Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

I want to study the following market:

$$S_1(t)=S_1(t)(\mu_1dt + \sigma_1dW_1(t))$$

$$S_2(t)=S_2(t)(\mu_2dt+\sigma_2dW_2(t))$$

for $t\in [0,T]$, constants $\mu_i,\sigma_i$, initial values $S_i(0)>0$, $i\in\{1,2\}$ and $W_1,W_2$ two Brownian Motion with the following bracket:

$$\langle W_1,W_2\rangle =\rho t$$ with $\rho\in\mathbb{R} $. We look at the following payoff:

$$(S_1(T)-S_2(T))^+$$

and I want to determine the price at $0$ of an option with the above payoff. It was hinted (on an old exercise from last year) to find a measure $Q$ under which the quotient $S_1/S_2$ is a martingale and calculate the price at $0$ of a European call option with strike $K=1$. Since no bank account was mentioned in the exercise text, I'm a little bit confused. Supposing there is a bank account $B(t)$, and assuming absence of arbitrage, we know there is a EMM $Q^1$ under which the discounted assets are martingales, i.e. $S_i(t)/B(t)$ are martingales. In this case I would define the density (of $Q$ w.r.t. to $Q^1$) on $\mathcal{F}_T$ as follows:

$$Z_T:=\frac{1}{B(T)}\frac{S_2(T)}{S_2(0)}$$

and $Z_t:=E[Z_T|\mathcal{G}_t]=\frac{1}{B(t)}\frac{S_2(t)}{S_2(0)}$. Then using Bayes formula I'm able to prove that $\frac{S_1}{S_2}$ is a $Q$-martingale. Are you think my observations are correct so far?

Now I want to compute $$E_Q[(\frac{S_1(t)}{S_2(t)}-1)^+|\mathcal{F}_0]$$

I wrote down:

$$E_Q[(\frac{S_1(t)}{S_2(t)}-1)^+|\mathcal{F}_0]=E_Q[(\frac{S_1(t)}{S_2(t)}-1)\mathbf1_{\{\frac{S_1(t)}{S_2(t)}>1\}}|\mathcal{F}_0]$$

Using Bayes gives

$$E_Q[(\frac{S_1(t)}{S_2(t)}-1)\mathbf1_{\{\frac{S_1(t)}{S_2(t)}>1\}}|\mathcal{F}_0]=\frac{1}{Z_0}(E_{Q^1}[\frac{S_1(t)}{S_2(t)}Z_t\mathbf1_{\{\frac{S_1(t)}{S_2(t)}>1\}}|\mathcal{F}_0]-E_{Q^1}[Z_t\mathbf1_{\{\frac{S_1(t)}{S_2(t)}>1\}}|\mathcal{F}_0])=\frac{1}{B(0)}(E_{Q^1}[\frac{S_1(t)}{B(t)}\mathbf1_{\{\frac{S_1(t)}{S_2(t)}>1\}}|\mathcal{F}_0]-\frac{1}{S_2(0)}E_{Q^1}[\frac{S_2(t)}{B(t)}\mathbf1_{\{\frac{S_1(t)}{S_2(t)}>1\}}|\mathcal{F}_0])$$

Under $Q^1$, $\frac{S_i}{B}$ is given by

$$S_i(t)=S_i(0)\exp{(\sigma_iW^*-\frac{1}{2}\sigma_i^2t)}$$

Where $W^*=W_t+\frac{\mu_i-r}{\sigma}t$ is $Q^1$ Brownian motion. Hence the above is under $Q^1$, $\exp{(Z)}$ distributed, where $Z$ is normal distributed with mean $-\frac{1}{2}\sigma_i^2t$ and Variance $\sigma^2t$. Is it true, that this conditional expectations reduce to expectations? If not, how would you calculate this?

