Ito's lemma: $$\text{if } dX_t = \sigma_tdW_t+\mu_t dt, \text{ where $W_t$ is a Wiener process, then } df(X_t)=\sigma_tf'(X_t)dW_t + (\mu_tf'(X_t)+ \frac{1}{2}\sigma_t^2f''(X_t))dt$$

now I want to apply this to $$Y_t=e^{\sigma W_t+\mu t}$$

Where $W$ is a Wiener process.

There are two ways I can think of to apply Ito's lemma to $Y_t$ but they give different results:

(1). We can of course set $X_t = \sigma W_t + \mu t$, and set $f(x)=e^x$. Then Ito's Lemma gives $df= f(X_t)(\sigma dW_t+(\mu + \frac{1}{2} \sigma ^2 )dt)$

(2). However, I am wondering why I get a different result if instead, I set $X_t=W_t$, and $f(x)=e^{\sigma x + \mu t}$.

Then Ito's lemma gives: $df=f(X_t)(\sigma dW_t + \frac{1}{2}\sigma^2dt)$.

Which of these applications is correct, and more importantly, why is the other incorrect?

  • $\begingroup$ Because in the second case it's $f(t,x)$, not $f(x)$. $\endgroup$ – zhoraster Mar 29 '17 at 14:12

The first one is correct.

In the second one, $f$ depends explicitly on $t$. This is not a problem per se, but it means that the correct form of Ito's lemma should include an additional $\frac{\partial f}{\partial t} dt$ term. This is an advantage of the otherwise misleading notation $\frac{dX_t}{dt}=\mu_t + \sigma_t \frac{dW_t}{dt}$: it reminds us what we are differentiating with respect to, so that we know not to forget about any explicit dependence on that variable.

To see the point, consider a more extreme example: $f(t,x)=t$. If I then define $Y_t=f(t,W_t)$, $dY_t$ can't possibly be $0$. Yet this is what your form of Ito's lemma without the $\frac{\partial f}{\partial t}$ term would imply.

  • $\begingroup$ you mean with respect to $t$? Are we differentiating with respect to $t$? I find this very confusing. If we are really differentiating with respect to $t$ then why don't we just write it in the form you just wrote? Why is this notation "otherwise misleading" as you say? $\endgroup$ – user56834 Mar 29 '17 at 14:22
  • $\begingroup$ @Programmer2134 Roughly speaking, we are differentiating with respect to $t$, yes. Put another way, you might think of $df$ as $\frac{df}{dt} dt$. The notation I suggested is only misleading because $\frac{dW_t}{dt}$ does not exist in the classical sense. $\endgroup$ – Ian Mar 29 '17 at 14:27

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.