2
$\begingroup$

If $X$ and $Y$ are two independent random variables with means and variances $\mu_X,\mu_Y,\sigma^2_X,\sigma_Y^2$ respectively. If $\mu_X+\mu_Y>0$, by Cantelli's inequality then $P(X+Y\leq0)\leq\dfrac{\sigma_X^2+\sigma_Y^2}{\sigma_X^2+\sigma_Y^2+(\mu_X+\mu_Y)^2}$. Is this bound sharp? If yes, an example?

Thanks

$\endgroup$
  • $\begingroup$ Are you assuming $X$ and $Y$ are independent? $\endgroup$ – Robert Israel May 19 '17 at 1:10
2
$\begingroup$

Let $Z = X + Y$. $Z$ has mean $\mu=\mu_1 + \mu_2$ and, assuming $X$ and $Y$ are independent, variance $\sigma^2 = \sigma_1^2 + \sigma_2^2$. The inequality is then Cantelli's inequality for $Z$. Now Cantelli's is equivalent to

$$I_{z > 0} \ge 1 - (z - b)^2/b^2 $$ with $$ b = \mu + \sigma^2/\mu$$ Thus the inequality is sharp if and only if almost surely $Z$ is one of the two values of $z$ that make this an equality, namely $0$ and $b$. But the only way the sum of two independent random variables can have only two possible values is that one of $X$ and $Y$ is almost surely constant.

$\endgroup$
  • $\begingroup$ Thanks Robert. I was wondering if one can get an optimal bound (assuming independency between $X$ and $Y$) with the limited information of the first two moments of these random variables, of course without one being almost surely constant. $\endgroup$ – Quema May 19 '17 at 12:18
  • $\begingroup$ If one is not quite constant, but nearly so, you can get arbitrarily close to this bound. $\endgroup$ – Robert Israel May 19 '17 at 17:06
  • $\begingroup$ I am not so sure this bound is optimal (given the information we have). $\endgroup$ – Quema May 20 '17 at 21:16

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.