Abstract: GARCH-type models have been broadly used to forecast volatility.But it's ignored to evaluate the performance of volatility forecasting. The reason is mainly lack of appropriate benchmark to evaluate. We estimate and forecast the return of SZZS using GARCH-type models. Realized volatiliky is computed as benchmark using 5-minuets high frequency data. Volatility forecasting performance is measured using M-Z regression and loss function. The conclusion is that GARCH type models have a very goodforecasting performance both in sample and out of sample, and GJR(1,1) under skewed t-distribution assumption is the most powerful to forecast.

Key words: GARCH:realized volatility, M-Z regression, loss function