B0908
Title: The minimum shrinkage covariance determinant estimator
Authors: Kris Boudt - Vrije Universiteit Brussel and VU Amsterdam (Belgium)
Steven Vanduffel - Vrije Universiteit Brussel (Belgium)
Tim Verdonck - KU Leuven and UAntwerpen - imec (Belgium)
Yukai Yang - Uppsala University (Sweden) [presenting]
Abstract: The Minimum Covariance Determinant (MCD) estimator is often used for the high breakdown point calculation of robust Mahalanobis distances requiring the covariance matrix to be invertible, which is problematic in high dimensions. We propose the Minimum Shrinkage Covariance Determinant (MSCD) estimator, which preserves the high explosion breakdown point properties of the MCD estimator, and has the additional advantages that it is well-conditioned and has a higher implosion breakdown point. A simulation study confirms the good properties of the estimator. Finally, we illustrate the advantages of the MSCD estimator for outlier detection in high dimensions and for portfolio selection on a large investment universe.
