Composite forecasting of vast-dimensional realized covariance matrices using factor state-space models

Abstract We propose a dynamic factor state-space model for the prediction of high-dimensional realized covariance matrices asset returns. Using block LDL decomposition joint matrix assets and factors, we express individual similar to an approximate model. parts, i.e., residual covariances as well loadings, independently via tractable approach. This results in closed-form Matrix- F predictive densities distinct elements Student’s t loadings. In out-of-sample forecasting portfolio selection exercise compare performance proposed under different specifications dynamics. These includes diagonal residuals based on GICS sector classifications strict diagonality assumptions combinations both using linear shrinkage. find that performs very empirical application 225 NYSE-traded stocks well-known Fama–French factors sector-specific represented by exchange traded funds.