Let's say I have two processes $X_t =X_0 \exp((a-\frac{1}{2}\sigma_X^2)t +\sigma_X dW_t^1)$ and $Y_t=Y_0 \exp((b-\frac{1}{2}\sigma_Y^2)t +\sigma_Y dW_t^2)$ and I then multiply them together (like converting a foreign asset into domestic currency). I arrive at

$X_tY_t = X_0Y_0 \exp((a+b-\frac{1}{2}\sigma_X^2-\frac{1}{2}\sigma_Y^2)t +\sigma_X dW_t^1+\sigma_Y dW_t^2)$

If I then assume the $W_t$ have a correlation of $\rho$, how would I obtain the expectation of $X_tY_t$. The thing that is troubling me is because I have 2 Brownian motions and I'm not sure how to compute from there.

Edit: I would like to make this question a little bit more general. What would I do if I wanted to compute $E[X_t Y_s]$, where say $s < t$. So something to do with independent increments would play a factor.


1 Answer 1


Just use the fact that $$ \sigma_X W_t^1 + \sigma_Y W_t^2 = \sqrt{ \sigma_X^2 + \sigma_Y^2 + 2\rho\sigma_X\sigma_Y } W_t $$ holds in probability assuming that $W_t^1$ and $W_t^2$ are 2 correlated Brownian motions with $$ d\langle W_t^1, W_t^2 \rangle_t = \rho dt $$ and $W_t$ is a new standard Brownian motion defined over the same probability space.

Simply put, just replace your sum of two correlated Gaussians (LHS above) by a single Gaussian (RHS above) exhibitting the exact same statistical properties (for a Gaussian identical mean/variance is enough). Doing so, you can now use the formulas you are accustomed to.

Applying this shows $$ X_t Y_t = X_0Y_0 \exp((a+b-\frac{1}{2}\sigma_X^2-\frac{1}{2}\sigma_Y^2)t +\sqrt{ \sigma_X^2 + \sigma_Y^2 + 2\rho\sigma_X\sigma_Y } W_t) $$ is lognormally distributed with mean $$ \mu = \ln(X_0Y_0)+(a+b-\frac{1}{2}\sigma_X^2-\frac{1}{2}\sigma_Y^2)t $$ and variance $$ \sigma^2 = (\sigma_X^2 + \sigma_Y^2 + 2\rho\sigma_X\sigma_Y)t $$

hence applying the usual formula for the mean of a lognormally distributed variable $$ E[X_t Y_t] = e^{\mu + \sigma^2/2} $$ is a function of $\rho$.


I think you are completely mixing up two different problems (here two different probability measures).

I understand from your comment that you define the dynamics of the FOR/DOM instantaneous exchange rate $X_t$ (i.e. $X_t = x$ meaning that, at time $t$, 1 unit of foreign currency = x units of domestic currency) under the foreign risk-neutral measure $\mathbb{Q}^f$ (or rather the probability space $(\Omega,\mathcal{F},\mathbb{Q}^f)$ along with the price process of an equity underlying denominated in the foreign currency.

In that case, under the domestic risk-neutral measure $\mathbb{Q}^d$ (or rather the probability space $(\Omega,\mathcal{F},\mathbb{Q}^d)$), one should have, by absence of arbitrage opportunities and assuming market completeness: $$ \frac{Y_t X_t}{B_t^d} \text{ is a } \mathbb{Q}^d \text{- martingale} $$

with $B^d_t$ representing the time-$t$ value of a risk-free money market account in the domestic economy in which 1 unit of currency has been invested at $t=0$. Using the martingale property it then entails that: $$ \frac{Y_0 X_0}{B_0^d} = Y_0 X_0 = E^{\mathbb{Q}^d} \left[ \frac{Y_t X_t}{B_t^d} \vert \mathcal{F}_0 \right] $$ and further assuming deterministic rates: $$ E ^{\mathbb{Q}^d} \left[ X_t Y_t \vert \mathcal{F}_0 \right] = e^{-r_d t} X_0 Y_0 $$ which is independent of $\rho$.

This is known as a compo adjustment in deriatives lingo.

That being said, under the measure $\mathbb{Q}^f$ where you have originally defined $X_t$ and $Y_t$, the expectation $E^{\mathbb{Q}^f} [ Y_t X_t ]$ does dependent on $\rho$ as hinted above.

More info on the quanto/compo change of measure technique here

  • $\begingroup$ Thanks for that, but I guess I might be having trouble applying this to my problem. My problem has $X_t =X_0\exp((r_d-r_f-\frac{1}{2}\sigma_X^2)t+\sigma_X W_t^1)$ and $Y_t =Y_0\exp((r_f-\frac{1}{2}\sigma_Y^2)t+\sigma_Y W_t^2)$. What would be the expectation of $X_tY_t$ because my answer involves $\rho$, when I know for certain that it should not. $\endgroup$
    – Jim
    May 3, 2016 at 11:36
  • $\begingroup$ @Jim, you are actually mixing up two different problems, please see my edits $\endgroup$
    – Quantuple
    May 3, 2016 at 12:52
  • $\begingroup$ The relationship you are trying to show is only true for a specific situation known as compo adjustment (has to do with quanto changes of measures), I would advise to change the title of your question to reflect that, for it is an interesting question, not asked in the right way. $\endgroup$
    – Quantuple
    May 3, 2016 at 12:54
  • $\begingroup$ Great explanation of my two minds. Would you be able to show how you arrive at the expectation for the compo adjustment (as you call it). Do I need to use Girsanov? $\endgroup$
    – Jim
    May 3, 2016 at 13:06
  • $\begingroup$ Do you agree that if you manage to show that $Y_t X_t / B_t^d$ is a $\mathbb{Q}^d$-martingale then you are done? $\endgroup$
    – Quantuple
    May 3, 2016 at 13:07

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.