Let $B_t$ be the standard Brownian motion process, $a > 0$, and let $H_a = \inf \{ t : B_t > a \}$ be a stopping time. I want to show that the Laplace transform of $H_a$ is $$\mathbb{E}[\exp(-\lambda H_a)] = \exp (-\sqrt{2\lambda} H_a)$$ by considering the martingale $$M_t = \exp \left(\theta B_t -\frac{1}{2}\theta^2 t\right)$$

There's an obvious argument to follow here: assuming the optional stopping theorem applies, we have $$1 = \mathbb{E}[M_{H_a}] = \mathbb{E} \left[ \exp \left(\theta a - \frac{1}{2}\theta^2 H_a\right) \right] = \exp(\sqrt{2\lambda} a) \mathbb{E} \left[ \exp(-\lambda H_a) \right]$$ where $\theta = \sqrt{2\lambda}$. This is exactly what we wished to show. However, as far as I can tell, the hypotheses of the optional stopping theorem are not satisfied here. Here is the statement I have:

If $(X_n)$ is a martingale and $T$ is an a.s. bounded stopping time, then $\mathbb{E}[X_T] = \mathbb{E}[X_0]$.

I think not all is lost yet. $M_t > 0$ for all $t$, so the martingale convergence theorem applies, and $M_t \to M_\infty$ a.s. for some integrable random variable $M_\infty$. For each $t$, $H_a \wedge t = \min \{ H_a, t \}$ is a bounded stopping time, so certainly $\mathbb{E}[M_{H_a \wedge t}] = \mathbb{E}[M_0]$. But, $$\mathbb{E}[M_{H_a \wedge t}] = \mathbb{E}[M_{H_a} \mathbf{1}_{\{H_a \le t\}}] + \mathbb{E}[M_t \mathbf{1}_{\{H_a > t\}}]$$ and clearly what one wants to do is to take $t \to \infty$ on both sides. But here's where I get stuck: I'm sure I need a convergence theorem here in order to conclude that the equation remains valid in the limit.

Now, $0 < M_{H_a} = \exp \left(\theta a - \frac{1}{2} \theta^2 H_a \right) \le \exp(\theta a)$, so the dominated convergence theorem applies, and so $$\lim_{t \to \infty} \mathbb{E}[M_{H_a} \mathbf{1}_{\{H_a \le t\}}] = \mathbb{E}[M_{H_a} \mathbf{1}_{\{H_a < \infty\}}]$$ and I believe Fatou's lemma gives me that $$\liminf_{t \to \infty} \mathbb{E}[M_t \mathbf{1}_{\{H_a > t\}}] \ge \mathbb{E}[M_{\infty} \mathbf{1}_{\{H_a = \infty\}}]$$ but I think what I need is the equality $$\lim_{t \to \infty} \mathbb{E}[M_t \mathbf{1}_{\{H_a > t\}}] = \mathbb{E}[M_\infty \mathbf{1}_{\{H_a = \infty\}}]$$ and as far as I can tell neither the monotone convergence theorem nor the dominated convergence theorem applies here. Is there anything I can do to rescue this line of thought?

  • 3
    $\begingroup$ Use $0\le M_t\mathbf{1}_{\{H_a>t\}}\le\exp(\theta a-\frac12\theta^2t)$. $\endgroup$
    – Did
    Mar 10 '11 at 21:05
  • $\begingroup$ @Didier: Thanks! Not sure why I keep missing the obvious. $\endgroup$
    – Zhen Lin
    Mar 10 '11 at 21:24
  • $\begingroup$ This is to second Byron's comment. In fact, I wanted to mention this in my first comment but I forgot. $\endgroup$
    – Did
    Mar 11 '11 at 6:12
  • 2
    $\begingroup$ @Didier: If you post your comment as an answer, we can vote it up and put this question to bed. $\endgroup$ Jun 9 '11 at 19:24
  • $\begingroup$ @ZhenLin There is a typo in the first equation. The RHS should be $\exp(-\sqrt{2\lambda}a )$ $\endgroup$
    – d.k.o.
    May 24 at 10:18

Use the fact that $0\leqslant M_t\mathbf{1}_{\{H_a>t\}}\leqslant\exp\left(\theta a−\frac12\theta^2t\right)$.


I m not sure that I understand properly the question.
We have $M_{H_a\wedge t}\to M_{H_a}$ almost surely, and
$M_{H_a\wedge t}<e^{\theta.a}$ The right hand side is constant so integrable, so doesn't the dominated convergence readily applies for $t \to +\infty$ ? Am i missing something here ?


  • $\begingroup$ No, I'm afraid uniform integrability is something I never really understood well and have completely forgotten. $\endgroup$
    – Zhen Lin
    Mar 11 '11 at 7:27
  • $\begingroup$ It is not true that a positive martingale converges in $L^1$. It converges a.s. and is bounded in $L^1$, but may not be uniformly integrable. The critical binary branching process is an example. $\endgroup$
    – user940
    Mar 11 '11 at 16:06
  • $\begingroup$ In fact, the OP's positive martingale does converge to zero almost surely and therefore it cannot be UI. $\endgroup$
    – user940
    Mar 11 '11 at 16:10
  • $\begingroup$ @Byron Schmuland : You are right I edited my post consequently, sorry about that Zhen Lin. $\endgroup$
    – TheBridge
    Mar 11 '11 at 16:53

I don't think you need to worry about the case $H_a = \infty$ - for the simple reason that this is an event of probability zero.

  • $\begingroup$ This does not work. Even when the probability of a stopping time being infinite is zero, if the probability of it being very large is sufficiently large, the optional stopping theorem fails. The standard example is Gambler's Ruin: if the game is to flip a coin and we have infinite capital, then a winning strategy is $b_n = 1 + \sum_{i=1}^{n-1} b_i$, $b_1 = 1$. (Or in other words $b_n = 2^{n-1}$.) This strategy has expected winnings $1$, but when the optional stopping theorem applies, the expected winnings from any strategy with a fair game is zero. $\endgroup$
    – Ian
    Oct 8 '14 at 19:54

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.