8
$\begingroup$

If we assume probability density distribution functions of random variables $\alpha$, $\beta$ and $\alpha/ \beta$, we would like to devise a joint distribution of $\alpha$ and $\beta$. Although there does not exist a unique solution, we would like to just find one. Any thoughts ?

$\endgroup$
3
  • $\begingroup$ If we assume that they are positive, then we can take their logarithms and reduce the problem to the case if we know $\alpha, \beta$ and $\alpha+\beta$, which sounds maybe simpler. $\endgroup$
    – domotorp
    Jan 24, 2010 at 19:39
  • 1
    $\begingroup$ I have two guesses: First, the case where alpha can be 0 is quite different from if you exclude that possibility. Second, that you can encode some hard problems this way. $\endgroup$ Jan 24, 2010 at 21:06
  • $\begingroup$ Btw, not exactly the same thing, but related: en.wikipedia.org/wiki/Sicherman_dice $\endgroup$
    – domotorp
    Jul 18, 2010 at 11:42

1 Answer 1

1
$\begingroup$

I am not entirely certain what your question is. It might be

(i) Is it always possible to find a joint distribution of $(\alpha, \beta)$ for any prescribed distributions of $\alpha, \beta$ and $\alpha / \beta$ ?

(ii) Is it possible to find/calculate a joint distribution from the three distributions when you know the joint distribution exists e.g. because these are observations in an experiment?

(i) is not possible in general. Set $\alpha = \exp(X)$ and $\beta = \exp(-Y)$ then $\log(\alpha / \beta) = X + Y$. Now let $X$ and $Y$ be uniform on $[0,1]$ and choose a distribution for $X+Y$ so that $P(X+Y < 0.5) = 1$. This means $P(X > 0.5) = 0$ a contradiction to uniform. A way to visualize this might be looking at mass distributions on the square $[0,1]\times[0,1]$. Prescribing the margins (here uniform) is a restriction on the projections to the axes (i.e. $0\times[0,1]$ and $[0,1]\times 0$) and the remaining freedom is distributing the mass in the square.

(ii) Looks more like statistics than probability. There are a number of ways of coming up with a joint distribution. But you would need to specify more context to find a reasonable approach.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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