1
$\begingroup$

$W_{t}$ - Stochastic process ( Brownian motion). I need to check if $ (W_{2t}-W_{t})_{t \geqslant0}$ is a Brownian motion.

I look at the independent increment property. I want to find contradiction for this property.

So $t>s\geqslant u$ for any $s$ and $u$ that are less, less and equal, respectively.

So I look at: $$cov(W_{2t} - W_t - W_{2s} - W_s, W_{2u} - W_u) = cov(W_{2t} - W_t,W_{2u} - W_u) - cov(W_{2s} - W_s,W_{2u} - W_u)$$

if I suppose that $t > 2u$ then I can say that my first covariance is $0$ due to independence of increments.

Then I have left only:

$$ = 0 - cov(W_{2s} - W_s,W_{2u} - W_u) $$ which is definitely an overlapping intervals. So they have $cov \neq 0 $, or simply if I put $s = u$ then $cov = s=u$

Is it correct?


One more question:

Can I check the $(W_{2t}-W_{t}) - (W_{2s}-W_{s})$ is $N(0,\text{smth})$ ? or not? Why not?

Why I am asking, I did write that $W_{2t}-W_{t}$ $N(0,t)$, logically $W_{2s}-W_{s}$ $N(0,s)$, so the difference is $N(0,t+s)$ (by difference on normal r v). BUT!

If I rearrange other way, which also seems to be ok, $(W_{2t}-W_{2s}) - (W_{t}-W_{s})$ I get $N(0,3t-3s)$.

This is nonsense: I got 2 different distributions. But, which law I break when I do such calculations?


And the last question: suppose I have stochastic process $(-W_{t})_{t \geqslant0}$.

The property that $W_{t}-W_{s}$ is $N(0,t-s)$ in our case holds:

$W_{s}-W_{t}$ has $-N(0,t-s)$ which is $N(0,t-s)$ as normal distribution is symmetric? Is it right?


I copy-pasted from wiki:

The results for the expectation and variance follow immediately from the definition that increments have a normal distribution, centered at zero. Thus $W_t = W_t-W_0 \sim N(0,t)$ Does this leads to the following. Each element of this stochastic process is normally distributed? I am getting confused as $W_t \sim N(0,t)$ from the previous formula.

$\endgroup$
  • $\begingroup$ $(W_t)$ is a Brownian Motion, I'm assuming? $\endgroup$ – Shalop Oct 29 '15 at 20:49
  • 1
    $\begingroup$ $Z_t$ and $Z_s$ are not independent if $\frac t s\in (.5,2)$, so their difference is not just a normal with difference of means and sum of variances. $\endgroup$ – A.S. Oct 29 '15 at 22:03
  • $\begingroup$ I'm not really understanding your last question. Yes, it is true that each individual $W_t=W_t-W_0$ is normally distributed, simply by the definition of the Brownian Motion. $\endgroup$ – Shalop Oct 29 '15 at 22:03
  • $\begingroup$ My last question is basically: Is $W_t$ a r.v. that is normally distributed? So can we treat the $(W_t)_{t\geqslant 0}$ as sequence of r.v. n.d.? $\endgroup$ – Ievgenii Oct 29 '15 at 22:14
  • 1
    $\begingroup$ Yes, each $W_t$ is Normally distributed with variance $t-0=t$ and mean $0$. $\endgroup$ – Shalop Oct 29 '15 at 22:15
3
$\begingroup$

There is a slightly simpler solution. Let $Z_t:= W_{2t}-W_t$.

If $(Z_t)$ was a Brownian Motion, then it would be true that $cov(Z_t, Z_s) = s \wedge t$ for all $s,t \geq 0$. But, as one can check, this fails when $s=1$ and $t=\frac{3}{2}$, since $cov(Z_1,Z_{3/2})=1/2 \neq 1$.

In general, you should remember that a Gaussian Process with continuous paths is completely determined (in distribution) by its means and its covariances. For a Brownian Motion, the means are $0$ and covariances are $s \wedge t$.


To answer your second question: yes, it is true that $(W_{2t}-W_t)-(W_{2s}-W_s)$ is normally distributed. To prove this, assume $s<t$ without loss of generality.

When $2s\leq t$ note that $W_{2t}-W_t$, and $W_{2s}-W_s$ are independent Normals with variances $t$ and $s$ respectively, and hence their difference is Normal with variance $s+t$.

When $2s>t$, we can write $(W_{2t}-W_t)-(W_{2s}-W_s) = (W_{2t}-W_{2s})-(W_t-W_s)$ and by independence of increments we again see that this is a difference of two independent Normals with variances $2(t-s)$ and $t-s$, respectively, hence is Normal with variance $3(t-s)$.

So the different variances are a result of considering two separate cases, $2s \leq t$, and $2s>t$. Remember, increments are independent only when the intervals don't overlap.


To answer your last question, yes.

$\endgroup$

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.