Predicting Volatility: Getting the Most Out of Return Data Sampled at Different Frequencies