The Comparison Of Different VaR Methods: An Empirical Research Based On The Chinese Stock Market

At present, Value at risk is a wildly used method to valuate the market risk. This thesis considers three classes of Value at risk method: parametric method, non-parametric method and semi-parametric method. After introducing the three kinds of VaR methods, firstly, we choose fifteen VaR models which contain eight parametric methods, four non-parametric methods and three semi-parametric methods to calculate the VaR number, using ShangZheng(code:0000001) and ShenCheng(code: 399001) index in Chinese stock market. Then, in order to compare and evaluate the performance of these VaR models, We try to construct a comprehensive backtesting system which can evaluate a VaR models from three aspects—accuracy, conservatism and efficiency. Using this system and the recent data of the two indexes, we compare the out-of-sample predictive performance of specific implementations of each of the fifteen VaR models in the confidence level 95% and 99%. The result of backtesting tells us that none of the available methods produces a uniformly superior risk forecasts for all performance criteria, however, it is possible to draw some broad conclusions. The first conclusion is that, the performance of the models considered in this paper is not greatly dissimilar across most of the performance criteria, excluding some exceptions. Secondly, in 95% confidence level, EWMA and GARCH-t have better performance compared to any other methods in general; GARCH-GED and. EGARCH-M seem a bit conservative while Hybrid-HS and bootstrap2 are more positive. Finally, in 99% confidence level, across three dimensions of model performance (conservatisnvaccuracy and efficiency), SMA is a better model. However, be a positive risk manager, index-HS and bootstrap2 may be the better choices. On the contrary, the GARCH-t and GARCH-GED methods are more attractive to supervisors and negative risk managers because of their extreme conservatism.