1
$\begingroup$

I am interested in a model like:

$y_{i} = \sum_{k\in K}{\beta_{k} z_{k}}$, with $z_{k} \tilde{} N(\mu_{k}, \sigma_{k})$. where $\beta \equiv(\beta_{k})_{k\in K}$ is not known, but all else is.

I assume that there is a prior $\pi (\beta)$ of an arbitrary distribution and I want to calculate the posterior (of $\beta$), given an observation of $(y_{i})_{i\in N}$ (but not of $z_{k}$).

I wonder if there are conjugate prior distributions for this case (any other information might be useful as well). I understand that for unknown mean and variance normal occurrences you can use normal mean and inverse chi square variance priors, but here they must have (EDIT:) related distributions.

Although I'm interested in the general case, any information about the case $K = \{k\}$ would be helpful already.

$\endgroup$
1
$\begingroup$

The likelihood function is $p(y_i | \beta) = N(y_i; \sum_k \beta_k \mu_k, \sum_k \beta_k^2 \sigma_k)$, so a conjugate prior for $\beta$ would be proportional to this. Unfortunately, this is not a standard distribution unless $\mu=0$ and $K=\{k\}$, in which case it is a generalized Gamma distribution.

$\endgroup$

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.