Accurate volatility predictions are crucial for the successful implementation of risk management. The use of high frequency data approximately renders volatility from a latent to an observable quantity, and opens new directions to forecast future volatilities. Our goals in this paper are to find a general powerful forecasting procedure for volatilities based on high frequency data, to evaluate the predictive potential of volatility forecasts for the true latent volatility, and to analyze the impact of more reliable volatility predictions on the quality of two widely used risk measures. For that purpose, we explore the performance of various models and modern prediction tools.