Statement of problem

From Oksendal SDEs question 5.18:

The geometric mean reversion process is a solution to:

$$ dX_t = k (a - \log X_t) X_t dt + \sigma X_t dB_t $$

In showing that solution is given by:

$$ X_t = \exp ( e^{-k t}\log X_0+(a-\sigma^2/2k)(1-e^{k t})+\sigma e^{-k t}\int_0^t e^{k s}dB_s)$$

It says to take $Y_t = \log X_t$


$$ Y_t= e^{-k t}\log X_0+(a-\sigma^2/2k)(1-e^{k t})+\sigma e^{-k t}\int_0^t e^{k s}dB_s$$

and go from there..


I have seen elsewhere that the next step should lead to:

$$dY_t = \left( -kY_t+k\left(a-\frac{\sigma^2}{2k}\right)\right)dt+\sigma dB_s$$

But I am unsure about $Y_t = f(t,B_t)$, and when using Itô's formula to evaluate $$dY_t = \partial_t f(t,B_t)dt + \partial_x f(t,B_t)dB_t + \frac{1}{2}\partial_{xx}f(t,B_t) dt$$

I do not know how to determine the function $f(t,x)$ for use in the formula due to the stochastic integral contained in $Y_t$.

After saz post

$$f(t,x)=\sigma e^{-k t}x + \left(a -\frac{\sigma ^2}{2k }\right)\left(1-e^{k t}\right)$$

For the formula you need the following partial derivatives:

$$ \partial_t f(t,x)=-e^{-k t} k x \sigma -e^{t k } k \left(a -\frac{\sigma ^2}{2 k }\right) $$

$$ \partial_x f(t,x) = e^{-k t} \sigma $$

Then: $$ Z_t = \left(\frac{1}{\sigma }Y_0+\int_0^t e^{k s} \, dB_s\right) $$

$$Y_t = f(t,Z_t)$$


$$dY_t =\left[ -e^{-k t} k \left(\frac{1}{\sigma }Y_0+\int_0^t e^{k s} \, dB_s\right) \sigma -e^{t k } k \left(a -\frac{\sigma ^2}{2 k }\right) \right] dt + e^{-kt} \sigma dZ_t $$

$$ dY_t = \left[ k\left(a-\frac{\sigma^2}{2k}\right) -kY_t \right] dt + e^{-kt} \sigma dZ_t $$

$$ dY_t = \left[ k(a-Y_t)-\frac{\sigma ^2}{2}\right]\text{dt} + \sigma \text{dB}_t $$


$$ dY_t = \left[ k(a-Y_t) - \frac{\sigma^2}{2}\right]dt + \sigma dB_t $$

Then by Ito's with $X_t = e^{Y_t}$:

$$ dX_t = k(a-\log X_t) X_t dt + X_t \sigma dB_t \,\,\,\, \blacksquare$$


No, you cannot simply set

$$g(t,x) = \int_0^t e^{ks} \, dx$$

since the stochastic integral $\int_0^t e^{ks} \, dB_s$ is not pathwise defined. In order to find the differential of $(Y_t)_{t \geq 0}$, you have to apply Itô's formula for Itô processes. Namely, if

$$dZ_t = \sigma_t \, dB_t$$

then Itô's formula states that

$$\begin{align*} f(t,Z_t)-f(0,Z_0) &= \int_0^t \partial_x f(s,Z_s) \sigma_s \, dB_s + \int_0^t \left( \frac{1}{2} \partial_{xx} f(s,Z_s) \sigma_s^2 + \partial_t f(s,Z_s) \right) \, ds. \end{align*}$$

Apply this to

$$\begin{align*} Z_t &:= \left(\frac{1}{\sigma} Y_0 + \int_0^t e^{ks} \, dB_s \right) \\ f(t,x) &:= \sigma e^{-kt} x + \left(a- \frac{\sigma^2}{2k} \right) (1-e^{kt}). \end{align*}$$

  • $\begingroup$ I assumed you mean $Y_0$ in $Z_t$, is this correct? I have got slightly further in my attempt, I think. $\endgroup$ – shilov Sep 21 '14 at 14:06
  • $\begingroup$ @shilov Yes, you are right. I didn't check all your calculations, but it looks pretty good. $\endgroup$ – saz Sep 21 '14 at 14:38

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.