VaR Models Based On Different Distribution

The paper takes the return series in stock market, futures market and foreign exchange market for example, comparing the performance of out-of-sample VaR measurement of some main VaR models. The empirical results show: the parametric VaR models under the normal distribution assumption can’t get the satisfying results in the VaR measurement since the normal distribution can’t embody the skewed, leptokurtic and fat tailed characteristics of return series. However, the parametric VaR models under the AEPD, AST, SKT and ALD distribution assumption have done well in the VaR measurement since these distributions can describe the stylized facts of financial assets’ return. As for CAViaR model, it doesn’t have obvious advantages in the VaR measurement comparing other parametric models. At last, this paper also shows that the financial assets’ returns of futures market and foreign exchange market don’t have obvious leverage effect.