Abstract:  The volatility accuracy of several volatility forecast models is examined for the case of daily spot returns for the Mexican peso - US Dollar exchange rate. The models applied are univariate GARCH, a multi-variate GARCH (the BEKK model), option implied volatilities, and a composite forecast model. The composite specification includes time-series (ARCH-type) and option implied volatility forecasts. Different to most of the literature, this paper includes a statistical evaluation of the forecast accuracy of a composite model and models that are not combined. The results show that the composite volatility forecasts are superior to the other models in terms of mean squared errors (MSE). In forecast evaluations of the MSE it was found that estimates were statistically significantly different between composite forecast estimates and its counterparts. According to these results conclusions are as follows: the composite model is superior and both type of data -historical and implied in option prices- must be used when available.