HAR-based realized volatility clustering

Financial time series are often clustered considering conditional volatility, estimated from GARCH models that rely on daily squared returns. Realized mea- sures provide, however, a better estimation of the volatility. Consequently, clustering approaches based on realized volatility should be preferred. Assuming that real- ized volatility dynamics can be explained by different trading frequencies of market partecipants, we propose a new approach for fuzzy clustering of financial time series based on the Heterogenous Autoregressive Realized Volatility (HAR-RV) model. We perform an empirical analysis on the clustering structure of the U.S. stocks belonging to the Dow Jones Industrial Average (DJIA) index.