Let $X$ and $Y$ be random variables with finite means $\mu_X$ and $\mu_Y,$ finite variances $\sigma_X^2$ and $\sigma_Y^2,$ and correlation $\rho.$ Let $A$ be the event that $X \leq \mu_X + k\sigma_X$ and $Y \leq \mu_Y + k\sigma_Y,$ where $k \gt 1.$ Using the one-sided univariate Chebyshev inequality, I can find this lower bound on the probability of $A$ occurring: $$P[A] \geq {{k^2-1} \over {k^2+1}}$$ But this does not use the correlation. If the correlation is known, what is the tightest bound possible?

  • 1
    $\begingroup$ Are you sure you don't want $A$ to be the event $|X-\mu_X|\le k\ \sigma_X$ and $|Y-\mu_Y|\le k\ \sigma_Y$? That would be closer to the univariate Chebyshev situation. $\endgroup$
    – whuber
    Feb 12, 2014 at 19:35
  • 1
    $\begingroup$ Formulas for that case are available from Wikipedia's article on Chebyshev's inequality. I couldn't find anything on the one-sided scenario. $\endgroup$
    – soakley
    Feb 12, 2014 at 20:10

1 Answer 1


Using the results from wikipedia page of Chebyshev's inequality (first inequality in section "two correlated variables"), I dont see anything for your case with one-sided inequalities. The more usual case with double inequalities, one obtains, with the event $A$ defined as $A=|\frac{X-\mu_X}{\sigma_X}|\le k ~~\text{and}~~ |\frac{Y-\mu_Y}{\sigma_Y}| \le k$ we get the inequality $$ Pr(A) \ge 1 - \frac{2k^2+\sqrt{4k^4-4k^4\rho}}{2k^4} = 1- \frac{1+\sqrt{1-\rho}}{k^2} $$ where $\rho$ is the correlation between $X$ and $Y$. Observe that negative correlation reduces the bound, while positive correlation increases it.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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