VaR Measurement Method Based On Hidden Markov Model

Financial sector in China is developing very fast and more openable to overseas’ investors. Many new financial products appear in the market. In this environment, the risk our financial sector facing is getting more and more complicated. The drastic drop of A-shares during Jun, 2015 to Sep, 2015 also reminds us the potential risk of our financial system. Analysis of financial market volatility structure and forecast of financial risk are quite important for both investors and regulators. Va R(Value at Risk) provides an intuitive, comprehensive and forward-looking risk measurement to investors and regulators. Under the promotion of Basel Committee, Va R has been adopted by various financial institutions. In this paper, we propose a Va R measurement method based on Hidden Markov Model to take mechanism transition of financial market and non-normal feature of financial return into account.Firstly, we propose the research background and significance of this paper, and then review the domestic and foreign research status of Va R models, find that models which take mechanism transition of financial market into account can’t handle the tail risk very well, especially left tail risk and models which take the tail risk into account can’t capture the mechanism transition of financial market. We introduce Hidden Markov Model and Gaussian Mixture Model to capture the mechanism transition of financial market and the tail risk at the same time. Based on that, we form the HMM-GMM-Va R method.We select the Va R measurement method proposed by Yanfei Wu in 2013 which combine HMM and GARCH together to estimate Va R and Va R measurement method based on GMM as compare group. After that we use simulation data to evaluate the accuracy of Va R models and find that the accuracy of HMM-GMM-Va R and GMMVa R are very close, both models outperform HMM-GARCH-Va R. Then we use Shanghai Composite Index to conduct empirical research and test the performance in reality of Va R models use Kupiec test and Christofferson test, find HMM-GMM-Va R outperform other models. Lastly, we put Va R models in the stressed condition of Ashares during Jun, 2015 to Sep, 2015 to evaluate their performance, find that only HMM-GMM-Va R gives signal when market volatility increased significantly. If investors adopt HMM-GMM-Va R in their risk management process, they can reduce their losses.