Multivariate elliptically contoured and ө-generalized normal models
The author introduces the conception of the elliptically contoured distribution and the matrix variate \(\theta\)-generalized normal distribution. A \(p\times n\) random matrix \(X\) has an elliptically contoured matrix variate distribution (e.c.m.v.d.) if its characteristic function has the form \[ \Psi_{X}(T)=\exp\{ i{\text Tr}(T'M)\} \varphi({\text Tr}(T'\Sigma T\Phi)), \] where \(\Sigma>0\) and \(\Phi >0.\) The Wishart distribution is a particular case of e.c.m.v.d. A \(p\times n\) random matrix \(Y\) has a matrix variate \(\theta\)-generalized normal distribution (m.v.g.n.d.) if it can be written as \(Y=AXB+M\), where \(A\colon p\times p\), \(B\colon n\times n\), \(M\colon p\times n\) are constant matrices with \(A\) and \(B\) nonsingular and the elements of the \(p\times n\) random matrix \(X\) are independent with the probability density function \[ f(X_{ij})=\{2\Gamma(1+\theta^{-1})\}^{-1}\exp\{-| X_{i} |^{\theta}\},\quad \theta>0. \] The author gives properties of these distributions. For example, he proves that among the random matrices \(X\) such that \[ E\| AXB+M\|_{\theta}= k_0, \quad \| Y\|_{\theta}= \sum\limits_{i=1}^{p}\sum\limits_{j=1}^{n} | y_{ij}|^{\theta}, \] the matrix \(X\) has maximal entropy if \(X\) has m.v.g.n. distribution. (The entropy of \(X\) is \(E_{X}(-\log f(X)).\)
This page was built for publication: Multivariate elliptically contoured and ө-generalized normal models
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4258899)
