Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Let $B$ and $W$ be independent Brownian motions, let $\tau$ be a stopping time adapted to $\mathcal{F}^{W}$, do we always have $E[\int_{0}^{\tau}B_{s}dW_{s}]=0$?

I know that $\int_{0}^{t}B_{s}dW_{s}$ is a square integrable martingale, and I know that if the integrand were nonrandom, then the answer would be no. But I get stuck in the original question. I tried to use the tower property, but I don't know how to play with this Ito stochastic integral.

Thanks for your help!

share|cite|improve this question
What do you know? What have you tried? – Sasha Aug 17 '12 at 21:02
Nicely done! Now to @Sasha's question... – Did Aug 18 '12 at 8:16
@did Could you please comment on this question? – user7762 Aug 28 '12 at 0:37
up vote 0 down vote accepted

By the integration by parts formula $\int B_sdW_s = B_tW_t-\int W_s dB_s$ . Get rid of the first term by agreeing that we'll only look at stopping times where $W_{\tau} = 0$. Conditioning on all of $\mathcal F_W$ the r.h.s is $N(0, \int^{\tau}W^2(s) ds)$. Choose $\tau$ so that unconditioning yields a non $\mathbb L_1$ function. More specifically, let $X = e^{e^{W_1}}$ and choose a stopping time $\tau$ satisfying 3 conditions: $\tau > 1, W_{\tau} = 0, \int^{\tau}W^2(s) ds) \ge X$. The second condition can be satisfied because $W_t = 0$ for some arbitrarily large $t$, and the third can be satisfied because the integral increases to $\infty$. By the first remarks the integral we're looking at can be represented as $\sqrt{\int^{\tau}W^2(s) ds)} Z$ where $Z \sim N(0,1)$ and is independent of $\int^{\tau}W^2(s) ds)$. Since you have the distribution fairly explicitly, you can calculate $\mathbb E(\vert W_{\tau} \vert)$ by first conditioning on $\mathcal F_W$ and find that it is the same as $\sqrt{\frac 2 {\pi}} \mathbb E \sqrt{\int^{\tau}W^2(s) ds)} \ge \sqrt{\frac 2 {\pi}} \mathbb E \vert X \vert = \infty$ so the r.v. is not $\mathbb L_1$.

share|cite|improve this answer
Could you explain what do you mean by saying "choose $\tau$ so that unconditioning yields a non L1 function"? – user7762 Aug 28 '12 at 0:42
tried to do that. all r.v.'s of your form are symmetric so they are not going to have non-zero expectation, non $\mathbb L _1$ is the onl way you can fail to have expectation $0$. – mike Aug 28 '12 at 11:26
Could you explain why $\mathbb{E}[\int_{0}^{\tau}W_{s}dB_{s}|\mathcal{F}^{W}]$ has the same distribution as $\sqrt{\int_{0}^{\tau}W^{2}(s)ds}Z$ ? I don't know how to apply independence lemma to an ito integral. Thanks! – user7762 Aug 28 '12 at 23:09
When you condition on all of $\mathcal F_X$ the integral term is constant because it is $\mathcal F_W$ measurable, and by independence $B_t$ is the same brownian motion it was originally. So the condition distribution can be gotten from the distribution of terms like $\int f(t) dB_t$ where $f(t)$ is a deterministic fctn of $t$, and that is gaussian with mean $0$ and variance $\int f^2(t) dt$. – mike Aug 29 '12 at 11:39
I know heuristically this is how the independence lemma works. But I'm having trouble giving a rigorous proof. For example, what is an explicit formula for the r.v. $Z$? – user7762 Aug 29 '12 at 13:30

Edit: Oops, I missed the $\mathcal{F}^W$ in the original problem. So this doesn't answer the question, but I'll leave it anyway.

If you allow arbitrary stopping times (or even just those adapted to the filtration $\mathcal{F}^{B,W}$), the answer is no.

Let $X_t = \int_0^t B_s\,dW_s$. We note that the quadratic variation of $\newcommand{\X}{\langle X\rangle}X_t$ is $\X_t = \int_0^t B_s^2\,ds$. In particular, we have $\X_t \uparrow \infty$ almost surely.

For $a<0<b$, let $\tau_{a,b} = \inf\{t : X_t \notin (a,b)\}$. Also let $\tau_a = \inf\{t : X_t = a\}$. Since $X_t^2 - \X_t$ is a martingale, for any $a,b$ we have $$E[\X_{t \wedge \tau_{a,b}}] = E[X_{t \wedge \tau_{a,b}}^2] \le (|a| \vee |b|)^2.$$ But since $\X_t \uparrow \infty$, we must have $\tau_{a,b} < \infty$ almost surely.

Now $X_{t \wedge \tau_{a,b}}$ is a bounded martingale, so by optional stopping we have $$0 = E[X_{\tau_{a,b}}] = a P(\tau_a < \tau_b) + b (1-P(\tau_a < \tau_b)).$$ Thus $P(\tau_a < \tau_b) = b/(b-a)$. Letting $b \to \infty$, we have $P(\tau_a < \infty) = 1$. But then $E[X_{\tau_a}] = a \ne 0$.

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.