This paper evaluates several GARCH models regarding their ability to forecast volatility n Tehran Stock Exchange (TSE). These include GARCH models with both Gaussian and fattailed residual conditional distribution, concerning their ability to describe and forecast volatility from 1day to 22day horizon. Results indicate that AR(2) MRSGARCHGED model outperforms other models at oneday horizon. Also, the AR(2) MRSGARCHGED as well as AR(2)MRSGARCHt models outperform other models at 5day horizon. In 10 day horizon, three models of AR(2)MRSGARCH outperform other models. Concerning 22 day forecast horizon, results indicate no differences between MRSGARCH models with that of standard GARCH models. Regarding Risk management outofsample evaluation (95% VaR), a few models seem to provide reasonable and accurate VaR estimates at 1day horizon, with a coverage rate close to the nominal level. According to the risk management loss functions, there is not a uniformly most accurate model.
