Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

How can we compute the mean and variance of $e^{W_tW_s} $ where $(W_t)_{t \geq 0} $ is a Brownian motion?

If we want to compute $ \mathbb{E}(W_tW_s) $, the usual thing to do is to assume that $ s \leq t$ and write $W_tW_s = (W_t - W_s)W_s + W_s^2$, then we can use independence of increments to finish that computation, but due to the exponential function, that trick does not work here unless we can write $W_tW_s$ as a sum of independent increments, so that $e^{W_tW_s}$ can be written as a product of functions of the independent increments.

Does anyone have any suggestions?

share|cite|improve this question
up vote 5 down vote accepted

Recall that $(W_s, W_t)$ are jointly Gaussian with zero mean and covariance matrix, assuming without loss of genarility $s < t$ $$ \begin{pmatrix}\mathbb{Cov}(W_s,W_s) & \mathbb{Cov}(W_s,W_t) \\ \mathbb{Cov}(W_t,W_s) & \mathbb{Cov}(W_t,W_t) \end{pmatrix} = \begin{pmatrix}s & s \\ s & t \end{pmatrix} $$ We thus faced with finding an expectation of a function of two random variables. To make it simple, we recall that the Gaussian random variable $V = W_t-W_s$ is independent of $W_s$. Thus, usign $(W_s, W_t) \stackrel{\text{law}}{=} (W_s, W_s+V)$, we have $$ \mathbb{E}\left(\exp\left(W_s W_t\right)\right) = \mathbb{E}\left(\exp\left(W_s \left(W_s + V\right)\right)\right) = \mathbb{E}\left(\exp\left(W_s^2 \right) \exp\left(W_s V \right) \right) $$ Now, making use of the law of the total expectation: $$\begin{eqnarray} \mathbb{E}\left(\exp\left(W_s^2 \right) \exp\left(W_s V \right) \right) &=& \mathbb{E}\left( \mathbb{E}\left(\exp\left(W_s^2 \right) \exp\left(W_s V \right) |W_s \right) \right) \\ &=& \mathbb{E}\left( \exp\left(W_s^2 \right) \mathbb{E}\left(\exp\left(W_s V \right) |W_s \right) \right) \tag{1} \end{eqnarray} $$ As $V$ is a Gaussian random variable, we have, for any real $w$, $$\mathbb{E}\left( \exp(w V) \right) = \exp\left( \mathbb{E}(V)w + \frac{w^2}{2} \mathbb{Var}(V) \right) = \exp\left( \frac{w^2}{2} \left(t-s\right) \right) $$ Therefore, eq. $(1)$ now reads: $$ \mathbb{E}\left( \exp\left(W_s^2 \right) \mathbb{E}\left(\exp\left(W_s V \right) |W_s \right) \right) = \mathbb{E}\left( \exp\left( W_s^2 \left( 1 + \frac{t-s}{2} \right) \right)\right) $$ Recall that $W_s \stackrel{\text{law}}{=} \sqrt{s} Z$, for standard normal random variable $Z$, thus $W_s^2 \stackrel{\text{law}}{=} s Z^2 \stackrel{\text{law}}{=} s X$, where $X$ follows $\chi^2$-distribution with one degree of freedom. Thus $$\begin{eqnarray} \mathbb{E}\left( \exp\left( W_s^2 \left( 1 + \frac{t-s}{2} \right) \right)\right) &=& \mathbb{E}\left( \exp\left( s X \left( 1 + \frac{t-s}{2} \right) \right)\right) = \mathcal{M}_X\left( s \left( 1 + \frac{t-s}{2} \right) \right) \\ &=& \left. \frac{1}{\sqrt{1-2u}} \right|_{u = s \left( 1 + \frac{t-s}{2} \right)} = \frac{1}{\sqrt{(s-1)^2 - s t}} \end{eqnarray} $$ Of course, the moment generating function is only defined if $(s-1)^2 > s t$.

share|cite|improve this answer
Wow. +1. $\quad$ – Stefan Hansen Jan 24 '13 at 17:33

$\mathbb E e^{(W_t - W_s)W_s + W_s^2} = \mathbb E(\mathbb E (\text{same}\mid \mathcal F_s) )= $$\mathbb E e^{(t- s)W_s^2 + W_s^2} $ and the square is of the same form.

share|cite|improve this answer
What do $same$ and $\mathcal{F}_s$ refer to? – Anna Liebermann Jan 24 '13 at 16:53
I think this might be on the right track, but it needs work. You don't need to condition on the whole history up to time $s$; it's enough to condition on $W_s$. – Michael Hardy Jan 24 '13 at 18:21
yes, its missing a factor of $\frac 12$, and an outside expectation , which I stuck in – mike Jan 24 '13 at 20:03

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.