Large-Dimensional Portfolio Selection with a High-Frequency-Based Dynamic Factor Model

Abstract This article proposes a new predictive model for large-dimensional realized covariance matrices. Using high-frequency data, we estimate daily realized covariance matrices for the constituents of the S&P 500 Index and a set of observable factors. Using a standard decomposition of the joint covariance matrix, we express the covariance matrix of the individual assets similar to a dynamic factor model. To forecast the covariance matrix, we model the components of the covariance structure using a series of autoregressive processes. A novel feature of the model is the use of the data-driven hierarchical clustering algorithm to determine the structure of the idiosyncratic covariance matrix. A simulation study shows that this method can accurately estimate the block structure as long as the number of blocks is small relative to the number of stocks. In an out-of-sample portfolio selection exercise, we find that the proposed model outperforms other commonly used multivariate volatility models in extant literature.