I have to derive the Geometric Brownian motion (with not constant drift and volatility), and to find the mean and variance of the solution.

$\quad \left\{\begin{aligned} & d X_t = \mu(t) X_t d t + \sigma(t) X_t d W_t \\ & X_0 = \xi \end{aligned}\right.$

The solution can be obtained in a classical manner by Ito's Lemma:

$X_t = \xi e^{\int_0^t \left(\mu(s) - \frac{\sigma^2(s)}{2}\right) d s + \int_0^t \sigma(s) d W_s}$

And we can find the mean and variance:

$\mathbb{E}[X_t] = \xi e^{\int_0^t \left(\mu(s) - \frac{\sigma^2(s)}{2}\right) ds} \mathbb{E}\left[e^{\int_0^t \sigma(s) dW_s}\right]$

$Var(X_t) = \xi^2 e^{\int_0^t \left(2\mu(s) - \sigma^2(s)\right) d s} \left(\mathbb{E}\left[e^{2 \int_0^t \sigma(s) d W_s}\right] - \mathbb{E}\left[e^{\int_0^t \sigma(s) d W_s}\right]^2\right)\\$

These expression are not really simple, as they are when $\mu$ and $\sigma$ are constant.

Does someone know if we have a general expression for the expectation of the exponential of an Itô's Integral? (in this case of a deterministic function). I.e:

$\mathbb{E}\left[e^{\int_0^t \sigma(s) dW_s}\right]$



Given that $\sigma(s)$ is a deterministic function, then the process $(X(t))_{t \geq 0}$, where $$X(t) = \int_{0}^{t} \sigma(s) dW_s,$$ is a Gaussian process with zero mean and covariance function $\rho(s,t) = \displaystyle \int_{0}^{\min(s,t)} \sigma(u)^2 du$. A proof of this theorem can be found in Schreve's stochastic calculus for Finance II. Hence, $$\mathbb{E}\left[\exp\left(\int_{0}^{t} \sigma(s) dW_s\right)\right],$$ is the moment generating function of a Normally distributed random variable with zero mean and variance $\rho(t,t)$ evaluated in $1$.

  • $\begingroup$ Fast and precise, thanks! $\endgroup$ – Edouard Berthe May 11 '16 at 15:09

I would like to add something to the excellent answer of Siron: although it is good to know the Ito Integral of a deterministic function is Gaussian, we can solve my problem without that property.


$\quad g \colon [0,T] \longmapsto \mathbb{R}$

a deterministic function. Then if we define:

$\quad Y_t = \int_0^t g(s) \mathrm{d} W_s \iff \mathrm{d} Y_t = g(t) \mathrm{d}W_t\\ \quad f \colon (t, x) \mapsto e^x \\ \quad Z_t = f(t, Y_t) = e^{Y_t}$

Then by Ito's Lemma: $\def\d{\mathrm{d}}$ $\d Z_t = \left(0 + 0 \times Z_t + \frac{1}{2}g^2(t)Z_t\right)\d t + g(t)Z_t\d W_t = \frac{1}{2}g^2(t)Z_t\d t + g(t)Z_t\d W_t$


$\quad Z_t = Z_0 + \int_0^t \frac{1}{2}g^2(t)Z_t\d t + \int_0^t g(t)Z_t\d W_t$


$\def\E{\mathbb{E}}$ $\E[Z_t] = \E\left[Z_0 + \int_0^t \frac{1}{2}g^2(t)Z_t\d t\right] = \E[Z_0] + \int_0^t \frac{1}{2}g^2(s)\E[Z_t]\d t$

Because expectation of Ito Integral is null + Fubini.

So $h \colon t \mapsto \E[Z_t]$ checks: $\quad h'(t) = \frac{1}{2}g^2(t)h(t)$.

Therefore: $\quad h(t) = h(0)e^{\int_0^t \frac{1}{2}g^2(s)\d s} = e^{\int_0^t \frac{1}{2}g^2(s)\d s}$

This is what we want!


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.