3
$\begingroup$

For $(M_t)_{t\geq0}$ to be a martingale w.r.t a filtration $\mathcal{F}_t$, we require

  1. $M_t \in L^1$
  2. $E[M_t \mid \mathcal{F}_s] = M_s$, $t\geq s$

If it is known that $M_t \geq 0$, does one have to prove condition 1? Or does it fall out of condition 2 since $$E[M_t] = E[E[M_t \mid \mathcal{F}_0]] = E[M_0] < \infty?$$ My concern is whether the law of iterated expectations holds if we don't know apriori that $M_t \in L^1$. Relatedly, are there examples of stochastic processes where $M_t \notin L^1$ but $E[M_t \mid \mathcal{F}_s] = M_s$?

$\endgroup$
3
$\begingroup$

It is part of the definition of the conditional expectation that $$E[M_t 1_A] = E[E[M_t \mid \mathcal{F}_0] 1_A]$$ for any $A \in \mathcal{F}_0$. By taking $A = \Omega$ we see that the law of iterated expectations is an immediate consequence of the definition (without assuming that $M_t \in L^1$).

As a result, if $M_0 \in L^1$ and $(M_t)_{t \geq 0}$ satisfies condition $2$ then by your reasoning we have that $M_t$ is a martingale.

It's also fairly easy to see that there exist processes $M_t \not \in L^1$ such that the conditional expectations exist and $E[M_t \mid \mathcal{F}_s] = M_s$. For example, let $X$ be a non-integrable, non-negative random variable and for every $t \geq 0$, let $\mathcal{F}_t = \sigma(X)$ and $M_t = X$. Then we obviously have $$E[M_t \mid \mathcal{F}_s] = E[X \mid \sigma(X)] = X.$$

$\endgroup$

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.