The PDF of a Beta distribution is $$ f_X(x) = {{x^{a-1}(1-x)^{b-1}}\over {B(a,b)}} $$ and CDF $$ F_X(x) = I_x(a,b) $$

The PDF of a uniform distribution is $$ f_Y(y) = {1 \over {b-a}} \space for \space x \in [a,b] $$ and CDF $$ { x - a} \over {b - a} $$

I've know that if X and Y are two independent, continuous random variables, described by probability density functions above of $f_X(x)$ and $f_Y(x)$ then probability density function of $Z = XY$ is $$ f_Z(z) = \int_\infty^\infty f_X(x)f_Y({z\over x}){1\over abs(x)}dx $$ but I'm struggling with the algebra to get from this to the PDF and the CDF of the joint distribution. I'm using the uniform distribution between 0 and 1.

  • $\begingroup$ Is the mod(x) in your equation supposed to be abs(x)? $\endgroup$ – R Carnell Feb 4 at 13:43
  • $\begingroup$ yes - sorry I've amended $\endgroup$ – andrew Feb 4 at 14:58
  • $\begingroup$ Although your question is unclear--you never specify what the distributions of $X$ and $Y$ are--the title suggests they are both Beta distributions, indicating the duplicate thread provides an answer. $\endgroup$ – whuber Feb 4 at 15:31
  • $\begingroup$ sorry - amended to make clear that i'm looking for the joint PDF and CDF of a uniform distribution and a beta distribution $\endgroup$ – andrew Feb 4 at 15:48
  • 1
    $\begingroup$ Because the duplicate at stats.stackexchange.com/questions/3359/… has no explicit answers and because the uniform distribution (which is a Beta(1,1) distribution) is special, conceivably a simple, effective answer could be provided in this special case, so I have voted to reopen this question. $\endgroup$ – whuber Feb 5 at 15:59

The density of a $B(a,b)$ variable $X$ is $$f(x,a,b) = \frac{1}{B(a,b)}x^{a-1}(1-x)^{b-1}$$

for $0\le x\le 1.$ Let $F(x,a,b)$ be its distribution function

$$F(x,a,b) = \int_{-\infty}^x f(x,a,b)\,\mathrm{d}x = \int_0^x f(x,a,b)\,\mathrm{d}x,$$

so that (as usual) $\frac{d}{dx} F(x,a,b) = f(x,a,b).$

When $U$ is a uniform variable on $[0,1]$ and $0\le z\le 1,$ $\Pr(U\le z) = z.$ Otherwise, when $z\gt 1,$ $\Pr(U\le z)=1.$ These (defining) formulas can be expressed as

$$\Pr(U\le z) = z\wedge 1$$

(the smaller of $z$ and $1$), provided $z\ge 0.$

To evaluate the distribution of $UX$ (where $U$ and $X$ are independent), note it's certain that $0\le UX \le 1.$ So let $z$ be in this range and, assuming $a\gt 1,$ compute

$$\begin{aligned} F_{UX}(z) &= \Pr(UX\le z) = \Pr\left(U\le \frac{z}{X}\right) = E\left[\frac{z}{X}\wedge 1\right]\\ &= \frac{1}{B(a,b)}\int_z^1 \frac{z}{x} x^{a-1}(1-x)^{b-1}\,\mathrm{d}x + \int_0^z (1)f(x,a,b)\,\mathrm{d}x\\ &= \frac{z B(a-1,b)}{B(a,b)} (1-F(z,a-1,b)) + F(z,a,b)\\ &= \frac{z (a+b-1)}{a-1} (1-F(z,a-1,b)) + F(z,a,b). \end{aligned}$$

That determines the CDF of $UX$ (by definition). To obtain its PDF, differentiate with respect to $z,$ giving

$$f_{UX}(z) = \frac{d}{dz}F_{UX}(z) = \frac{a+b-1}{a-1}\left(1 - F(z,a-1,b) - z f(z,a-1,b)\right) + f(z,a,b).$$

Here are the results of four separate simulations of one million values of $UX$ for various values of $(a,b):$


The red curves plot $f_{UX}:$ they agree perfectly with the simulations.

Since this question appears to be for educational purposes, I leave the evaluation of the case $0\lt a \le 1,$ should that be of interest, as an exercise. Note that even in this case $UX$ exists, and there is no problem evaluating integrals from $z$ to $1$ even when the exponent of $x$ (equal to $a-2$ in the formula) is less than $-1;$ however, such integrals do not correspond to Beta distributions and need to be expressed in terms of the Incomplete Beta function.


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.