Let's say that I have two correlated GBMs:

$$dA_t = A_t \sigma^A dW^A_t$$ $$dR_t = R_t \sigma^R dW^R_t$$ $$dW^R_t dW^A_t = \rho dt$$

I am trying to price a derivative which payoff at time $T$ is:

$$\text{Payoff}_T = (A_TR_T - A_T \lambda)^+ $$

My idea was to apply Margrabe's formula, but for this I need to formulate the two processes $X_t = A_t R_t$ and $Y_t = \lambda A_t$ as GBMs as well in order to find their respective volatilities and their correlation.

The first one is quite trivial:

$$dY_t = d(\lambda A_t) = \lambda dA_t + \frac{1}{2} 0 = \lambda A_t \sigma^A dW^A_t = dY_t \sigma^A dW^A_t$$

which is clearly a GBM and $\sigma_Y = \sigma^A$.

But I'm a struggling to express the second one, what I came up with so far is:

$$ \begin{align} d(A_t R_t) & = & A_t dR_t + R_t dA_t + dA_tdR_t \\ & = & A_t R_t \sigma^R dW^R_t + A_t R_t \sigma^A dW^A_t + A_t R_t \sigma^R \sigma^A \underbrace{dW^A_t dW^R_t}_{\rho dt} \\ & = & A_t R_t \left[ \sigma^R dW^R_t + \sigma^A dW^A_t + \sigma^R \sigma^A \rho dt \right] \end{align}$$

But this is where I'm stuck, I can't figure out how to express this as a "simple" GBM as it's quite clearly multivariate... Am I missing something?

Is there a way I can still use the Margrabe formula to price my option?

  • 1
    $\begingroup$ As $(A_TR_T - A_T \lambda)^+ = A_T(R_T - \lambda)^+$, you can use $A_T$ as the numeraire, and then price using Black's formula. $\endgroup$ – Gordon Oct 20 '15 at 13:39
  • 1
    $\begingroup$ I am not experience with approaches using numeraires. Does it mean that one prices $(R_T-\lambda)^+$ and how precisely can I use the numeraire? This really sound interesting! $\endgroup$ – Ric Oct 21 '15 at 14:50
  • 1
    $\begingroup$ @Richard agreed I'll ask another question about this. $\endgroup$ – SRKX Oct 22 '15 at 1:02

Coming back to the line where you are stuck. If we define $$ Z_t = \sigma^A/\bar{\sigma} W_t^A + \sigma^R/\bar{\sigma} W_t^R, $$ with $\bar{\sigma}^2 = (\sigma^A)^2 + 2 \sigma^A \sigma^R \rho + (\sigma^R)^2$, then $Z_t$ is a Brownian motion in its own filtration and the first and second moment are correct.

Then we write your last line using $X_t = A_t R_t$ as $$ dX_t = X_t (\bar{\sigma} dZ_t + \sigma^A\sigma^R \rho dt), $$ with the solution $$ X_t = X_0 \exp((\sigma^A\sigma^R \rho - \bar{\sigma}^2/2) t + \bar{\sigma} Z_t), $$ which is a GBM with (!) drift.

In Magrabe's formula you need the covariance (vol times vol times correlation) of the two diffusion terms: $$ \begin{eqnarray} Cov(\bar{\sigma} Z_t, \lambda \sigma^A W^A_t) = Cov(\sigma^A W_t^A + \sigma^R W_t^R, \lambda \sigma^A W^A_t) &= \\ Cov(\sigma^A W_t^A,\lambda \sigma^A W^A_t) + Cov(\sigma^R W_t^R,\lambda \sigma^A W^A_t) &= \\ \sigma^A \lambda Cov(W_t^A,W_t^A) + \sigma^R \lambda \sigma^A Cov(W_t^R,W_t^A) &= \\ \sigma^A \lambda t + \sigma^R \lambda \sigma^A \rho t.& \end{eqnarray} $$

  • $\begingroup$ And then $dZ_t dW^A_t = ( \frac{\sigma^A }{\bar{\sigma}} + \frac{ \sigma^R }{\bar{\sigma}} \rho ) dt$? So the correlation to be used withing Margrabe's formula is $\frac{\sigma^A }{\bar{\sigma}} + \frac{ \sigma^R }{\bar{\sigma}} \rho$? $\endgroup$ – SRKX Oct 20 '15 at 16:25
  • $\begingroup$ @SRKX: That is correct. $\endgroup$ – Gordon Oct 20 '15 at 20:47
  • $\begingroup$ See my thoughts above about this. $\endgroup$ – Ric Oct 21 '15 at 8:28

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.