Let $M_1=X_1+1$ and let $M_n=(\log(M_{n-1})+1)^{X_{n}}$ where $(X_n)$ is an i.i.d. sequence of Poisson(1) distributed random variables. Also, let $(\mathbb{F}_n)$ be the filtration defined by $\mathbb{F_n}=\sigma(X_1,...,X_n)$.

I can see that $M_n$ is indeed $\mathbb{F}_n$ measurable, but how to prove the martingale property $\mathbb{E}(M_n|\mathbb{F_{n-1}})=M_{n-1}$?


Let $X\sim\operatorname{Poi}(\lambda)$ be a Poisson rv with expectation $\lambda$, let $s\in\mathbb R$. Then $$ Es^X = \sum_{k\ge0} s^k \frac{\lambda^k }{k!} e^{-\lambda} = \sum_{k\ge0} \frac{(s\lambda)^k }{k!} e^{-\lambda} = e^{s\lambda}e^{-\lambda} = e^{\lambda(s-1)}.$$ Use this with $X=X_n$, $\lambda=1$ and $s=\log M_{n-1}+1$ to get the desired martingale expectation formula $E(M_n|\mathbb F_{n-1}) = M_{n-1}$.

  • $\begingroup$ Why do we calculate $\mathbb{E}(s^{X{_n}})$ with respect to the law of $X_n$ only instead of the joint law of $X_1,...,X_n$? $\endgroup$ – Joogs Jan 16 '18 at 12:02
  • 1
    $\begingroup$ $X_n$ is independent of the previous $X_i$. By conditioning on $\mathbb F_{n-1}$ we have "taken into account" the previous $X_i$, whose effect is manifested in the value of $M_{n-1}$. On each atom of $\mathbb F_{n-1}$ the quantities $M_{n-1}$ and the previous $X_i$ are constants and $X_n$ is an independent Poisson. Atom by atom the expectation of $M_n$ works out to the "constant" $M_{n-1}$, which of course would typically be different on different atoms. $\endgroup$ – kimchi lover Jan 16 '18 at 12:23

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.