This study is to explore the in-the-sample predictive ability and to compare the out-of-sample forecasting performance by using ten stock price indexs and six time series models. We use two loss functions and DM statistic to evaluate and compare the volatility predictive ability for future fluctuation. Furthermore, we consider a version of short-term volatility by means of GARCH (1,1) model and introduce the possibility of an asymmetric volatility effect of GJR (1,1) model. The best model decision is sensitive to different evaluations. If we use sum of daily squared returns as realized volatility, our conclusion is that the GJR (1,1) model has relatively better forecast ability no matter in the long-term and short-term.