I have a process: $$dr_t = (W_t^1 - ar_t)dt +\sigma dW_t^2$$ where $W_t^1$ and $W_t^2$ are brownian motions with instantaneous correlation coefficient $\rho$. I want to show that the solution of this process can be written in a form: $$r_t = r_0 e^{-at}+ \int_0^tk(t,s)dW_s^1 + \sigma\int_0^th(t,s)dW_s^2$$. I started with the substitution $\tilde{r}_t = e^{at}r_t$ which led me to the solution. $$r_t = r_0 e^{-at}+ \int_0^t e^{a(s-t)}W_s^1ds + \sigma\int_0^te^{a(s-t)}dW_s^2 $$ which differs from the asked solution in the first integrand ($ds$ instead of $dW_s^1$). I thought I could use Fubini theorem but I am not quite sure how. So essentially what I would like to know is how to show that these two solutions are equivalent. Thanks for any suggestions.

  • 3
    $\begingroup$ I would try the following: Consider $h(t) W_t$ where $h$ is a function of time only. Then using Ito, this can be written in integral form as $\int_0^t h'(s) W_s \mathrm{d}s + \int_0^t h(s) \mathrm{d}W_s$. So $h'(s) = e^{a (s - t)}$ in your case. Then $h(s) = e^{a(s - t)} / a$ and $\int_0^t e^{a (s - t)} W_s \mathrm{d}s = \int_0^t \left( 1 - e^{a (s - t)} / a \right) \mathrm{d}W_s$. Warning: typed on my phone. $\endgroup$ Mar 11, 2018 at 19:32
  • 1
    $\begingroup$ You lost a $d$ in front of $r_t$ in your first equation. @LocalVolatility's has given you the right answer, so I am not going to repeat it. $\endgroup$
    – Hans
    Mar 11, 2018 at 20:11
  • $\begingroup$ @LocalVolatility: Would you like to check out my question regarding SABR model solution derivation? quant.stackexchange.com/q/38719/6686 $\endgroup$
    – Hans
    Mar 11, 2018 at 20:12
  • $\begingroup$ excellent solution LocalVolatility..cheers $\endgroup$ Mar 11, 2018 at 23:20
  • 1
    $\begingroup$ @LocalVolatility you could write an answer out of your comment if you have time. $\endgroup$ Mar 12, 2018 at 12:17

2 Answers 2


This is a slightly extended and corrected (thanks @DaneelOlivaw) version of my comment.

Consider a process $h(t) W_t$, where $h$ is a function of time only. Using the Ito product rule, this can be expressed in integral form as

\begin{equation} h(t) W_t = \int_0^t h'(s) W_s \mathrm{d}s + \int_0^t h(s) \mathrm{d}W_s, \end{equation}

where we used that $h(0) W_0 = 0$. Matching terms we find that

\begin{equation} h'(s) = e^{a (s - t)} \end{equation}

and thus

\begin{equation} h(s) = \frac{1}{a} e^{a (s - t)}. \end{equation}

Rearranging yields

\begin{eqnarray} \int_0^t e^{a (s - t)} W_s \mathrm{d}s & = & \frac{1}{a} W_t - \frac{1}{a} \int_0^t e^{a (s - t)} \mathrm{d}W_s\\ & = & \frac{1}{a} \int_0^t \left( 1 - e^{a (s - t)} \right) \mathrm{d}W_s. \end{eqnarray}


"I thought I could use Fubini theorem but I am not quite sure how."

Note that we can write:

$$ \begin{align} \int_0^t e^{a(s-t)}W^1_s\text{d}s & = \int_0^t e^{a(s-t)}\left(\int_0^s\text{d}W^1_u\right)\text{d}s \\[6pt] & = \int_0^t \int_0^se^{a(s-t)}\text{d}W^1_u\text{d}s \end{align}$$

In the Appendix of Heath et al. (1992) you can find a nice statement of the stochastic Fubini theorem, which they themselves take from Ikeda and Watanabe (1981).

Let $s \leq T$ with $T \in \mathbb{R}^+$, we then define the continuous function $\phi(s)=e^{a(s-t)}$ for $s \in [0;T]$ $-$ note that in our case the function only depends on one variable thus for $u \in [0;T]$ we have $\phi(s,u)\triangleq\phi(s)$ which makes things easier. The function $\phi(s)$ is well-defined for any value of $t \in \mathbb{R}$. Therefore:

  1. The function $\phi(s)$ is deterministic thus mesurable w.r.t. the filtration generated by $W^1$;
  2. $\phi(s)$ is clearly square-integrable over $[0;T]$;
  3. For any $\tau \in [0;T]$: $\int_0^T (\int_0^{\tau}\phi(s)\text{d}W^1_u)\text{d}s=\int_0^T \phi(s)(\int_0^{\tau}\text{d}W^1_u)\text{d}s=W_{\tau}^1\int_0^T \phi(s)\text{d}s$ $=cW_{\tau}^1$ for some $c \in \mathbb{R}$, which is continuous in $\tau$ by property of Brownian Motion.

The hypothesis of Lemma 0.1 of Heath et al. are fulfilled (p.98), thus applying Corollary 2 (p.99):

$$\begin{align} \int_0^t e^{a(s-t)}W^1_s\text{d}s & = \int_0^t \int_0^se^{a(s-t)}\text{d}W^1_u\text{d}s \\[6pt] & = \int_0^t \left(\int_u^te^{a(s-t)}\text{d}s\right)\text{d}W^1_u \\[6pt] & = \int_0^t \left(\frac{1-e^{a(u-t)}}{a}\right)\text{d}W^1_u \end{align}$$

@LocalVolatility's comment gives an alternative derivation avoiding Fubini, although I believe the $1/a$ must be outside the parenthesis.

  • $\begingroup$ I am still unclear how to finish from here. Could you please explain how to justify switching the order of integration? Thanks $\endgroup$ Mar 12, 2018 at 20:55
  • $\begingroup$ @MichaelMark I have added some additional details. $\endgroup$ Mar 13, 2018 at 13:39

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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