The volatility specification of the Markov-switching Multifractal (MSM) model is proposed as an
alternative mechanism for realized volatility (RV). We estimate the RV-MSM model via Generalized
Method of Moments and perform forecasting by means of best linear forecasts derived via the
Levinson-Durbin algorithm. The out-of-sample performance of the RV-MSM is compared against
other popular time series specfications usually employed to model the dynamics of RV as well as other
standard volatility models of asset returns. An intra-day data set for five major international stock
market indices is used to evaluate the various models out-of-sample. We find that the RV-MSM seems
to improve upon forecasts of its baseline MSM counterparts and many other volatility models in terms
of mean squared errors (MSE). While the more conventional RV-ARFIMA model comes out as the
most successful model (in terms of the number of cases in which it has the best forecasts for all
combinations of forecast horizons and criteria), the new RV-MSM model seems often very close in its
performance and in a non-negligible number of cases even dominates over the RV-ARFIMA model.