0
$\begingroup$

Given $M$ a martingale starting from $0$ with continuous sample paths and $(M_t)_{t\ge 0}$ is a Gasussian process. I want to show $\langle M,M\rangle_t$ (the quadratic variation of $M_t$) is a monotone nondecreasing continuous function of $t$).

I was considering using Doob-Meyer. By Doob-Meyer there exists a nondecreasing stochastic process $(A_t)$ with continuous sample paths such that $A_t=<M_t,M_t>_t.$

Is there a different way to proceed by using the fact that $M_t$ is Gaussian?

$\endgroup$
2
$\begingroup$

In fact (and maybe this is what you are saying you want to show), the quadratic variation of $M$ is deterministic; more precisely, $\langle M\rangle_t = E[M^2_t]$ a.s. Here is the outline of a proof. The function $q(s):=E[M_s^2]$ is continuous and non-decreasing. A Gaussian martingale necessarily has independent increments. Fix $t>0$. For a given positive integer $n$ choose $0=t_0<t_1<t_2<\cdots<t_{n-1}<t_n=t$ so that $q(t_{k+1})-q(t_k)=q(t)/n$ for $k=0,1,2,\ldots,n-1$. The increments $M_{t_1}, M_{t_2}-M_{t_2},\ldots,M_{t_n}-M_{t_{n-1}}$ are independent Gaussian random variables with mean $0$ and variance $q(t)/n$. The sum of squares $\sum_{k=1}^n[M_{t_k}-M_{t_{k-1}}]^2$ therefore has the same distribution as $(q(t)/n)\sum_{k=1}^n Z_k^2$, where $Z_1,Z_2,\ldots$ are iid standard normal; by the law of large numbers, this latter sum converges a.s. to $q(t)$ as $n\to\infty$. On the other hand, $\sum_{k=1}^n[M_{t_k}-M_{t_{k-1}}]^2$ converges in probability to $\langle M\rangle_t$. It follows that $\langle M\rangle_t=q(t)$ a.s.

| cite | improve this answer | |
$\endgroup$

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.