure are analyzed,and the basic type of the model is set.Firstly,the operation mechanism of fortune treasure and the statistical characteristics of its return rate are analyzed.It is found that the series is non-normal,which has a long tail,and its trailing property of autocorrelation coefficient and truncation property of partial autocorrelation coefficient.Secondly,Peaks Over Threshold model(POT model)is used to describe the tail risk of return rate according to its long tail characteristics,the quantile autoregressive model is used to describe the overall risk according to the trailing and truncated.In the second part,the POT model is introduced to calculate the VaR of the tail risk.Firstly,according to the POT model set in the first part,the basic risk measurement formulas of VaR and ES with three undetermined parameters are derived.The threshold parameter was determined by the sample mean excess function,and the other two parameters were solved by the MCMC algorithm of bayesian estimation.Thus,the bayesian POT model measuring VaR and ES is completely determined,which can be seen in expression(3.18)and(3.19).Secondly,the model established above is used to calculate the VaR and ES of Fortune Treasure's return rate,and the results are shown in table 3.2 and table 3.3.Comparison between the calculation results of bayesian POT model and the results of POT model based on maximum likelihood estimation show that the results of bayesian POT risk measurement model are more higher.Through the backtest for VaR,although both of the two estimation methods have passed the test,bayesian estimation method fails less,so the bayesian POT risk measurement model is better.Finally,this thesis forecasts the VaR of the Fortune Treasure's return rate from January 1,2019 to March 31,the VaR values predicted at 99%,97.5% and 95% are 5.7366,7.7176 and 11.4852 respectively.In the last part,the overall risk of fortune treasure was measured by the bayesian quantile autoregressive risk measurement model.Firstly,according to the quantile autoregressive model set in the first part,the basic formula of VaR risk measurement is given,and the autoregressive order of the model is determined according to the AIC criterion.The bayesian quantile autoregressive risk measurement model is established while estimating the parameters of the model at different quantile levels by bayesian estimation.Secondly,the VaR of the fortune treasure's return rate from January 3,2015 to December 31,2018 is measured through this model.The value in December 2018 is given in table 4.6,and the VaR trend of the whole sample is given in figure 4.4.The model has passed the backtest for VaR at each level of quantile.Compared with traditional estimation methods,bayesian estimation method has a lower failure rate in risk measurement and is more reliable.Finally,the VaR of the Fortune Treasure's return rate from January 1 to March 31,2019 is predicted in table 4.8 and figure 4.5 through bayesian quantile autoregressive risk measurement model.By comparing the results of the above two established models,it is found that the bayesian POT model is more suitable to describe tail risk,while the bayesian quantile autoregressive model is more suitable to measure overall risk. |
PDF Full Text Request