We evaluate predictive performance of a selection of conditional and unconditional Value-at-Risk (VaR)models. By filtering various time varying volatility models we predict risk forecasts and compare them with that of unconditional VaR models. We consider VaR models based on ARCH models, historical simulation, Monte Carlo, nonparametric quantile regressions, and extreme value theory (EVT). In addition, we also employed what is often called as conditional VaR (CVaR). The return series are standardized using various conditional variance models such as GARCH, FIGARCH, EGARCH, models. We apply these methods to the foreign exchange of several countries that suffered recent financial crisis, and compare their performances in terms of various evaluation criteria using the tests of White (2000) and Hansen (2001).