This is a re-post of a question on the Mathematica stack exchange, as per the advice of another user (see here). I am pursuing a computational solution there, but thought it might be worth looking for a pen-and-paper method.

I would like to calculate the PDF for the product of three independent Beta random variables. Specifically, I would like to find the distribution of the product of the following: $X_1\sim \textrm{Beta}(1,3/2)$, $X_2\sim \textrm{Beta}(3/2,1)$ and $X_3\sim \textrm{Beta}(2,1/2)$.

Does anyone have any idea how I can do this? The reason I state $n$ in the question is because I would like eventually to generalise this calculation to the product of more $\textrm{Beta}$ random variables.

  • 1
    $\begingroup$ Isn't this just the "product" where the value within the unit cube at x,y,z is just X1(x)*X2(y)*X3(z) where those "X" functions are densities? Or in R notation: dbeta(x, 1,1.5)*dbeta(y,1.5,1)*dbeta(z,2,0.5) $\endgroup$ – DWin Aug 12 '15 at 1:56
  • $\begingroup$ AKA Dirichlet distribution? $\endgroup$ – DWin Aug 12 '15 at 2:04
  • $\begingroup$ They are independent, but not identical, I am wondering if they still can be multiplied together. $\endgroup$ – Deep North Aug 12 '15 at 7:00
  • $\begingroup$ And pdf is not probability, the dbeta(x,1,1.5) is not probability, since Beta distribution is continuous, isn't it? $\endgroup$ – Deep North Aug 12 '15 at 7:08
  • $\begingroup$ @DeepNorth: Agree densities are not probabilities, but wouldn't the densities of a multivariate distribution where these were uncorrelated be the product of the ednsities times some normalization constant? $\endgroup$ – DWin Aug 14 '15 at 1:58

I read this as about the distribution of the product of independent beta distributed random variables. A good tool to study the (distribution of) product of independent random variables is the Mellin transform. The Mellin transform of a beta random variable can be calculated as (an elementary integral) $$ \DeclareMathOperator{\E}{\mathbb{E}} M_X(s) = \E X^s = \frac{B(\alpha+s,\beta)}{B(\alpha,\beta)} $$ where $B( , )$ is the Beta function. The Mellin transform of the (independent) product $X_1 \dotsm X_n$ is the the product of the Mellin transforms $M_{X_i}(s)$.

If that cannot be inverted exactly, the saddlepoint approximation for instance can be used to find a very good approximation for the density. (I will post here some examples, later). This paper finds some general expressions using the Mellin transform approach above, but general solutions do not have a simple form. So the approximation route above could be more practical.


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.