In this paper we introduce a new method of forecasting covariance matrices of large dimensions by exploiting the theoretical and empirical potential of using mixed-frequency sampled data. The idea is to use high-frequency (intraday) data to model and forecast daily realized volatilities combined...