2
$\begingroup$

The problem is very straightforward and it says:

Let $X_1,X_2,X_3$ independent random variables with a common density $f_{x_i} \sim Uniform(1,2)$. Define $Y = X_1 X_2 / X_3$ Find the numeric value of $var(Y)$.

I propose the following transformation in order to compute the density of $Y_1$ through the margin distribution, and then compute the variance of $Y_1$:

$$ Y_1 = X_1 X_2 / X_3 $$ $$ Y_2 = X_2/X_3 $$ $$ Y_3 = X_3 $$

But I'm not getting any results. Is there an easy path to follow?

Disclaimer: I tried to solve the problem with this technique because I wanted to practice the topic of transformation of random variables.

$\endgroup$
5
  • 2
    $\begingroup$ You have asked two different questions: the question in the title is to find the distribution of $Y$; the question in the text is to find $\text{Var}(Y)$. It is quite easy to find the latter without the former; though if you need the former, then you need the former. $\endgroup$
    – wolfies
    Commented Oct 25, 2021 at 5:44
  • 1
    $\begingroup$ As $X_1,X_2,X_3$ are independent, $E\left[Y\right]=E\left[\frac{X_1X_2}{X_3}\right]=E\left[X_1\right]E\left[X_2\right]E\left[\frac1{X_3}\right]$. Similarly find $E\left[Y^2\right]$ and hence the variance. Please add the self-study tag and read its wiki. $\endgroup$ Commented Oct 25, 2021 at 9:55
  • $\begingroup$ Inspirationally a la Ramanujan, looks like the variance should be approximately half of Euler's constant. $\endgroup$
    – wolfies
    Commented Oct 25, 2021 at 11:40
  • 2
    $\begingroup$ Taking logarithms of everything ought to be revealing ;-). However, your title question and post ask two different things: finding the variance is a matter of straightforward integration (because the uniform distribution has such a simple PDF). Which question are you trying to ask? $\endgroup$
    – whuber
    Commented Oct 25, 2021 at 13:41
  • $\begingroup$ Thanks for all your feedback. I already have edited my question in order to get more accurate answers. I tried to find the variance of Y by proposing a transformation of random variables since i wanted to practice that topic. $\endgroup$ Commented Oct 25, 2021 at 18:30

2 Answers 2

5
$\begingroup$

One easy path is to apply standard definitions. In particular, when the $X_i$ are independent and identically distributed with a density $f,$ the moment of $Y=X_1X_2/X_3 = y(X_1,X_2,X_3)$ of order $k$ is given by

$$\mu_k(Y) = \iiint y(x_1,x_2,x_3)^kf(x_1)f(x_2)f(x_3)\,\mathrm{d}x_1 \mathrm{d}x_2 \mathrm{d}x_3.$$

For the Uniform$[1,2]$ distribution, $f(x) = 1$ for $1\le x \le 2$ (and otherwise is zero). The domain of integration thereby reduces to the cube $[1,2]^3,$ giving

$$\mu_k(Y) = \int_1^2\int_1^2\int_1^2 \left(\frac{x_1x_2}{x_3}\right)^k\,\mathrm{d}x_1 \mathrm{d}x_2 \mathrm{d}x_3.$$

The integrand factors, enabling this triple integral to be computed as a product of three elementary univariate integrals,

$$\mu_k(Y) = \int_1^2x_1^k\,\mathrm{d}x_1 \int_1^2x_2^k\,\mathrm{d}x_2\int_1^2 x_3^{-k}\, \mathrm{d}x_3 = \left(\frac{x_1^{k+1}}{k+1}\bigg|^2_1\right) \left(\frac{x_2^{k+1}}{k+1}\bigg|^2_1\right) \left(\frac{x_3^{1-k}}{1-k}\bigg|^2_1\right)$$

(understanding that any expression evaluating to "$x^0/0$" means $\log(x)$).

Apply this result to the cases $k=1,2$ and plug them into the formula

$$\operatorname{Var}(Y) = \mu_2(Y) - \mu_1(Y)^2.$$

$\endgroup$
3
$\begingroup$

Graphical comment: A simulation (in R) may give you some clues.

set.seed(2021);  m = 10^5
x1 = runif(m, 1,2)
x2 = runif(m, 1,2)
x3 = runif(m, 1,2)
y1 = x1*x2/x3
summary(y1);  sd(y1)
   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
 0.5134  1.1573  1.4795  1.5611  1.8731  3.8708 
[1] 0.5396972  # aprx SD(X1), about 2 place accuracy
y2 = x2/x3
summary(y2)
   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
 0.5026  0.8214  1.0018  1.0407  1.2195  1.9966 

enter image description here

par(mfrow=c(2,2))  # enable 4 panel plot
 hist(y1, prob=T, col="skyblue2")
 hist(y2, prob=T, col="skyblue2")
 hist(x1*x2, prob=T, col="skyblue2")
 plot(y1,y2, pch=".")
  abline(v=1:2, col="red")
par(mfrow=c(1,1))  # return to 1 panel plot
$\endgroup$

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.