This paper employed three kinds of different statistical methods, parametric, nonparametric, and semi-parametric methods, to investigate the precise of predicting the one-day-ahead value-at-risk (VaR) measured by EWMA, GARCH, EGARCH, TGARCH, APARCH, HS, and FHS models in three types of markets (stock exchanges, commodities, and exchange rates). Unlike prior studies focused on the long trading position only, we calculated both for short and long trading positions of the VaR. Finally, the performances of all models were tested by the evaluating methods of conservatism, accuracy and efficiency. The empirical results show that TGARCH model has the best forecasting performance.