Let $X$ be a random variable and let $g,f$ be real-valued, nondecreasing, and bounded.

Show that $\mathbb{E}[f(X)g(X)]\geq \mathbb{E}[f(X)]\mathbb{E}[g(X)]$

Having a hard time seeing where to start for some reasons, any hints?

I managed to write:

$\mathbb{E}[f(X)g(X)] = \frac{1}{4}\mathbb{E}[(f(X)+g(X))^2 - (f(X)-g(X))^2]$

But I'm not sure if this is in the right direction or not.


Hint: Let $X_1,X_2$ be independent copies of $X$, and note that $$(g(X_1)-g(X_2))(f(X_1)-f(X_2))\geq 0.$$

  • $\begingroup$ Didn't know this trick, very nice indeed! $\endgroup$ – zhoraster Sep 22 '15 at 14:13
  • $\begingroup$ This is really beautiful. $\endgroup$ – drhab Sep 22 '15 at 14:29

The fact itself is pretty intuitive: Two increasing transformations of a random variable are positively correlated.

First note that $X$ maybe assumed to have a uniform distribution on $[0,1]$ (via the quantile transformation).

Now let $f(x) = \mathbf{1}_{[x_0,1]}(x)$, $x_0\in[0,1)$. Since $$ h(x) = \frac{x-x_0}{1-x_0}\le x,\quad x\in[x_0,1], $$ we get $$ E[f(X) g(X)] = E[\mathbf{1}_{X\ge x_0} g(X)]\ge E[\mathbf{1}_{X\ge x_0} g(h(X))] \\ = \int_{x_0}^1 g(h(x))dx = (1-x_0)\int_0^1 g(z) dz = P(X\ge x_0)E[g(X)] = E[f(X)] E[g(X)]. $$

By linearity, the desired inequality holds for non-negative non-decreasing step functions $f$. But for a constant $f$ we have an equality, so the inequality holds for non-decreasing step functions $f$ of any sign. The proof is completed by approximating any function with step functions.


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.