5
$\begingroup$

Let $X$ and $Y$ be independent random variables, $X,Y \sim unif(0,1)$. Let $U = \min \{X,Y\}$ and $V = \max\{X,Y\}$. Find the correlation coefficient of $U$ and $V$.

I think we can assume that $U = X$ and $V = Y$ because they both have the same distribution. Further, $E[U] = E[V] = 1/2$ and $Var(U) = Var(V) = 1/12$. Next is $E[UV]$ and this is where I am stuck, because I can't solve this $E[UV] = \int_{}^{}\int_{\mathbb{R}^2}^{}uvf(u,v)dudv$. I think I can substitute $f(u,v)$ with $1$, but I'm not sure.

$\endgroup$
5
  • 1
    $\begingroup$ $U$ and $X$ do not have the same distribution. $P(U\geq s) = P(X\geq s)P(Y\geq s) = (1-s)^2$. Also you have not properly defined $V$. $\endgroup$
    – Alex R.
    Feb 10, 2015 at 23:00
  • $\begingroup$ Then could you please tell me how to find E(U)? $\endgroup$
    – user44697
    Feb 10, 2015 at 23:22
  • $\begingroup$ Well $F(s)=P(U\leq s)$, the cdf, you could differentiate that and get a density. Or maybe you remember that for a non-negative random variable $U$, $E[U]=\int_0^\infty (1-F(s))ds$ $\endgroup$
    – Alex R.
    Feb 10, 2015 at 23:24
  • $\begingroup$ $U$ and $V$ are not independent, as $U<V$ ! $\endgroup$
    – user65203
    Feb 13, 2015 at 15:03
  • $\begingroup$ math.stackexchange.com/q/2146136/321264 $\endgroup$ May 2, 2021 at 14:35

3 Answers 3

4
$\begingroup$

This problem has several nice shortcuts.

For the first term, you can note that $UV=XY$. This is because one or the other of $X,Y$ is the smaller one and the other is the larger, and multiplication doesn't care what order they come in. So since $ X, Y $ are independent, $ E [UV]= E [X] E [Y] = 1/4$.

We can use a similar trick to calculate $E[U]$ and $E[V]$. Let's think about $U$. We first consider $X=x$. Now with probability $1-x$, $Y>x$, in which case $U=x$. With probability $x$, $Y \leq x$, in which case $U$ is uniform on $[0,x]$. I'll write $U_x$ for a uniform variable on $[0,x]$. Now linearity of expectation tells us that:

$$E[U|X=x]=x (1-x) + x E[U_x]=x (1-x) + x^2/2$$

Now the total expectation formula gives:

$$E[U]=\int_0^1 E[U|X=x] f_X(x) dx = \int_0^1 (x (1-x) + x^2/2) dx.$$

The situation is similar to calculate $E[V]$, $E[U^2]$, and $E[V]^2$, then you can just combine everything to get the correlation coefficient.

$\endgroup$
0
$\begingroup$

You can also check whether your answer is correct using a computer algebra system. In this instance, since $X$ and $Y$ are independent standard Uniform random variables, their joint pdf, say $f(x,y)$ is:

f = 1;   domain[f] = {{x, 0, 1}, {y, 0, 1}};

Then, you seek:

Corr[{Min[x, y], Max[x, y]}, f]

$\frac12$

where I am using here the Corr function from the mathStatica package for Mathematica. As disclosure, I should perhaps add that I am one of the authors.

The expectation you are having trouble with is $E[U V]$:

Expect[Min[x, y] Max[x, y], f]

$\frac14$

Also, your calculation $E[U] = E[V] = 1/2$ is plainly wrong: this should be obvious since $E[X] = \frac12$, so $E[min[X,Y]]$ must be less than $\frac12$.

$\endgroup$
0
$\begingroup$

Let's find the joint distribution. When $u \leq v$ then \begin{align*} P(U<u,V<v)&=P(\min\{X,Y\}<u,\max\{X,Y\}<v)\\ &=P(X<u,Y<u)+P(X<u,u<Y<v)+P(u<X<v,Y<u)\\ &=P(X<u)P(Y<u)+P(X<u)P(u<Y<v)+P(u<X<v)P(Y<u)\\ &=u^2+2u(v-u)=2uv-u^2 \end{align*} When $v<u$ then \begin{align*} P(U<u,V<v)&=P(\min\{X,Y\}<u,\max\{X,Y\}<v)=v^2 \end{align*} Hence $f(u,v)=2$ when $u \leq v$ and $f(u,v)=0$ when $v<u$. Then $f(v)=2v$, $f(u)=2(1-u)$. Then $E[U]=1/3$, $E[U^2]=1/6$, $E[V]=2/3$, $E[V^2]=1/2$ and $$E[UV]=\int_0^1\int_0^{1}uv f(u,v)dudv =2\int_0^1\int_0^{v}uv dudv =\int_0^1v^3 dv=\frac{1}{4}$$

Now using this, I get the correlation is $1/2$.

btw. More generally if $X$ and $Y$ are independent and have distributions $F_x$ and $F_y$. Then when $u \leq v$, \begin{align*} P(U<u,V<v)&=P(\min\{X,Y\}<u,\max\{X,Y\}<v)\\ &=F_x(u)F_y(u)+F_x(u)(F_y(v)-F_y(u)) +F_y(u)(F_x(v)-F_x(u))\\ &=F_x(v)F_y(u)+F_x(u)F_y(v)-F_x(u)F_y(u) \end{align*} and when $v<u$, $$ P(U<u,V<v)=P(\min\{X,Y\}<u,\max\{X,Y\}<v)=F_x(v)F_y(v) $$ so $f(v,u)=f_x(v)f_y(u)+f_x(u)f_y(v)$ when $u \leq v$ and $f(v,u)=0$ when $v<u$. Using this you can calculate $E[g(U,V)]$ for any function $g$.

$\endgroup$
8
  • $\begingroup$ The PDF of (U,V) is wrong and not useful. The value of E(UV) is wrong. $\endgroup$
    – Did
    Feb 13, 2015 at 14:37
  • 1
    $\begingroup$ That value of $E[UV]$ doesn't even make sense, $U \in [0,1]$ and $V \in [0,1]$ so $E[UV] \in [0,1]$... $\endgroup$
    – Ian
    Feb 13, 2015 at 16:23
  • $\begingroup$ @Did PDF is corrected. The PDF is useful to calculate $E[UV]$. $\endgroup$
    – user103828
    Feb 14, 2015 at 10:01
  • $\begingroup$ @Ian You are correct. I made a mistake in the PDF calculation. It's been corrected now. $\endgroup$
    – user103828
    Feb 14, 2015 at 10:03
  • $\begingroup$ @user103828 In general yes, of course, but not in this case. Did you read Ian's answer? $\endgroup$
    – Did
    Feb 14, 2015 at 20:06

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.