2
$\begingroup$

I need to find two dependent random variables with standard normal distribution, but with zero covariance. It is easy too find just two dependent random variables with such a distribution (X and -X, for example), but how I can reach zero covariance?

Thanks in advance.

$\endgroup$
  • 1
    $\begingroup$ Answer by Google. $\endgroup$ – David Mitra Mar 2 '14 at 11:15
  • $\begingroup$ This article has also some example distributions which are dependent but have zero correlation. $\endgroup$ – hauntergeist Mar 2 '14 at 11:21
6
$\begingroup$

Try $X \sim N(0,1)$ and $Y=X$ when $|X|\lt k$ and $Y=-X$ when $|X|\ge k$ for some non-negative $k$.

Then $X$ and $Y$ have standard normal distributions and $cov(X,Y)$ is a continuous increasing function of $k$, negative when $k$ is close to $0$ and positive when $k$ is large. So for some $k$ you will have $cov(X,Y)=0$

| cite | improve this answer | |
$\endgroup$
  • 1
    $\begingroup$ Some experimentation suggests $k$ should be slightly more than $1.53817$ $\endgroup$ – Henry Mar 2 '14 at 11:49
  • 1
    $\begingroup$ and to be more precise, $k$ is the square root of the median of a chi-squared distribution with $3$ degrees of freedom, which using R code sqrt(qchisq(0.5,3)) gives about $1.538172$ $\endgroup$ – Henry Apr 30 '18 at 23:42
1
$\begingroup$

Let $X$ and $Z$ be independent standard normal random variables, and take $Y=Z$ on $\{X\cdot Z\ge 0\}$ and $Y=-Z$ on $\{X\cdot Z <0\}$.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ It might help if you said that you were comparing the normal random variables $Y$ and $Z$, which have zero covariance despite $|Y|=|Z|$ with probability $1$. $\endgroup$ – Henry Feb 20 '16 at 18:34
-1
$\begingroup$

Although I haven't worked through all aspects that must be checked, I believe that if you specify

$$[X \mid Y] = Y$$

i.e. that $X$ becomes $Y$ when conditioning on $Y$, you obtain

$$\text {Cov}(X,Y) = E(XY) = E[E(XY\mid Y)] = E[YE(X\mid Y)]= E[YE(Y)] =0$$

Note that $$E(X) = E[E(X\mid Y)] = E[Y] = 0$$

and by the decomposition-of-variance formula

$$\text {Var}(X) = \text {Var}(E[X\mid Y]) + E[\text {Var}(X\mid Y)]= \text {Var}(E[Y])+E[\text {Var}(Y)] = 0 + 1 = 1$$

i.e. this conditional specification is consistent with the unconditional moments.

So these variables are uncorrelated but not independent.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ What is the meaning of $[X \mid Y = Y]$? $\endgroup$ – Did Mar 22 '14 at 18:05
  • $\begingroup$ @Did The verbal description compensates for the confusing notation, which I just attempted to improve (and which still remains a shorthand, but a widespread one) $\endgroup$ – Alecos Papadopoulos Mar 23 '14 at 0:07
  • $\begingroup$ Sorry but the notation $[X|Y=Y]$ is NOT widespread, and for good reasons. For your interest, the only reasonable interpretation of the thing I can think of does not lead to covariance zero either. $\endgroup$ – Did Mar 23 '14 at 6:38

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.