8

We know that given a continuous square integrable martingale there exists unique (up to indistinguishability) continuous, natural and increasing process which is quadratic variation process of the given martingale. I would like to know the converse i.e., given any continuous, natural and increasing process is it possible to get a martingale whose quadratic variation will be the given process? If not, please provide a counter example. Thanks

saz
  • 123,507
chandu1729
  • 3,871

1 Answers1

7

Yes, just take the Brownian motion with the change of time $A_s$ that's it to define $M_s = B_{A_s}$. This process will have the quadratic variation $A_s$.

The resulting $M$ is a local martingale with respect to the canonical filtration of the process $B$. If you want this to be square integrable you will have to ask $E(A_s) < \infty$ for each $s$. This is a consequence of the Dubins-Shwartz theorem.

Bunder
  • 2,493
  • This works only if you know that $A_s$ is a quadratic variation of some process! But my question was whether you can find any such process whose quadratic variation will be the given process. – chandu1729 Nov 29 '13 at 18:47
  • @chandu1729: No, this works. Just take $B$ to be a Brownian motion independent of $A$. (Actually, it should be added that in general you just get a local martingale, and not necessarily a proper martingale) – George Lowther Nov 30 '13 at 11:53
  • @George Lowther: I am not able to establish that $B_{A_s}$ is a martingale adapted to $\mathcal{F}_t$. Can you please give some ideas on it. Thanks. – chandu1729 Dec 01 '13 at 09:14