$X$ and $Y$ are random variables uniformly distributed on $[a,b]$. Could the random variable $Z=X+Y$ be uniformly distributed if $X$ and $Y$ are dependent and correlation between them doesn't equal $1$ or $-1$?

  • $\begingroup$ $1/12$ of the square of the range of $Z$ is the variance of $Z$ if $Z$ is uniformly distibuted. That means the range of $Z$ would be less than $2(b-a)$. The variance of $Z$ is also $(b-a)^2(1+\rho)$ where $\rho$ is the correlation. Shortening the range of $Z$ necessarily means chopping triangles off corners of the square $[a,b]^2$. $\endgroup$ – Michael Hardy Dec 27 '12 at 0:12

Yes. For instance, let $X$ be uniformly distributed on $[0,1]$, and let $Y$ be equal to $\left\{ X+\frac{1}{2}\right\}$, where $\{x\}$ represents the fractional part of $x$. Then $Z=X+\left\{ X+\frac{1}{2}\right\}$ is uniformly distributed on $[1/2,3/2]$. There is an infinite family of such examples, parametrized by the number of jump discontinuities of $Y(X)$, in which $Z$ is uniformly distributed over $[1-1/n,1+1/n]$ and the correlation between $X$ and $Y$ approaches $-1$ as $n\rightarrow\infty$.

  • $\begingroup$ @cardinal Oh, I see that I was typing stupid comments without thinking. Thanks. $\endgroup$ – Learner Dec 27 '12 at 1:45
  • $\begingroup$ @Learner: It happens (to all of us). :-) $\endgroup$ – cardinal Dec 27 '12 at 1:46

We produce a discrete example, because the details are more pleasant to verify. The same idea leads to a continuous example.

Let $X=-3,-1,1,3$ each with probability $1/4$. Suppose that when $X=-3,-1,1,3$ respectively, then $Y=1,3,-3,-1$ respectively. Then $X+Y$ takes on the values $-2$ and $2$, each with probability $1/2$.

Since $X$ and $Y$ have mean $0$, their covariance is $E(XY)$ is $-3$. The variance of $X$ and of $Y$ are each $5$. So the correlation coefficient of $X$ and $Y$ is $-\frac{3}{5}$.

Continuous examples can be made using a similar strategy.

  • $\begingroup$ This is the same answer as @mjqxxxx has previously supplied. $\endgroup$ – cardinal Dec 27 '12 at 3:11
  • $\begingroup$ @cardinal: I have removed the continuous example. $\endgroup$ – André Nicolas Dec 27 '12 at 3:16

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.