The solution to the SDE

$$dx_t= -kx_t dt + cx_t dW_t$$


$$x_t = x_0 e^{\left(c - \frac{k^2}{2} \right)t}e^{-k W_t}$$

with mean

$$\mathbb{E} \left[ x_t \right] = x_0 e^{\left(c - \frac{k^2}{2}\right)t}$$

where $W_)$ is the Wiener process.

I'm looking to compute

$\mathbb{E} \left[ (W_s + W_t - 2W_0)^2 \right]$

but am unsure of how to proceed.

  • $\begingroup$ I don't see how the computation of the expectation is related to the solution of the GBM SDE... $\endgroup$ – SRKX Jul 20 '14 at 21:11

I would calculate it this way,

$\mathbb{E}[(W_s+W_t−2W_0)^2] = \mathbb{E}\left[\left((W_s-W_0)+(W_t-W_0)\right)^2\right]\\ \hspace{4cm}=\mathbb{E}[(W_s-W_0)^2]+\mathbb{E}[(W_t-W_0)^2]+2\mathbb{E}[(W_s-W_0)(W_t-W_0)] \\ \hspace{4cm}=s+t+2\mathbb{E}[W_sW_t]\\ \hspace{4cm}=s+t+2\min(s,t)$

  • 1
    $\begingroup$ How do you come to $E(W_s,W_t)=\min(s,t)$? $\endgroup$ – emcor Jul 19 '14 at 21:56
  • 4
    $\begingroup$ Assume WLOG that $s<t$ then $E[W_sW_t]=E[W_s (W_t - W_s )+W_s^2]$ independent increments and the distributional properties and you are done. $\endgroup$ – htd Jul 20 '14 at 14:44
  • $\begingroup$ @Henrik I see that formula, but it does not depend on $s<t$, you can rewrite it both ways getting $E(W_sW_t)=s=t$ which is obviously wrong if $s<t$? $\endgroup$ – emcor Jul 21 '14 at 9:59
  • $\begingroup$ @emcor: Henrik answer is right! The formula depends on $s<t$. Indeed, $(W_t-W_s)$ is independent of $W_s$, it's Markov property that says that $(W_t-W_s)$ behave as a new brownien process independent of all what happens in the past $F_s$. Due to this, $\mathbb{E}[W_s(W_t-W_s)]=\mathbb{E}[W_s]\mathbb{E}[(W_t-W_s)]=0$. $\endgroup$ – aajajim Jul 22 '14 at 12:04
  • 1
    $\begingroup$ @emcor: It's Stochastic Calculus basics. math.stackexchange.com/questions/139260/… $\endgroup$ – aajajim Jul 28 '14 at 15:47

Your solution $x_t$ is wrong.

Your mean is wrong too. Note that $\mathbb{E}\left[ e^{W_t}\right] = e^\frac{t}{2}$. I corrected the typo that was pointed out by Richard.

  • 1
    $\begingroup$ $W_t$ is Gaussian - right? Then $E[e^{W_t}]$ is just the moment generating function of a Gaussian random variable with variance $t$ evaluated at $1$. So your formula is wrong. $\endgroup$ – Ric Jul 16 '14 at 6:51
  • $\begingroup$ $E[e^{W_t}] = \exp(t^2/2)$. $\endgroup$ – Ric Jul 16 '14 at 7:10
  • 1
    $\begingroup$ Richard: thanks for pointing out the typo. If $X \sim \mathcal{N}(\mu, \sigma^2)$, $\mathbb{E}\left[e^X\right] = e^{\mu + \frac{1}{2} \sigma^2}$. $\endgroup$ – wsw Jul 19 '14 at 16:00
  • 2
    $\begingroup$ Richard: Note that $W_t \sim N\Big(0, \big(\sqrt{t}\big)^2\Big)$. Then $E\big(e^{W_t} \big) = e^{\frac{t}{2}}$. $\endgroup$ – Gordon Aug 12 '14 at 13:33

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.