Harnessing the decomposed realized measures for volatility forecasting: Evidence from the US stock market

Abstract

This study explores the predictive ability of three decomposed realized measures for the US stock market using the mixed data sampling (MIDAS) framework. From the in-sample analysis, we find that all the decomposed realized measures have a significant positive impact on future stock volatility. Moreover, the predictive model, including moderate and extreme volatility, outperforms the related competing models via out-of-sample analysis. We also investigate the predictive sources of moderate and extreme volatility by considering sub-sample and high and low volatility level, and find that the main ability of them is reflected in the low fluctuation period. Furthermore, using a portfolio exercise, we show that the decompositions of moderate and extreme volatility can substantially increase the economic value. Finally, we extend our empirical analysis considering different forecast horizons and non-linear model with regime-switching.