The VaR Risk Measurement Based On HMM And Its Empirical Analysis

In the new century, especially since the global financial crisis in2008, theinternational and domestic financial markets have undergone profound changes andtheir risks have increased sharply, so analyzing the characteristics of financialvolatility and measuring financial market risks are of great significance both forinvestment and regulation. The method of VaR (Value at Risk) provides investorswith intuitive and comprehensive quantitative indicators of risks for its highgeneralization ability, which has become the world’s mainstream risk measure. Inthis paper, we study the improvements and applications of VaR models of thefinancial market risk measure for the characteristics presented by financial datasuch as high peak and fat tail, volatility persistence, structural transformation andso on.Firstly, we elaborate briefly the research background and significance of thispaper, domestic and foreign research status about the VaR models, and put forwardthe main content of this paper. Secondly, we introduce the basic concept andalgorithm of the hidden markov model, and point out its applications in theabnormal state detection. Then the fundamental principal and the method ofcalculation of VaR are explained, and the insufficiency of the common used methodof ARCH model to estimate volatility is indicated. On this basis, we put forwardHMM-ARMA-GARCH model in this paper, which uses the hidden markov model’sstate variables to describe the normal state fluctuations and abnormal statefluctuations in financial markets, and makes the unobservable state variables be agood explanation to the fluctuation of agglomeration phenomenon. We alsoestablish the ARMA-GARCH model to estimate volatility for each state sequencerespectively and give the specific steps to calculate VaR. Finally, we analyze theShanghai enterprise debt index sequence by using the presented model and thetraditional ARMA-GARCH model respectively. The accuracy of the VaR is testedby the Kupiec failure frequency method. The results show that the presented model has a good estimate effect, and reduce effectively the problem of overestimatefluctuation persistence of the GARCH model.