I already asked this question on math.stackexchange but got no answer.

For a non-integer number of degrees of freedom $\nu > p-1$, one can simulate the central Wishart distribution $W_p(\nu, \Sigma)$ with the help of the Bartlett decomposition.

How to simulate the noncentral Wishart distribution $W_p(\nu, \Sigma, \Theta)$ for such a $\nu$ (and $\Theta$ a $p\times p$ semi-definite positive matrix)? According to Letac & Massam's tutorial on non central Wishart distributions, this distribution exists for $\Sigma=I_p$, in the sense that there is a distribution whose characteristic function is the same as the one of $W_{p}(``\text{an integer} \geq p``, I_p, \Theta)$ with $\nu$ instead of the integer. For $\Sigma \neq I_p$ I don't know actually.

For $\nu>2p-1$ it suffices to apply the equality $W_p(\nu, \Sigma, \Theta) = W_{p}(\nu-p, \Sigma) \ast W_{p}(p, \Sigma, \Theta)$. How to do in the case when $p-1 < \nu \leq 2p-1$?

  • $\begingroup$ Namely: math.stackexchange.com/q/2542561 (I believe that in principle the policy, though, is to give your math.SE question a week before asking it here too.) $\endgroup$ – Jules Lamers Nov 30 '17 at 12:28

Efficient Simulation of the Wishart model (2009) shows how to simulate the fractional non-central Wishart distribution for all $\nu>p-1$. See section 6.1.2.b for $\nu\geq p+1$ and page 41 and following for $p-1<\nu<p+1$.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.