Recently, the financial industry in Taiwan increasingly uses Value-at-Risk (VaR) in portfolio risk management, risk capital allocation and performance attribution. Risk managers are rightfully concerned with the precision of VaR and the related expected shortfall (ES) techniques. The purposes of this thesis are to estimate the predicating accuracy of VaR and ES computed by different models, and to assess the ex ante magnitude of the error through the construction of confidence intervals around the VaR and ES measures. We apply GARCH to build dynamic variation models, and use normal and t distributions to construct the Hill, FHS, and GCCF estimators. Then, the bootstrap method is used to assess the predicating accuracy of VaR and ES. Finally, by analyzing simulated data, we conclude the FHS and Hill estimators have the higher predicating accuracy in estimating VaR and ES. The result should be useful in practice.