This study uses 12 bivariate generalized autoregressive conditional heteroskedasticity (GARCH) models to forecast the volatility of stock-based portfolios in Asia, and then evaluates the forecast performance for the above models. Empirical results show that, the Student’s t does not own better forecast performance, the standard and nonstandard approaches have the same forecast performance, and the constant conditional correlation (CCC) has the best forecast performance among three covariance specifications. Regarding the six GARCH models composed of three covariance specifications and two parameter estimate methods the CCC specification with the standard approach has the best forecast performance irrespective of return distribution setting.