Analytical nonlinear shrinkage of large-dimensional covariance matrices (Q2215772)

This note provides a closed form expression for the nonlinear shrinkage estimation of large covariance matrices. The work can be viewed as the analytic estimation analogue of \textit{O. Ledoit} and \textit{M. Wolf} [J. Multivariate Anal. 88, No. 2, 365--411 (2004; Zbl 1032.62050)] in which estimation is accomplished by numerical techniques. As opposed to Ledoit and Wolf [loc. cit] where linear shrinkage estimation is employed, the present approach relies on the connection between nonlinear shrinkage and nonparametric estimation of the Hilber transform of the sample spectral density. The advantage is the high accuracy and increased speed of the approach as well as the fact that it covers the case where the dimension exceeds the sample size.