Suppose $(\Omega,\mathcal F,\{\mathcal F_n\}_n,\mathbb P)$ is a filtered probability space. Let $\tau$ be a stopping time, and define a collection of sets $\mathcal F_\tau:=\{A\in\mathcal F: A\cap\{\tau=n\}\in\mathcal F_n\forall n<\infty\}$. We can easily show that $\mathcal F_\tau$ is a $\sigma$-algebra.

  1. Let $M=\{M_n\}_n$ be a martingale. Show that $\mathbb E[M_n|\mathcal F_\tau]=M_{n\wedge\tau}, n\ge 0$.

  2. Let $M=\{M_n\}_n$ be a uniformly integrable martingale with last element $M_\infty\in\mathcal F_\infty$. Show that $\mathbb E[M_\infty|\mathcal F_\tau]=M_\tau$.

My attempt:

I would use the definition of conditional expectation to show both of the identities.

For any $S\in\mathcal F_\tau$, we want to show that $\mathbb E[M_{n\wedge\tau};S]=\mathbb E[M_n;S]$. But then I got completely stuck for hours. Even after checking a lot of posts dealing with this kind of problem, I am still quite confused with the definition of $\mathcal F_\tau$. Can someone help me out? Thank you!


1 Answer 1


First we have to show that $M_{n\wedge \tau}$ is $\mathcal F_{\tau}$ measurable. For this we have to show that $(M_{n\wedge \tau} \leq x)\cap (\tau=k) \in \mathcal F_k$. But $(M_{n\wedge \tau} \leq x)\cap (\tau=k)=(M_{n\wedge k} \leq x)\cap (\tau=k)$ which is an intersection of two sets in $\mathcal F_k$.

Next we have to show that $\int_A M_{n\wedge \tau} dP=\int_A M_n dP$ for $A \in \mathcal F_{\tau}$. It is enough to show that $\int_{A\cap (\tau=k)} M_{n\wedge \tau} dP=\int_{A\cap (\tau=k)} M_n dP$ for each $k$. In other words we have to show $\int_{A\cap (\tau=k)} M_{n\wedge k} dP=\int_{A\cap (\tau=k)} M_n dP$. This follows by martingale property if $n > k$ and trivially holds if $n \leq k$.

The second part is just obtained by taking limit as $n \to \infty$ since a uniformly integrable martingale converges almost surely and in $L^{1}$.

  • $\begingroup$ Thank you for your answer! But could please explain the intuition of $\mathcal F_\tau$? I am very confused with the definition of $\mathcal F_\tau$. $\endgroup$
    – Bach
    Dec 10, 2020 at 7:43
  • $\begingroup$ It is the sigma algebra of events up to to the random time $\tau$. Since $\mathcal F_n$'s are orignally defined only for fixed time $n$ we have to consider $A \cap (\tau=n)$ to get a meaningful definition of $A \in \mathcal F_{\tau}$. @Bach $\endgroup$ Dec 10, 2020 at 7:47
  • $\begingroup$ It seems to me that saying $A\in\mathcal F_\tau$ is the same as partitioning $A$ so that each part represents $A$ stops at time $i$ where $i$ varies in $1,2,......$. And if we pick up one part from those partitions, then the corresponding value of the stopping time on that part gives enough infomation that whether we stop or not? $\endgroup$
    – Bach
    Dec 10, 2020 at 8:07

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.