2
$\begingroup$

Let $X$, $Y,$ and $Z$ be random variables. (There are no restrictions on these variables, but you may assume that these are continuous random variables if you want.) Suppose that $X$ and $Z$ are independent, and also suppose that $Y$ and $Z$ are independent. Does it follow that $\mathrm{cov}(XY,Z)=0$? (I understand that $XY$ and $Z$ may not be independent, but this does not rule out the zero covariance.)

Under the assumption that $X$ and $Z$ are independent and that $Y$ and $Z$ are independent, I am able to show that $$\mathrm{cov}(X,YZ) = \mathrm{cov}(Y,XZ) = \mathrm{cov}(XY,Z) + \mathrm{E}[Z]\cdot \mathrm{cov}(X,Y).$$ Showing this is fairly straightforward: $\mathrm{cov}(XY,Z)=\mathrm{E}[XYZ]-\mathrm{E}[XY]\cdot\mathrm{E}[Z]$; in addition, $\mathrm{cov}(X,YZ)=\mathrm{E}[XYZ]-\mathrm{E}[X]\cdot\mathrm{E}[YZ]=\mathrm{E}[XYZ]-\mathrm{E}[X]\cdot\mathrm{E}[Y]\cdot\mathrm{E}[Z]$, implying that $\mathrm{cov}(X,YZ)=\mathrm{cov}(XY,Z)+\mathrm{E}[XY]\cdot\mathrm{E}[Z]-\mathrm{E}[X]\cdot\mathrm{E}[Y]\cdot\mathrm{E}[Z]$, which is obviously equal to $\mathrm{cov}(XY,Z) + \mathrm{E}[Z]\cdot\mathrm{cov}(X,Y).$ And, of course, $\mathrm{cov}(X,YZ) = \mathrm{E}[XYZ]-\mathrm{E}[X]\cdot\mathrm{E}[Y]\cdot\mathrm{E}[Z]= \mathrm{cov}(Y,XZ)$, proving the above result.

However, to proceed further, my intuition tells me that $\mathrm{cov}(XY,Z) = 0$ and thus that $\mathrm{cov}(X,YZ) = \mathrm{cov}(Y,XZ) = \mathrm{E}[Z]\cdot\mathrm{cov}(X,Y)$. Am I wrong in thinking that $\mathrm{cov}(XY,Z) = 0$? But if it is true that $\mathrm{cov}(XY,Z) = 0$, is there a simple proof that does not possibly involve measure theory? Thanks.

$\endgroup$
  • 1
    $\begingroup$ The revised suggestion is wrong as well, try $(X,Y)$ uniform on $\{-1,1\}^2$ and $Z=XY$. Then any two random variables from $X$, $Y$ and $Z$ are independent and $\mathrm{cov}(XY,Z)=1$. $\endgroup$ – Did Feb 7 '16 at 14:54
  • $\begingroup$ @Did:I don't see why this is the case. $P(X\in [-1/2,1/2]|Z>0.5)=0\neq P(X\in [-1/2,1/2])$. So $X$ and $Z$ are not independent in this case. $\endgroup$ – adrido Mar 5 '18 at 0:29
  • 1
    $\begingroup$ @adrido Did you get that $X$, $Y$ and $Z$ are all two-valued random variables, always equal to $+1$ or $-1$? Then yes, $X$ and $Z$ are independent. (By the way, this counterexample is, under different but equivalent presentations, ultra classical.) $\endgroup$ – Did Mar 5 '18 at 6:31
  • $\begingroup$ ops! Sorry I didn't. nice example! $\endgroup$ – adrido Mar 16 '18 at 0:56
3
$\begingroup$

Edit: The original question asked about independence, and was answered by the example below. This example also settles the modified question about covariance, since $\text{Cov}(XY,Z)\ne 0$.

Toss a fair coin twice. Let $X=1$ if we have head on the first toss, and $0$ otherwise. Let $Y=1$ if we have head on the second, and $0$ otherwise. Let $Z=1$ if the two tosses give different results, and $0$ otherwise. Then the random variables $X$, $Y$, and $Z$ are pairwise independent.

The random variables $XY$ and $Z$ are not independent.

$\endgroup$
  • $\begingroup$ I have edited my question to ask about the zero covariance instead of independence between XY and Z. If X and Z are independent and Y and Z are independent random variables, is it true that cov(XY, Z) = 0? If true, is there a simple proof of this? Thanks. $\endgroup$ – g123 Feb 7 '16 at 9:26
  • $\begingroup$ Sorry this is off topic but can you tell me what will be the formula for $Cov(X,Y|Z)$? $\endgroup$ – Ronald Feb 3 '18 at 9:57

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.