Stock prices are one of the most volatile economic variables and forecasting stock prices and their returns has proved very challenging, if not impossible. In this paper, we apply a battery of linear and nonlinear models to forecast the returns in nine international stock exchanges for the period 1998-2008. The models are random walk, historical mean, moving average, exponentially something, AR, and GARCH class models including ARCH, GARCH, GJR- GARCH" and EGARCH. Volatility is defined as within- month standard deviation of continuously compounded daily returns (log-returns) on the indices of main stock exchanges. We compare the forecasting results of the eight major international stock exchanges with the Tehran stock exchanges (TSE), where the market is highly regulated and therefore less subject to volatility. To evaluate the forecasting results, we use three symmetric loss functions including the mean absolute error, root mean squared error, and the mean absolute percentage error. Results suggest that the GJR- GARCH model provides the superior forecasting performance in comparison with other volatility forecasting models in international exchanges. However, the simple smoothing model provides superior performance in TSE. While random walk model provides the worst performance for international exchanges, it is a good performing model, second in order, in TSE. Historical average model provides the worst performance and ARCH class models do not rank high in forecasting competition for TSE.