0
$\begingroup$

Let $X$ be a non-negative random variable. In a proof for $E[X]=\int_0^\infty P(X>t)dt$ from the answer of this question, we use Fubini for the middle quality. Why do we need $X$ to be non-negative? We basically have a double integral over a function $f(X,t)$ which is $1$ if $X>t$ and else 0. So this function is non-negative for any $X$ not just for non-negative random variables $X$, thus we could use Fubini regardless. Where is the flaw in my reasoning?

$\endgroup$
1
1
$\begingroup$

We require $X$ to be non-negative, because otherwise \begin{equation} X \neq \int_{[0,X)} 1 \ dt, \ \ \mbox{ so in general} \ \ E[X] \neq E \left ( \int_{[0,X)} 1 \ dt \right ). \end{equation}

For example, if for some $\omega_0 \in \Omega$ we have that $X(\omega_0) = -5 < 0$, then $[0,X(\omega_0)) = \emptyset$ and $\int_{[0,X(\omega_0))} 1 \ dt = 0 \neq X(\omega_0)$, so we are in trouble if $\mathbb{P} (\omega_0) >0$. Thus, $X$ is required to be non-negative in order for the first equality to hold, not because of the Fubini theorem.

Not that for a non-positive $X$ you get \begin{equation} E[X] = -E[-X] =- \int_0^{\infty} P (- X > t)\ dt= \int_{-\infty}^0 P (X <s )\ ds. \end{equation}

$\endgroup$
1
$\begingroup$

If $x \ge 0$, we can write $x = \int_0^\infty 1_{(t,\infty)}(x) dt$.

Then $E X = \int X d p = \int \int_0^\infty 1_{(t,\infty)}(X) dt d p = \int_0^\infty \int 1_{(t,\infty)}(X) dp dt = \int_0^\infty p \{ \omega|X(\omega)>t\} dt$.

$\endgroup$
1
  • $\begingroup$ Yes, this is just the proof given in the question I referenced. My confusion was with us not using the non-negativity for Fubini, but for the first equality. $\endgroup$
    – fstl
    Dec 19 '14 at 21:26
1
$\begingroup$

In general, $$\int_0^\infty P(X>t)dt=\int_0^\infty P(\max(X,0)>t)dt=E(\max(X,0)).$$

$\endgroup$
0
$\begingroup$

There is no flaw in the Fubini part. However, the first equality in that proof won't work if you allow $X$ to have negative values.

$\endgroup$

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.