Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

This comes from the book "A Probability Path". I'm just working through the problems trying to get a grasp of conditional expectations.

Suppose $X,Y$ are random variables with finite second moments such that for some decreasing function $f$ we have $E(X|Y)=f(Y)$. Show that $\operatorname{Cov}(X,Y)\leq 0$.

I can't figure out how to relate $E[X|Y]$ to $E[XY]$ which was my approach. If I could get a hint for this or how the decreasing function affects $E[X],E[Y],E[XY]$. Is there some work with bounding involved?

share|cite|improve this question
$E[X\mid Y]$ is the minimum-mean-square-error (MMSE) estimate of $X$ given the value of $Y$ and is given to be a decreasing function. The linear MMSE estimator of $X$ given the value of $Y$ is a straight line (passing through the mean point $(\mu_X, \mu_Y)$) whose slope has the same sign as $\text{cov}(X,Y)$. So intuitively one expects that $\text{cov}(X,Y) < 0$ so that the straight line is roughly "parallel" to $f(y)$. I think @Henry's answer formalizes this idea as $\text{cov}(Y, f(Y)) \leq 0$. – Dilip Sarwate Dec 2 '11 at 13:16

You could try something like the Law of total expectation to get $$E[XY]=E[E[XY|Y]]=E[Y\,E[X|Y]]$$

Added: So we now want to show $\operatorname{Cov}(Y,f(Y))\leq 0$. Let $Z$ be independent of $Y$ but identically distributed $Y$. Then $(Y-Z)(f(Y)-f(Z)) \le 0$ since $f()$ is decreasing.

So $E[(Y-Z)(f(Y)-f(Z))]\le 0$, and $E[(Y-E[Y]-(Z-E[Z]))(f(Y)-E[f(Y)]-(f(Z)-E[f(Z)]))]\le 0$ but since $Y$ and $Z$ are iid, this implies $E[(Y-E[Y])(f(Y)-E[f(Y)])]\le 0$ which is $\operatorname{Cov}(Y,f(Y))\leq 0$

share|cite|improve this answer
How do you prove then that $\operatorname{Cov}(Y,f(Y))\leq 0$ for decreasing $f$? – Ilya Dec 2 '11 at 7:52

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.