However I am not at all sure about this derivation so far. Since the bracket of $W_1$ and $W_2$ is explicitly given, I thought somewhere I have to use it. It would be appreciated, if I did a error in reasoning, that someone could explain how to solve the problem, i.e. calculate the price. Thanks in advance

math

share|improve this question
    
1. looks fine $\text{ }{}$ 2. you can look this up under option to exchange or better of ${}\text{ }$3. there are issues (dividends, r) but basically you factor $S_2(t)$ out and all that factor does is change the drift of $W_2$ –  mike Nov 26 '12 at 14:08
add comment

1 Answer

Let $\beta(t) := e^{rt}$, where $r$ is the domestic risk-free rate. Let $Q$ be the risk-neutral measure.

The value is:

$v_t := E^Q[\frac{\beta(t)}{\beta{T}}S_1(T) 1_{\{S_1(T) > S_2(T)\}} | \mathscr{F}_t] - E^Q[\frac{\beta(t)}{\beta{T}}S_2(T) 1_{\{S_1(T) > S_2(T)\}} | \mathscr{F}_t]$

$=: I_1 + I_2$.

To evaluate $I_1$, do a change of probability measure to $S_1(T)$ as the numeraire asset:

$ E^{P_1}[S_1(t) 1_{\{S_1(T) > S_2(T)\}} | \mathscr{F}_t] = S_1(t)P_1\{S_1(T) > S_2(T) | S_1(t),S_2(t)\}$;

where the Radon-Nikodym derivative to go from $Q$ to $P_1$ was:

$\frac{dP_1}{dQ} = \frac{S_1(T)\beta(0)}{S_1(0)\beta(T)}$

meaning that, we can read off this Doleans exponential using Girsanov's theorem to get our new Wiener process:

$dW_{P_1}(t) = dW_1^Q - \sigma_1 dt$

$W_{P_1}(t) = W_1^Q - \sigma_1 t$

Now, $S_1(t)P_1\{S_1(T) > S_2(T) | S_1(t),S_2(t)\} = S_1(t)P_1\{\text{ln}\frac{S_1(T)}{S_2(T)} > 0\}$

by the monotone increasing property of $\text{ln}$. Under measure $P_1$ and $\mathscr{F}_t$,

$S_1(T) = S(t) e^{(r+\frac12 \sigma_1^2)(T-t) + \sigma_1 (W_1(T) - W_1(t))}$.

Under $Q$, we can decompose $S_2(T)$'s Wiener process ($W_2^Q$) into Wiener processes $W_1^Q(t)$ and $W_1^\perp(t)$, where $W_1^\perp(t)$ is a BM uncorrelated to $W_1^Q(t)$. Under $Q$:

$W_2^Q = \rho W_1^Q + \sqrt{1-\rho^2}W_1^\perp$

... You then use Levy's characterization to solve the rest of the problem.

share|improve this answer
    
thanks for your answer. Just one question: Why can we decompose $W^Q_2=\rho W^Q_1+\sqrt{1-\rho^2}W^\perp_1$ ? Is this always possible, i.e. for a general Brownian motion? –  math Dec 1 '12 at 9:24
    
@math Unfortunately this is a black box to me (I lifted it from lectures). All I know is that it very very easily allows one to compute the solution to options such as the Magrabe option (which is the one you're asking about). –  Jase Dec 1 '12 at 10:55
    
Thanks for your quick response. I will check my book about stochastic calculus. There is another point, which is not completely clear to me: Why is this equality true: $P_1\{S_1(T)>S_2(T)|S_1(t),S_2(t)\} = P_1\{\log(\frac{S_1(T)}{S_2(T)}>0\}$? –  math Dec 1 '12 at 10:59
    
@math I made it ambiguous by skipping a few steps and also dropping the conditional $P\{...|...\}$ notation. Here's what I did: (1) Divide both sides by $S_2(T)$ (doesn't change inequality, GBM is strictly positive), (2) log both sides; $\ln{1} = 0$ so the RHS just becomes 0. Inequality still holds because $ln$ is monotone increasing. –  Jase Dec 1 '12 at 11:22
    
Actually the thing which was important for me is: "and also dropping the conditional P". The whole rest is clear to me :) thx again! –  math Dec 1 '12 at 11:24
show 1 more comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.