On the regular variation of elliptical random vectors (Q2497802)

The author considers the following mathematical context: let \(S = (S_1, S_2,\dots, S_d)^{\text{T}}\), \(d\geq 2\), be a spherical random vector in \(\mathbb R^d\), and let \(X = A^{\text{T}} S\) be an elliptical vector with \(A\in \mathbb R^{d\times d}\) a non-singular matrix (T is the transposition sign). Berman (1992) proved that, if the random radius \(R_d =(\sum_{i=1}^d S_i^2)^{1/2}\) is regularly varying with index \(\alpha>0\), then \(S\) and (any of) its components \(S_i, 1\leq i\leq d\), are regularly varying with index \(\alpha\). The aim of the present paper is to describe new equivalent conditions for the regular variation of the elliptical random vector \(X\). More precisely, the author proves that the regular variation of an (arbitrary) component of \(X\) entails a certain asymptotic result on \(S\) and its components, and obtains new equivalent conditions for this fact to hold. As a consequence, two improved asymptotic results concerning the sojourn and supremum of Berman processes are derived. The coefficient of upper tail dependence for elliptical random vectors with regularly varying components is briefly discussed.