I am taking a quant finance course, and am confused about my professor's methodology about the martingale method for pricing a standard European call, with payoff:

$c(T) = max[S(T) - K, 0]$,

which we write in terms of the martingale, $\hat{S}$ as:

$c(T) = exp(rT) max[\hat{S}(T) - exp(-rT)K,0]$

Then, he writes that:

$c(0) = \mathbb{E}_{\hat{\mathbb{P}}} \left[max [\hat{S}(T) - exp(-rT)K,0]\right]$

We can then write:

$c(0) = \int \left\{S(0) \exp[-1/2\sigma^2 T + \sigma x] - exp(-rT)K\right\} \times \frac{1}{\sqrt{T}\sqrt{2\pi}} \exp \left[-1/2 (x/\sqrt{T})^2\right]dx$.

I understand everything except the multiplication by the Brownian density: $\exp \left[-1/2 (x/\sqrt{T})^2\right]$. The reason is I thought this expectation is taken with respect to $\hat{\mathbb{P}}$ and not $\mathbb{P}$, where it is my understanding/confusion that this new probability density is obtained via the Radon-Nikodym derivative:

$\frac{d\hat{P}}{dP} = \exp\left[-\left(\frac{\mu -r}{\sigma}\right)W(t) - \frac{1}{2}\left(\frac{\mu -r}{\sigma}\right)^2 t\right]$ , so shouldn't the expectation have this extra term in it? I.e.,

$\hat{\mathbb{P}} = \mathbb{P} \times \frac{d\hat{P}}{dP}$?



The change of measure was done beforehand, $\hat{S}$ is a martingale under the $\hat{\mathbb{P}}$-measure, hence driftless. In the standard context that your professor adopted, $S$ is a log-normal process.

Therefore, one can write $$\hat{S(T)}=\hat{S(0)}e^{-1/2\sigma^2 T + \sigma W(T)}$$ where $W$ is a $\hat{\mathbb{P}}$-standard Brownian motion.

By replacing this formula in your professor's, we have

$$c(0) = \mathbb{E}_{\hat{\mathbb{P}}} \left[max [\hat{S(0)}e^{-1/2\sigma^2 T + \sigma W(T)}- exp(-rT)K,0]\right]$$

The only source of randomness in your formula is $W(T)$( centered,normally distributed with variance $T$) and it happens that we know its density function which is $$f_{W(T)}(x)=\frac{1}{\sqrt{T}\sqrt{2\pi}} \exp \left[-1/2 (x/\sqrt{T})^2\right]$$ Finally,

$$c(0) = \mathbb{E}_{\hat{\mathbb{P}}} \left[max [\hat{S(0)}e^{-1/2\sigma^2 T + \sigma W(T)}- exp(-rT)K,0]\right]=\int_{\mathbb{R}}{max [\hat{S(0)}e^{-1/2\sigma^2 T + \sigma x}- exp(-rT)K,0]f_{W(T)}(x)dx}$$

which is the result you are looking for ( be careful, you forgot the max function in your integral)

  • $\begingroup$ Hi. That makes sense, thanks. So, in general, if your only source of randomness is W(T), no matter what the Radon-Nikodym derivative is, you can always just integrate with $f_{W(T)}$? $\endgroup$ – Thomas Moore Sep 6 '17 at 22:22
  • 1
    $\begingroup$ In the answer, we are only working under one measure, so we could do that. However, if $W$ was a $\hat{\mathbb{Q}}$ Brownian motion, either you use your Radon-Nikodym derivative to change the expectation measure to $\hat{\mathbb{P}}$, or you convert $W$ into a $\hat{\mathbb{P}}$-measurable process. For example, we could have defined $S$ as $$dS_t=S_t(\mu dt+\sigma dW_t^{h})$$ where $W^{h}$ is a Brownian process under a measure $\mathbb{H}$, in that case we have to change the measure so it is compatible with the measure of the expectation, which is $\hat{\mathbb{P}}$ here. $\endgroup$ – Canardini Sep 7 '17 at 12:58

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.