MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

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

I have a sequence of continuous time random variables $X_n(t)$ where $t \in [0,1]$. Suppose that there is a filtration $F_t$ such that for each $n$, $X_n$ is a martingale with respect to this filtration. Note that the filtration does not depend on $n$. Also assume that $$\sup_n E[\sup_{0 \leq t \leq 1 } |X_n(t)|] < \infty.$$

Finally suppose for every $t \in [0,1]$ $X_n(t)$ converge uniformly to a limit $X(t)$ and $X_n(t)$ converges in $L^1$.

I would like to know what are the weakest conditions I can assume about the $X_n(t)$ so that $X(t)$ will also be a martingale. If I assume the $X_n(t)$ are continuous or even left continuous then I believe that the limit will also be a martingale. If I just want to assume right continuity, I think I need that the fixed time distributions of $X(t)$ are non-singular. What I'd really like to know is if some kind of continuity of the $X_n$'s is a requirement to push the martingale through to the limit.

share|cite|improve this question
    
Isn't saying $X(t)$ is a martingale for $(F_t)$ equivalent to saying $E[X(T)] = E[X(0)]$ for all stopping times $T$? If that is so, then won't your uniform convergence preserve this? – Gerald Edgar Dec 13 '11 at 1:51
    
@Gerald: Yes, but he didn't say anything about convergence of X(T) for stopping times T. – George Lowther Dec 13 '11 at 2:53
4  
But, the fact that $X_n(t)$ converges to $X(t)$ in $L^1$ for each time $t$ is enough. For any $0\le s < t\le1$ and $A\in F_s$, you have$$\mathbb{E}[1_A(X(t)-X(s))]=\lim_n\mathbb{E}[1_A(X_n(t)-X_n(s))]=0.$$ – George Lowther Dec 13 '11 at 2:55
    
@George. Yes, of course. I got confused about the definition of a martingale. Your calculation shows that for fixed $t,s$, $E[X(t)|F_s] = X(s)$ a.s., and I thought I wanted something like the statement, a.s. for every fixed $t,s$, $E[X(t)|F_s] = X(s)$ which in retrospect doesn't make sense since the equality is only defined up to measure zero sets. Thanks. – Ben Dec 13 '11 at 3:18
up vote 2 down vote accepted

Posting the answer already given in comments:

All you really need is that $X_n(t) \to X(t)$ in $L^1$ for each $t$. Conditional expectation with respect to any $\sigma$-field is continuous with respect to $L^1$ convergence; this is an elementary property of conditional expectation which follows from the inequality $\left|E[X \mid \mathcal{F}]\right| \le E [|X| \mid \mathcal{F}]$ and taking unconditional expectation of both sides.

Given this, since for each $s \le t$ and each $n$ we have $X_n(s) = E[X_n(t) \mid \mathcal{F}_s]$, you may pass to the $L^1$ limit on both sides to get $X(s) = E[X(t) \mid \mathcal{F}_s]$. So $X$ is a martingale.

share|cite|improve this answer

Your Answer

 
discard

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.