-2
$\begingroup$

Suppose I have some random variable $W$ along with its expectation $\mathbb{E}[W]$. My goal it to compute the integral

\begin{equation} \mathbb{E}\left[\int_{0}^{W}f(t)dt\right] = \int_{0}^{\mathbb{E}[W]}f(t)dt \end{equation}

Is the equality correct? If not, then what would be the correct result? Or is there any known theorem for this? Appreciate if anyone can provide such references.

$\endgroup$
3
  • $\begingroup$ Did you try with any function $f$? You are basically asking when $\mathbb{E}[F(W)] = F(\mathbb{E[W]})$ for a differentiable function $F$. $\endgroup$ Feb 6, 2017 at 6:35
  • $\begingroup$ Yes, that's what I mean basically, but I, in fact, did not come up with your representation. I have not tried any function yet, but my case will be continuous and differentiable functions with $\inf\{s > 0 : f(s) > 0\}$. $\endgroup$
    – Liäm
    Feb 6, 2017 at 10:39
  • $\begingroup$ Let me be more explicit: your equality will not hold in general, and examples are abundant. For example take $f(t) = 2t$. In that case you are asking if $\mathbb{E}[W^2] = \mathbb{E}[W]^2$, which holds if and only if $W$ has variance 0. In general, I expect that the only continuous $f$ for which your equality holds are constant on the range of $W$. $\endgroup$ Feb 6, 2017 at 13:51

1 Answer 1

1
$\begingroup$

\begin{align}\mathsf{E}\biggl[\int_0^{W}f(t)\mathrm{d}t\biggr] &=\mathsf{E}\biggl[\int_{-\infty}^{+\infty}f(t)(1_{\{0<t\le W\}}-1_{\{W\le t\le 0\}})\mathrm{d}t\biggr]\\ &= \int_0^{+\infty}f(t)\mathsf{P}(W\ge t)\mathrm{d}t-\int_{-\infty}^0f(t)\mathsf{P}(W\le t)\mathrm{d}t. \end{align}

$\endgroup$
3
  • $\begingroup$ Thanks! But I am not sure about the range of integration and the second indicator in your first equality. $W$ is always non-negative (maybe it was my fault not indicating this fact?). Why do we need to integrate over the entire entended $\mathbb{R}$? $\endgroup$
    – Liäm
    Feb 6, 2017 at 11:12
  • $\begingroup$ What is your concern? If $W$ is always non-negative, then the second term is 0 so you can ignore it. In the general case, the sign is negative since by convention $\int_{0}^w f(t) dt = -\int_{w}^0 f(t) dt$. $\endgroup$ Feb 6, 2017 at 14:34
  • $\begingroup$ @JamesMartin Thanks for your comment. $\endgroup$
    – JGWang
    Feb 7, 2017 at 1:59

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.