This thesis empirically compares the performance of conventional Value at Risk Models, EWMA model, GARCH(1,1) model and Historical Simulation model, under Conditional-VaR criterion and other conventional model verification criteria, Kupiec’s log-likelihood ratio and Basel rules, by using daily returns of KOSPI index from January 1998 to October 2003.
There was no difference in becktesting results based on failure rate among the three models. That is, no model is rejected under Kupiec’s log-likelihood ratio criterion and three models turned out to be in green zone which represents acceptable failure rate under Basel rule.
The difference of precision among the models, however, can be tested by comparison of absolute value of mean error(ME) of each model’s Conditional-VaR. EWMA model was superior to the other models in forecasting the conditional expectation of losses beyond the VaR.