Font Size: a A A

VaR Based On EVT And Its Application In Chinese Stock Market Risk Management

Posted on:2007-08-19Degree:DoctorType:Dissertation
Country:ChinaCandidate:W L YuFull Text:PDF
GTID:1119360242962691Subject:Western economics
Abstract/Summary:PDF Full Text Request
In recent years, factors like globalization of the economy, liberlization of the finance, competition and unregulation, financial innovation and technological advancement and so on, have rapidly enlarged the financial market scale and have significantly improved its efficiency, and as well as have largely increased its volatility and risk. Chinese have entered WTO, and with interest rate's marketization, capital item's openness and derivative financial market's establishment, market risk faced by financial assets will increasingly be complicated. Because of characteristics of quantify, synthesis, earthliness of VaR's measuring risk, VaR has been widely applied in many banks, financial institutions and supervision institutions, and now has been becoming the international standard. Therefore, introducing VaR to Chinese is practically very important. Considering no model which can evaluate effectively and exactly the VaR of Chinese stock market risk in every confidence level, the dissertation mended some models in different aspects in order to evaluate effective and exact VaR in every confidence level based on comparing some models evaluating VaR. Focusing on the daily return data sample of Shanghai Securities Trade Market's Complex Index and Shenzhen Securities Trade Market's Component Index from 1995 July to 2005 Nov., the dissertation adopted mainly empirical and comparative analysis method to analyse Chinese stock market risk based on using widely other people's research results for reference.To evaluate VaR exactly, appropriate models must be choosed to fit distribution of returns series. So the dissertation firstly analysed statistic features and distribution of returns series in order to choose sound models. The analysis showed that Chinese stock returns series have fat tails and excess kurtosis, weak auto-relation and volatility clustering, and although relation between Shanghai stock market and Shenzhen stock market is changing by time, the two stock markets have the long-term stable equilibrium relationship.There are many models that can evaluate VaR. The dissertation adopted several usual models like the normal model (including history average model and exponentially weighted moving average model (EWMA)) and GARCH model of parameter methods and the history simulation model of non-parameter methods and the extreme theory method to evaluate VaR, and used the back-testing of Kupiec failure to test validity of VaR. It was concluded that VaR based on the history average normal model is invalid in any confidence levels, VaR models based on EWMA, the history simulation and GARCH is invalid in high confidence levels and valid in low confidence levels, and VaR model based the extreme theory is on the contrary, that is, the method is valid in high confidence levels and invalid in low confidence levels. Therefore, directly using the models to evaluate VaR, valid VaR evaluation cannot be obtained in every confidence level. The dissertation mended VaR models in the following two ways.On one hand, the dissertation respectively introduced the model of bias-corrected conditional volatility to several models like the history average model, RiskMetrics EWMA model and GARCH model which are usually used to correct conditional volatility and evaluate VaR. it was concluded that VaR of bias-corrected conditional volatility is valid in every confidence levels, and more exact. But regarding GARCH model, although validity of VaR of bias-corrected conditional volatility is improved, veracity of the method isn't advanced in every confidence levels.On the other hand, while the basic assumption of POT model of the extreme theory is that excess thresholds are IID, actual excess thresholds is locally relative, which causes that the bias of VaR evaluation from the actual value is big. So two methods are used to remove the local relation. One is that extremal index is introduced to POT model. The analysis showed that the introduction of extremal index improves validity and veracity of VaR based on POT model. Still VaR based on POT model introducing extremal index is invalid in low confidence level. The other is to filter returns series using GARCH model. Since VaR based on extreme theory is valid in high confidence levels and is not as well as usual models in low confidence levels, the dissertation used history simulation and POT methods of extreme theory to fit the distribution of the residuals filtered by GARCH model to get VaR. It was concluded that VaR from the method is valid in every confidence levels, and close to the expected level.In conclusion, by directly using models in evaluating Chinese stock market risk, valid VaR evaluation cannot be obtained in every confidence level. However, regarding the history average model and the RiskMetrics EWMA model, VaR of bias-corrected conditional volatility is valid and exact in every confidence levels. Or, the mixture method of history simulation and extreme theory filtered by GARCH model can evaluate VaR validly and exactly in every confidence levels. And the later is better than the former.
Keywords/Search Tags:risk management, Value at Risk (VaR), extreme theory
PDF Full Text Request
Related items