2
$\begingroup$

Edit: Question Resolved. See below.

As a part of my bachelor thesis, I have to work through a paper about fake Brownian motion by Oleszkiewicz. In this paper he defines a stochastic process.

Let $G_1, G_2, W_t$ be independent. Here we have $G_1$, $G_2$ $\sim \mathcal{N}(0,1)$ and $W_t$ Brownian motion. For $a\geq 0$ and $t\geq e^{-a}$ define the filtration $\mathcal{F}_t^{(a)}=\sigma(G_1,G_2,(W_s)_{0\leq s\leq a+\ln t}),$ and the process $$X_t^{(a)} = \sqrt{t}\left(G_1\cos{W_{a+\ln t}}+ G_2\sin{W_{a+\ln t}}\right).$$

Now Oleskiewicz claims,

(1) $X_t^{(a)}$ is a continuous martingale,

(2) $X_t^{(a)}\sim \mathcal{N}(0,t)$,

(3) $X_e^{(a)}-X_1^{(a)}$ is not gaussian,

without actually proofing any of these statements.

While I managed to proof the first two statements, the proofs are rather long and involved, so I wanted to ask if I am missing something obvious. Furthermore, I am absolutely clueless regarding the third statement.

I hope you can help me.

$\endgroup$
  • $\begingroup$ The question has been resolved. Professor Oleszkiewicz took the time to answer it himself, after I sent him an E-mail. $\endgroup$ – Alex Jun 10 '14 at 21:55
  • $\begingroup$ Would you mind posting the answer? I found the question interesting. $\endgroup$ – Ian Jun 11 '14 at 0:10
1
$\begingroup$

This is the answer he send me regarding these increments.

"As for the increments, $X^{(a)}_e-X^{(a)}_1= (\sqrt{e}\cos W_{a+1}-\cos W_a)G_1+(\sqrt{e}\sin W_{a+1}-\sin W_a)G_2.$ Since the pair $(W_a, W_{a+1})$ is independent of the pair $(G_1, G_2)$, this increment has the same distribution as $[(\sqrt{e}\cos W_{a+1}-\cos W_a)^2+ (\sqrt{e}\sin W_{a+1}-\sin W_a)^{2}]^{1/2} G$, where $G$ is $\mathcal{N}(0,1)$ independent of the pair $(W_a, W_{a+1})$.

Now, if $Z$ and $G$ are independent, $G$ is $\mathcal{N}(0,1)$-distributed and $ZG$ is Gaussian then $|Z|$ must be constant a.s. -- indeed, $\mathbb{E}ZG=\mathbb{E}Z \mathbb{E}G=\mathbb{E}Z \cdot 0=0$, so that $ZG$ is a centered Gaussian random variable; now it suffices to notice that $\mathbb{E}G^4=(\mathbb{E}G^2)^2, \mathbb{E}(ZG)^4=3(E(ZG)^2)^2$, so that $EZ^4=(EZ^2)^2$ and hence $Var(Z^2)=0$, i.e., $Z^2$ is constant a.s.

However, the expression standing inside $[...]^{1/2}$ obviously is not constant a.s., as it is equal to $e+1-2\sqrt{e}\cos(W_{a+1}-W_a)$, and $W_{a+1}-W_{a}$ is a $\mathcal{N}(0,1)$ Gaussian random variable."

It's a quite impressive and interesting approach.

$\endgroup$

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.