Estimation And Analysis On The VaR Of Chinese Stock Market Based Upon ARCH Models

The VaR model was born in 1990's, which has only ten years history. Although it is still a very young member of empirical finance family, it has brought financial risk management substantial improvements and draws pretty much attention from both acedemia and practitioners. Accuracy of VaR measurement is still a big concern in application. This is because of the complicity in the VaR model, which is related to both the distribution and volatility of equity, and the volatility of equity return is featured as conditional heteroskedastic. This means that the estimation procedure of VaR should appropriately consider with these two issues. Fortunately, the very prominent ARCH model techniques give us the possibility of tackle with these difficulties. ARCH was originally introduced to deal with the problem of heteroskedasticity in econometric modeling, and it can help to estimate time varying conditional variance under various assumption of residual distribution, for example the standard econometric package EViews gives users three options, i.e. Gaussian, Student t and Generalized Error Distribution. Therefore, combining ARCH model with VaR estimation is a very hopeful direction to improve the precision of VaR estimation. This paper is thus motivated by this thinking and tries to apply the new combined models to estimate the VaR of Chinese stock market.The thesis first addresses the related conceptual framework of finance, a introduction of the financial risk and the finance risk management, and then summarizes several kinds of risk measure methods and carries out detailed review on VaR methods. The idea of introducing ARCH models into the VaR estimation is justified by concrete step of how to use ARCH models to calculate VaR. Empirical studies on the daily closing price of Shanghai SE Composite Index, Shanghai SE 180 Index, Shenzhen SE Component Index are carried out to illuminate the new combined model. Sample data covers August 2002 to September 2006, 1010 observations in total. The ARCH model fitting outputs, conditional residual series under various error distribution assumptions are used to calculate 95% and 99% quintiles, which in turn are applied to estimate the daily VaR of the sample index, inference and analysis afterwards. Finally the thesis proposes how to use this more effective methodology in the realistic finance world. Further discussion on the feasibility to research is included and a suggestion of developing a set appraisal system of ARCH based VaR is provided.