2
$\begingroup$

I am trying to find a reference (or, if it's false, a counterexample) for the following sort-of-intuitive fact: if $\tau$ is a stopping time with a subexponential probability distribution, and $(X_n)_{n\geq 1}$ are independent r.v.'s, also subexponential, then $\sum_{n=1}^\tau X_n$ aso has a subexponential distribution.

Specifically, I would like to know if the following statement is known:

Let $(X_n)_{n\geq 1}$ be independent random variables satisfying $\mathbb{E}[e^{X_n}] \leq 1$ for all $n$, and $\tau$ be a stopping time. Suppose $\mathbb{E}[e^{\alpha \tau}] \leq e^\beta$ for some $\alpha >0$ and $\beta<\infty$. Then $\mathbb{E}[e^{\sum_{n=1}^\tau X_n}] \leq 1$.

$\endgroup$

1 Answer 1

3
$\begingroup$

The hypothesis implies that $M_k=[e^{\sum_{n=1}^k X_n}]$ is a supermartingale, with $M_0=1$. Then the optional stopping theorem for positive supermartingales implies the requested inequality. (see e.g. Williams' book "Probability with martingales"). Note that no moment conditions on the stopping time $\tau$ are needed, just that it is an almost surely finite stopping time. Alternatively, look up "Wald's third identity"and apply it to the independent variables $e^{X_n}/[\mathbb{E}e^{X_n}]$ and the given stopping time. Note that the case of a general stopping time follows from the case of a bounded stopping time via Fatou's lemma.

$\endgroup$
2
  • $\begingroup$ Thank you! I can't believe it was so simple -- I got stuck at the almost sure boundedness assumption for $\tau$ in Wald's third identity. If I understand correctly, the assumption on $\tau$ can even be relaxed: all that's needed is almost sure finiteness. $\endgroup$
    – Clement C.
    May 29, 2020 at 23:09
  • 2
    $\begingroup$ You need some assumption on $\tau$ for Wald's identity to hold. But what you asked is an inequality, so you could apply Wald to $\min\{\tau,m\}$ and then let $m \to \infty$ using Fatou. $\endgroup$ May 29, 2020 at 23:47

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.