Realized Volatility Prediction in Stock Market

In our study, we examine various MIDAS (Mixed Data Sampling) regression models as a predictive tool for stock market volatility. These models vary in the regressors used, including squared return, absolute return, realized volatility, realized power, and return range.By analyzing equity return data, we find that the ability to see daily, consisting of 5-minute absolute returns, emerges as the most influential predictor of future volatility This is measured by an increase in quadratic variation. Notably, this approach outperforms models that rely solely on observed changes, which are based on prior increases in quadratic change. Surprisingly, our findings suggest that using high-frequency (5-minute) data directly does not improve the prediction of volatility.In summary, our study reveals the effectiveness of daily realized power, using 5-minute absolute returns, as a good indicator of future volatility in the stock market. We show that this approach outperforms models that rely solely on observed changes. Surprisingly there is no bet using high-frequency data.