3
$\begingroup$

I am looking at the process

$$X_t = \int_0^tB_udu$$

I know that this is a gaussian process with variance $t^3/3$. However, I would like to manually show the first statement directly.

For this, I would like to calculate the Laplace transform and show that it is the gaussian MGF. But the integral in the process makes that a tad difficult. My thought process that leads me to a slightly wrong result is:

We know that $B_t \sim \sqrt{t}B_1$. Thus: $$E\left[e^{\alpha X_t}\right] = E\left[e^{\alpha\int_0^tB_udu}\right] = E\left[e^{\alpha B_1\int_0^t\sqrt{u}du}\right],$$ which I can now integrate w.r.t. $B_1$. But that lands me at a MGF that is slightly off, so I suspect I can't do this transformation.

What would be the mathematically right way to calculate the MGF of $X$ here?

Edit: Just to be clear, of course you can do this with several ways, however I'm curious how to treat this specific situation - if possible at all.

$\endgroup$

1 Answer 1

3
$\begingroup$

We assume we work on a probability space $(\Omega,\mathcal{F},\mathbb{P})$ equipped with the filtration $\{\mathcal{F}_t\}_t$. By Itô's Lemma: $$B_t\text{d}t=\text{d}\left(tB_t\right)-t\text{d}B_t$$ Hence: $$X_t=tB_t-\int_0^tu\text{d}B_u$$ Let us define the function $\theta(t) :=\sqrt[3]{3t}$ and the filtration $\mathcal{F}^\theta_t:=\mathcal{F}_{\theta(t)}$. Introduce the following process: $$Y_t=\int_0^{\theta(t)}s\text{d}B_s$$ $Y_t$ is a local martingale with respect to the Brownian Motion with $Y_0=0$, hence its quadratic variation is: $$\begin{align} [Y,Y]_t=\int_0^{\theta(t)}s^2\text{d}s=\left[\frac{s^3}{3}\right]_0^{\theta(t)}=t \end{align}$$ It follows from Levy's characterization theorem that $Y_t$ is a Brownian Motion on the filtration $\mathcal{F}^\theta_t$ and thus is Gaussian.

Moreover, the time-changed Brownian Motion $\theta(t)B_{\theta(t)}$ also remains Gaussian with respect to the filtration $\mathcal{F}^\theta_t$. Indeed the function $\theta:\mathbb{R}^+\rightarrow \mathbb{R}^+$ is a mere deterministic bijection from $\mathbb{R}^+$ into itself with $\theta(0)=0$ hence $\theta(t)B_{\theta(t)}$ can be represented as $sB_s$ where $s\in\mathbb{R}^+$.

As a difference of Gaussian variables, $X_t$ is Gaussian with respect to the filtration $\mathcal{F}^\theta_t$ with same distribution as $\eta(t)+\xi(t)Z$, where $Z$ is a standard normal variable and $\eta(t), \xi(t)$ some deterministic functions of $t$. Changing back to the filtration $\mathcal{F}_t$, it is easy to see we are merely applying some deterministic transformation to the functions $\eta(t)$ and $\xi(t)$ thus $X_t$ remains Gaussian under the original filtration $\mathcal{F}_t$.

Background:

$\endgroup$
2
  • $\begingroup$ Good answer! However, since $\int f dB$ ist a gaussian process if f in L2, we would have the result immediately from application of the ito formula (difference of GP is GP). Currently revisiting a lot of that material for interview purposes and think I should revisit the time change, that sometimes yields really elegant solutions. However, it's not exactly what I'm looking for as I wanted to explicitly derive the MGF in this case (which involves an integral that one can't derive explicitly). $\endgroup$
    – not_sure95
    Mar 30, 2020 at 16:05
  • $\begingroup$ @not_sure95 The proof avoids any integral representation for any of the processes involved, given your question concerns a Gaussian integral. Can you include in your derivation how your proceeded to derive the MGF and what result you obtain? $\endgroup$ Mar 30, 2020 at 16:11

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.