I'm currently working on a project around the Bayesian approach to portfolio selection, and I can't manage to wrap my mind around the specification of the non-informative (diffuse) prior. Assuming gaussianity of the returns with parameters $\mu \in R^N$ and $\Sigma \in R^{N x N}$ the prior is usually expressed as follows:

$p(\mu,\Sigma) \propto |\Sigma|^{-(N+1)/2}$

Where $|\Sigma|$ is $det(\Sigma)$ and $N$ the number of variables (assets in my case). I understand that this comes from the marginal priors of $\mu$ and $\Sigma$, given by

$p(\mu) \propto c$

$p(\Sigma) \propto |\Sigma|^{-(N+1)/2}$

I understand the specification of $p(\mu)$, so that's not a problem. However, I do not understand the specification of $p(\Sigma)$. I understand that in the special case $N=1$, we obtain the usual prior for scale parameters $p(\sigma) \propto 1/\sigma$, but I don't get how we obtain $|\Sigma|^{-(N+1)/2}$.

I've checked already well-known references: Zellner, 1971 / Bawa, Brown and Klein 1974 / Geisser 1965 / etc but it doesn't help. It seems that there is something that I fundamentally don't grasp.

If anyone has an intuitive and clear explanation for this, it would be great !!

Many thanks in advance.


Your Answer

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

Browse other questions tagged or ask your own question.