• $(\Omega,\mathcal A,\operatorname P)$ be a probability space
  • $(\mathcal F_t)_{t\ge0}$ be a filtration of $\mathcal A$
  • $M$ be a $\operatorname P$-almost surely right-continuous $(\operatorname P,\mathcal F)$-martingale
  • $\sigma,\tau$ be $\mathcal F$-stopping times with $$\tau\le T\;\;\;\operatorname P\text{-almost surely}\tag1$$ for some $T\ge0$
  • $\mathcal F_\varsigma:=\left\{A\in\mathcal A:A\cap\left\{\varsigma\le t\right\}\in\mathcal F_t\text{ for all }t\in I\right\}$ for any $\mathcal F$-stopping time $\varsigma$

Now, let $$\sigma_n:=\frac1{2^n}\lceil 2^n\sigma\rceil\text{ and }\tau_n:=\frac1{2^n}\lceil 2^n\tau\rceil\;\;\;\text{for }n\in\mathbb N\;.$$ Note that $\sigma_n,\tau_n$ are $(\mathcal F_{k2^{-n}})_{k\in\mathbb N_0}$-stopping times and $(\sigma_n)_{n\in\mathbb N},(\tau_n)_{n\in\mathbb N}$ are nonincreasing with $$\sigma_n\xrightarrow{n\to\infty}\sigma\text{ and }\tau_n\xrightarrow{n\to\infty}\tau\tag2\;.$$

I want to conclude that $$\operatorname E\left[M_{\tau_n}\mid\mathcal F_{\sigma_n}\right]\xrightarrow{n\to\infty}\operatorname E\left[M_\tau\mid\mathcal F_\sigma\right]\tag3\;\;\;\operatorname P\text{-almost surely}\;.$$

Since $M$ is $\operatorname P$-almost surely right-continuous, we obtain $$M_{\tau_n}\xrightarrow{n\to\infty}M_\tau\;\;\;\operatorname P\text{-almost surely}\tag4\;.$$ However, I don't know how I need to proceed (especially with the indexed $\sigma$-algebra in the conditional expectation).

  • $\begingroup$ Chung's First Course in Probability has the proof of this theorem in later part of Chapter 9. I'll sketch the argument when I get home if there's no answer here by then $\endgroup$ – manofbear Jun 18 '17 at 17:15
  • $\begingroup$ @manofbear Thanks in advance. $\endgroup$ – 0xbadf00d Jun 18 '17 at 17:35
  • $\begingroup$ Bummer, turns out the theorem I was remembering is simpler than yours. I think the argument might adapt to your setting, but I'll work out the details first $\endgroup$ – manofbear Jun 18 '17 at 22:58
  • $\begingroup$ Is $M$ assumed to be square integrable? $\endgroup$ – saz Jun 19 '17 at 16:53

In my answer I assume that

$$\sup_{t \leq T+1} |M_t| \in L^1 \tag{5}$$

(for $T>0$ specified in $(1)$). By Doob's maximal inequality, this is, for instance, satisfied if $M_{T+1} \in L^p$ for some $p>1$.


$$|\mathbb{E}(M_{\tau_n} \mid \mathcal{F}_{\sigma_n})-\mathbb{E}(M_{\tau} \mid \mathcal{F}_{\sigma})| \leq I_1+I_2$$


$$\begin{align*} I_1 &:= |\mathbb{E}(M_{\tau} \mid \mathcal{F}_{\sigma_n})-\mathbb{E}(M_{\tau} \mid \mathcal{F}_{\sigma})| \\ I_2 &:= |\mathbb{E}(M_{\tau_n}-M_{\tau} \mid \mathcal{F}_{\sigma_n})|. \end{align*}$$

Since $\sigma_n \downarrow \sigma$, we have $\mathcal{F}_{\sigma_n} \downarrow \mathcal{F}_{\sigma}$, and therefore it follows from Lévy's convergence theorem that $I_1 \to 0$ almost surely as $n \to \infty$. Now fix some number $m \geq 1$. For $n \gg 1$ sufficiently large we have

$$I_2 \leq \mathbb{E} \left( \sup_{k \geq m} |M_{\tau}-M_{\tau_k}| \mid \mathcal{F}_{\sigma_n} \right).$$

Letting $n \to \infty$ it follows once again from Lévy's backward convergence theorem that

$$\limsup_{n \to \infty} I_2 \leq \mathbb{E} \left( \sup_{k \geq m} |M_{\tau}-M_{\tau_k}| \mid \mathcal{F}_{\sigma} \right). \tag{6}$$


$$\sup_{k \geq m} |M_{\tau}-M_{\tau_k}| \xrightarrow[]{m \to \infty} 0$$

and, by $(5)$,

$$\sup_{k \geq m} |M_{\tau}-M_{\tau_k}| \leq 2 \sup_{t \leq T+1} |M_t| \in L^1$$

an application of the dominated convergence theorem (for conditional expectations) shows that the right-hand side of $(6)$ converges to $0$ as $m \to \infty$. This finishes the proof.

| cite | improve this answer | |

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.