关键词: 已实现协方差矩阵, LASSO-CDRD, HAR-DRD, 均值方差模型

Abstract: The accurate prediction of high-dimensional covariance matrix is important for investment portfolio and risk management. In this paper, we use financial high-frequency data to obtain the realized covariance matrix, apply DRD decomposition on the realized covariance matrix and vectorize the realized volatility matrix D. To ensure the positive definiteness of the predicted values of the realized correlation coefficient matrix R, this paper applies Cholesky decomposition to matrix R, and vectorize the decomposed matrix. This paper employs Vector AutoRegressive (VAR) to model the two sets of vectors separately, uses LASSO method to estimate the parameters of high-dimensional VAR, establishes the dymanic models of realized volatility matrix D and the realized correlation coefficient matrix R, and constructs the LASSO-CDRD dynamic prediction model of covariance matrix. This paper also uses the mean variance model to conduct the economic evaluation of the covariance prediction models. The empirical analysis shows that, compared with other covariance prediction models, the LASSO-CDRD covariance matrix prediction model has higher prediction accuracy and Sharpe ratio, and its overall performance is the best.

Key words: Realized Covariance Matrix, LASSO-CDRD, HAR-DRD, Mean Variance Model