Let $B=(B_t)_{0\le t\le T}$ be a continuous semi-martingale and $\mathbb F=(\mathcal F_t)_{0\le t\le T}$ be its natural filtration. Denote by $\mathcal C_b(\Omega\times \mathbb R_+)$ the space of continuous bounded functions $F:\Omega\times \mathbb R_+ \to \mathbb R$, where $\Omega$ denotes the space of continuous functions defined on $[0,T]$ endowed with the uniform norm, and $\Omega\times\mathbb R_+$ is equipped with the product topology.

Now, consider a sequence of $\mathbb F$-stopping times $(\tau_n)_{n\ge1}$ s.t. there exists a random variable $\tau$ satisfying

$$\lim_{n\to \infty}\mathbb E[F(B,\tau_n)]~~=~~\mathbb E[F(B,\tau)],~~ \mbox{ for all } F\in\mathcal C_b(\Omega \times \mathbb R_+).$$

My question is: could we show that $\tau$ is an $\mathbb F$-stopping time? If not, what about assuming that $B$ is Markov or even a Brownian motion? Thanks a lot for the reply!


1 Answer 1


In the case of Brownian motion you can look at functions of the form $$g(\tau_n \wedge t) e^{i \sum \lambda_i(B_{t_i} - B_t)}$$ where $t_i > t$. Since the expectation is $$ \mathbb E(g(\tau_n \wedge t)) \mathbb E(e^{i \sum \lambda_i(B_{t_i} - B_t)})$$ which converges to $$ \mathbb E(g(\tau \wedge t)) \mathbb E(e^{i \sum \lambda_i(B_{t_i} - B_t)})$$ $g(\tau \wedge t)$ is independent of $\mathcal F (B_s - B_t), s>t$ which should make it $\mathcal F_t$ measurable. $${}$$ As pointed out below, that doesn't work, however the brownian case can be done completely differently by
1. continuous bounded functions of $B_{t_1},...,B_{t_n}$ are dense in $L^2(\mathcal F_{\infty}) $ because, e.g. lemma 5.3.1 of Revuz and Yor shows that linear combinations of $ sin(\sum \lambda_i B_{t_i}), cos(\sum \lambda_i B_{t_i})$ are dense. $${}$$2. Let $\epsilon > 0$. Approximate $\tau \wedge t$ to within $\frac {\epsilon} 2$ by a continuous bounded function of $B_{t_1},...,B_{t_n}$ which I'll call f. By the hypotheses $\mathbb E(f-\tau_n \wedge t)^2 \rightarrow \mathbb E(f-\tau \wedge t)^2 < \frac {\epsilon} 2$ Therefore by the triangle inequality $limsum E(t_n \wedge t - \tau \wedge t)^2 < \epsilon$. This shows that $\tau_n \wedge t \rightarrow \tau \wedge t$ in $L^2$, and therefore $\tau^t $ is a stopping time. (exercise 1.4.17 of revuz & yor). $${}$$

  • $\begingroup$ @ michael Thanks for the reply. Yes, I get the proof for a BM. Do you have an idea for a general diffusion process? If not, could we find some stopping time $\sigma$ s.t. $(B,\tau)$ has the same law of $(B,\sigma)$? $\endgroup$
    – CodeGolf
    Mar 30, 2016 at 7:28
  • $\begingroup$ I think with a lot of diffusions (bounded, smooth parameters say) you can use time changes & girsanov to convert it to the Brownian motion problem $\endgroup$
    – user83457
    Mar 30, 2016 at 8:58
  • $\begingroup$ @ michael I will try it. Could you specify a bit more on how to show that $\tau\wedge t$ is $\mathcal F_t$ measurable? $\endgroup$
    – CodeGolf
    Mar 30, 2016 at 9:37
  • $\begingroup$ Knowing that $\tau \wedge t$ is independent of $\mathcal F(B_s-B_t, s\ge t)$, how can we show $\tau \wedge t$ is $\mathcal F_t$ measurable? $\endgroup$
    – CodeGolf
    Mar 30, 2016 at 9:38
  • $\begingroup$ @michael's conclusion about the measurability of $\tau\wedge t$ needs some justification. Example: Toss two fair dice; let $X_i$ be the number showing on the top of die $i$. Let $Y$ be the indicator of the event that $X_1+X_2$ is even. The $\sigma$-algebra of interest is $\mathcal F=\sigma(X_1,X_2)$, and (i) $X_1$ and $X_2$ are independent, (ii) $Y$ is $\mathcal F$-measurable, (iii) $Y$ is independent of $X_2$, but (iv) $Y$ is not $\sigma(X_1)$ measurable. $\endgroup$ Apr 2, 2016 at 17:28

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.