Overall, the empirical results show that GJR with all the three distributions seems to provide superior forecasting performance at all oneday, five day ten-day, fifteen-day and twenty-day-ahead volatility forecasts horizons according to the statistical loss functions. So, it may be concluded that the asymmetric effect is central to estimating the quadratic effect for forecasting. The symmetric GARCH model performs poorly according to the statistical loss functions, especially at shorter forecast horizons. Moreover, non-normal distributions, generally, provide better out-of-sample results than the normal distribution.