I have a very fundamental question regarding simulation of DRIFTED geometric brownian motion.

We have the standard Blackos Scholes model:

$dS(t)=r S(t)dt+\sigma S(t) dW^{\mathbb{P}}(t)$, where $W^{\mathbb{P}}(t)$ is the standard Wiener process under probability measure $\mathbb{P}$.

If we want to simulate this, using constant $\Delta t$, we use the recursive formula:

$S_{t+1}=S_te^{(r-\frac{\sigma^2}{2})\Delta t+\sigma \sqrt{\Delta t} Z_t }$, where $Z_t \sim N(0,1)$.

Now assume that we want to change the drift such that:

$W^{\mathbb{Q}}(t) = W^{\mathbb{P}}(t) - \int_0^t \theta_s ds$ is a brownian motion under $\mathbb{Q}$ such that:

$dS(t)=(r + \sigma \theta )S(t)dt+\sigma S(t) dW^{\mathbb{Q}}(t)$

Now this is where I become unsure. If I want to simulate the drifted process, is it just fine to use the similar method as:

$S_{t+1}=S_te^{(r+\sigma \theta-\frac{\sigma^2}{2})\Delta t+\sigma \sqrt{\Delta t} Z_t }$, where $Z_t \sim N(0,1)$.

OR is it that I have to use $Z_t \sim N(\theta, 1)$?

Im not that strong in the change of measure part, so thats why I'm a bit unsure. Would appreciate for help.


  • $\begingroup$ Is there a typo in your equations for $S_{t+1} = ...$? Don't you mean $S_t$ instead of $S_0$? If not, then you won't simulate GBM paths even if you use the right drift. $\endgroup$ – Quantuple May 30 '16 at 9:29
  • $\begingroup$ Then the answer provided by Mark Joshi is spot on ;) $\endgroup$ – Quantuple May 30 '16 at 9:43
  • $\begingroup$ And the likelihood function is also only dependent on $Z_t \sim N(0,1)$ ? $\endgroup$ – Elekko May 30 '16 at 9:59
  • $\begingroup$ I don't know what you mean by that but yes, $Z_t \sim N(\theta,1)$ makes absolutely no sense $\endgroup$ – Quantuple May 30 '16 at 10:09
  • $\begingroup$ Ok, because I was reading this aueb.gr/pympe/hermis/hermis-volume-7/ACTUARIAL-RISK-MODELS/…, on equation (15), they define the $Z$ with non standard-normal densities. And then when they define the likelihood ratio (Radon-Nikodym derivative) I assumed that they used the "drifted mean" in the random $Z_t$. $\endgroup$ – Elekko May 30 '16 at 10:13

in the new measure, the stock has drift $r + \sigma \theta$ so yes you just proceed with that drift as you say. If $\theta$ is time dependent, it gets more complicated.

  • $\begingroup$ You mean $Z_t \sim N(0,1)$? How is the time dependance case then? $\endgroup$ – Elekko May 29 '16 at 23:37
  • $\begingroup$ you have to use the integral of theta against time instead of multiplying it, assuming it is deterministic. $\endgroup$ – Mark Joshi May 30 '16 at 4:37

Note that, under measure $Q$, the dynamics is of the form \begin{align*} dS_t = S_t \big[(r+ \sigma \theta_t) dt + \sigma dW_t^Q \big]. \end{align*} Then, for $\Delta>0$ sufficiently small, \begin{align*} S_{t+\Delta} &= S_te^{\left(r-\frac{1}{2}\sigma^2\right)\Delta + \sigma \int_t^{t+\Delta} \theta_s ds + \sigma \left(W_{t+\Delta}^Q-W_t^Q\right)}\\ &\approx S_te^{\left(r-\frac{1}{2}\sigma^2 + \sigma \theta_t\right)\Delta + \sigma \sqrt{\Delta} Z}, \end{align*} assuming that $\theta_t$ is continuous, where $Z\sim N(0, 1)$.


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.