A novel approach for dynamic modeling and forecasting of realized covariance matrices is proposed. Realized variances and realized correlation matrices are jointly estimated. The one-to-one relationship between a positive de nite correlation matrix and its associated set of partial correlations corresponding to any vine speci cation is used. A method to select a vine structure, which allows for parsimonious time-series modeling, is introduced. The predicted partial correlations have a clear practical interpretation. Being algebraically independent they do not underlie any algebraic constraint. The forecasting performance is evaluated through investigation of six-dimensional real data and is compared to Cholesky decomposition based benchmark models.
Personal website of Yarema Okhrin