I have two random variables

$Y \sim \operatorname{Beta}(a, 1 - a)$

$Z \sim \operatorname{Exp}(1)$

If $Y$ and $Z$ are independent, why is the distribution of $X = YZ \sim \operatorname{Gamma}(a, 1)$?

$f_X(x) = \int_0^\infty|\frac{1}{y}|f_Y(y)f_Z(\frac xy)dy$

$f_X(x) = \int_0^\infty \frac{1}{y}\frac{1}{\Gamma(\alpha)\Gamma(1-\alpha)}y^{\alpha-1}(1-y)^{-\alpha}e^{-\frac{x}{y}}dy$

but I can't derive more than it.

How can I proof $YZ \sim \operatorname{Gamma}(a, 1)$ ?


Here is a very familiar approach; nothing special about it.

Joint pdf of $(Y,Z)$ is $$f_{Y,Z}(y,z)=\frac{e^{-z}y^{a-1}(1-y)^{-a}}{\Gamma(a)\Gamma(1-a)}\mathbf1_{0<y<1,z>0}\quad,\,0<a<1$$

You can use a change of variables $(Y,Z)\to (U,V)$ such that $U=YZ$ and $V=Z$.

So the preimages are $z=v$ and $y=u/v$, and $0<y<1,z>0\implies 0<u<v$.

Absolute value of jacobian of transformation is $1/v$.

This gives the joint pdf of $(U,V)$:


Therefore, marginal pdf of $U$ is $$f_U(u)=\frac{u^{a-1}}{\Gamma(a)\Gamma(1-a)}\int_u^\infty e^{-v}(v-u)^{-a}\,dv\,\mathbf1_{u>0}$$

Substitute $v-u=t$, which converts the integral to a Gamma function, ultimately giving the answer $$f_U(u)=\frac{1}{\Gamma(a)}e^{-u}u^{a-1}\mathbf1_{u>0}$$

  • $\begingroup$ Pretty interesting, I thought making joint PDF would be more complicated. Did it via Mellin though... $\endgroup$ – Severin Pappadeux May 21 at 17:10

Use the following result:

Assuming $Y$ and $Z$ are independent, the PDF of $X = YZ$ is given by:

$$f_X(x) = \int_{-\infty}^{\infty} \frac{1}{|u|} f_{Y}(u) f_Z\left(\frac{x}{u}\right) du$$


You could use Mellin transform to get distribution for $YZ$. For a product of two RVs, there is a very simple theorem, which states that Mellin transform of product distribution is the product of Mellin transform of the constituent RVs. So there is simple algorithm - Mellin transform of $Y$ multiplied by Mellin transform of $Z$, and then do inverse Mellin transform to get final PDF. It is just like using Fourier transform for sum of two RVs.

$$ M(YZ) = M(Y) M(Z) $$

For exponential distribution $Y = \exp(-x)$

$$ M(Y) = \Gamma(s) $$

For $Z = B(a, 1-a)$ one could easily get

$$ M(Z) = \frac{ \Gamma(s+a-1) }{\Gamma(s) \Gamma(a)} $$

Therefore, for product

$$ M(YZ) = \frac{ \Gamma(s+a-1) }{\Gamma(a)} $$

which gives us pretty obvious inverse transform in the form of

$$ PDF(x|YZ) = M^{-1}\{\frac{ \Gamma(s+a-1) }{\Gamma(a)} \} = \frac{ \exp(-x) x^{a-1} }{\Gamma(a)} 1_{x>0} $$


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.