Normalized sample covariance matrix for multivariate t-distributed data (n = 1000, d = 500) with three degrees of freedom (left) and the corresponding true dispersion matrix (right). 

Contexts in source publication

Context 1

... the data stem from a leptokurtic or even regularly varying elliptically distributed random vector both the finite sample and asymptotic (co-)variances of the sample covariance or cross cor- relation matrix can be very large (see, e.g., Lindskog et al., 2003, Oja, 2003, van Praag and Wesselman, 1989. For example, Figure 6 contains a realization of the (normal- ized) sample covariance matrix for multivariate t-distributed data (n = 1000, d = 500) with three degrees of freedom (left hand side) and the corresponding true dispersion matrix (right hand side). ...

Context 2

... left hand side of Figure 8 contains a realization of the spectral estimator for the multivariate t-distributed data already used for calculating the sample covariance ma- trix in Figure 6. This can be compared with the corresponding true dispersion matrix on the right hand side of Figure 8 and the sample covariance matrix in Figure 6. ...

Context 3

... left hand side of Figure 8 contains a realization of the spectral estimator for the multivariate t-distributed data already used for calculating the sample covariance ma- trix in Figure 6. This can be compared with the corresponding true dispersion matrix on the right hand side of Figure 8 and the sample covariance matrix in Figure 6. Ob- viously, the spectral estimator provides a robust alternative to the sample covariance matrix. ...