PROCEEDINGS OF THE
AMERICAN MATHEMATICAL SOCIETY
Volume 123, Number 8, August 1995
ON DISGUISED INVERTED WISHART DISTRIBUTION
A. K. GUPTA AND S. OFORI-NYARKO
(Communicated by Wei-Yin Loh)
Abstract. Let A ~ Wp(n, I) and A = ZZ' where Z is a lower triangular
matrix with positive diagonal elements. Further, let B = A~l = W'W have
inverted Wishart distribution so that W = Z-1. In this paper we derive the
distribution of M = WLW1. It is also shown that a=£±±T'MT ~ Fp,„_p+1
where T ~ A/p(0, Ip) is independent of M .
1. Introduction
While deriving the minimax estimator of a normal covariance matrix when
additional information is available on some coordinates [2], we were confronted
with finding the distribution of a random variable of the type (Z'Z)~X where
Z is a lower triangular matrix with positive diagonal elements such that ZZ' =
A(p x p) ~ Wp(n,L). Here we first derive the distribution of M - WLW,
where W = Z~x. The distribution of T'MT is also derived where indepen-
dently T ~ Np(0, Ip). We call M a disguished inverted Wishart variable for
reasons explained in §3. Tan and Guttman [3] derived the distribution of a
disguised Wishart variable.
In §2 we present some preliminary results which are used in the sequel. In
§3 main results of the paper are derived.
2. Some preliminary results
The following lemmas are needed to derive the distribution of M. The
proofs of Lemmas 1.1 and 1.2 are given in [1].
Lemma 2.1. If M = MXM\, where Mx is a p x p lower triangular matrix with
positive diagonal elements, then
(2.1) J(M^Mx) = 2pf[mp-;)+[
i=i
where znil(i) is the ith diagonal element of Mx. Also the transformation N =
Received by the editors July 7, 1993.
1991 Mathematics Subject Classification. Primary 62H10; Secondary 62H12.
Key words and phrases. Minimax estimation, risk, Jacobian, lower triangular matrix, F-distribu-
tion.
© 1995 American Mathematical Society
2557
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use