I have a statement that reads:

If $Z_1,Z_2$ are random variables such that $Z_1 \geq Z_2$, then $\rho(Z_1) \geq \rho(Z_2)$.

where $\rho$ is a function.

What is the meaning of $Z_1 \geq Z_2$? I am particularly interested in the monotonicity of coherent risk measures. For discrete random variables, say:

$\Pr[Z_1=x_i] = p_i$ and $\Pr[Z_2=y_j] = q_j$

how can I write the condition $Z_1 \geq Z_2$ using $x_i,p_i,y_j,q_j$?

  • $\begingroup$ Inequality $Z_1 \geqslant Z_2$ holds surely, i.e for all $\omega \in \Omega$, $Z_1(\omega) \geqslant Z_2(\omega)$. Examples could be $Z_1 = X -1$, $Z_2 = X$ or, for independent uniform random variables $U_1 \sim \mathcal{U}(2,3)$ and $U_2 \sim \mathcal{U}(0,1)$, $Z_1 = U_1$ and $Z_2 = U_2$. $\endgroup$ – Sasha Aug 27 '12 at 3:36

First, a comment states that $Z_1\ge Z_2$ means the inequality holds surely, i.e., that it holds for every $\omega$ in the probability space $\Omega$. (There must be such a space, or else the notions of probability makes no sense, though it is quite common not to specify or even mention it explicitly.) But in practice, one uses the weaker statement $Z_1\ge Z_2$ almost surely (abbreviated a.s.), which means $\Pr[Z_1<Z_2]=0$. Quite frequently, I suspect this is the intended meaning even when the author omits the “a.s.” part.

You cannot write the condition $Z_1\ge Z_2$ using the quantities you list. For that, you need the combined probability distribution, i.e. you need to know the probabilities $P_{ij}=\Pr[Z_1=x_i \mathbin\& Z_2=y_j]$. In this case, $Z_1\ge Z_2$ a.s. if and only if $P_{ij}=0$ whenever $x_i<y_j$. If you don't include the “a.s.”, then even the full probability distribution won't provide the answer. You need the underlying probability space in order to answer the question in that case.

| cite | improve this answer | |
  • $\begingroup$ +1. And one knows exactly what are the marginals (the distributions of $Z_1$ and $Z_2$) such that there exists a joint distribution (of $(Z_1,Z_2)$) with the property that $Z_1\leqslant Z_2$ almost surely. Keyword: coupling. $\endgroup$ – Did Aug 27 '12 at 7:07

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.