Show that for any $\lambda \in \Re$, the process $Y_{\lambda,t}$ defined as:

$$Y_{\lambda,t} = (S_t/S_0)^\lambda e^{-(r\lambda-\lambda(1-\lambda)\sigma^2/2)t}$$

is a martingale under the risk neutral measure $Q$.

I was thinking that I could apply Ito's Lemma, in which to show that the $\text{d}t$ term will be zero. However, after doing the partial derivatives, the terms do not cancel each other out.

Would really appreciate all the help I can get!

  • 1
    $\begingroup$ What dynamics does $S_t$ follow? What is $x$ and $\lambda$? Your exponential is unclear too. $\endgroup$ Nov 8 '19 at 12:25
  • $\begingroup$ I have amended your equation to what I believe is the proper expression, and provided an answer based on that. Can you confirm that my correction is correct? $\endgroup$ Nov 8 '19 at 13:30
  • $\begingroup$ @DaneelOlivaw Sorry, I made a mistake! It's supposed to be 2 and not $x$! $S_t$ follows the dynamics of the Black & Scholes model! $\endgroup$
    – Math user
    Nov 8 '19 at 14:01
  • $\begingroup$ Ok, that is what I was wondering. You can check my answer below, I assume $S_t$ follows lognormal dynamics (i.e. as in the Black-Scholes model) with drift $r$ and diffusion/volatility $\sigma$ then? $\endgroup$ Nov 8 '19 at 14:20
  • $\begingroup$ Find the partial derivative with respect to t & S and plug it into Ito's lemma. Then use the fact that $S_t$ follows the black scholes model $dS_t=S_t(rdt+\sigma dW_t)$ and plug it back into $dS_t and (dS_t)^2$ should help in cancelling out the dt terms and you'll be left with only the $dW_t$ terms. $\endgroup$
    – Anon
    Nov 9 '19 at 0:19

We define the process $Y_t=Y(t,S_t)$ as follows: $$Y_t=\left(\frac{S_t}{S_0}\right)^\lambda \exp\left\{-\left(r\lambda-\lambda(1-\lambda)\frac{\sigma^2}{2}\right)t\right\}$$

Method 1

Let: $$\alpha=\lambda\left(r-(1-\lambda)\frac{\sigma^2}{2}\right)$$ Then by Itô's Lemma: $$\text{d}Y_t=-\alpha Y_t\text{d}t+\frac{\lambda}{S_t}Y_t\text{d}S_t+\frac{1}{2}\frac{\lambda(\lambda-1)}{S_t^2}Y_t\text{d}\langle S_t,S_t\rangle$$ Assuming $S_t$ follows a Geometric Brownian Motion with drift $\mu$ and diffusion $\beta$: $$\text{d}S_t=\mu S_t\text{d}t+\beta S_t\text{d}W_t$$ Then: $$\text{d}Y_t=\left(\lambda\mu+\lambda(\lambda-1)\frac{\beta^2}{2}-\alpha\right)Y_t\text{d}t+\lambda\beta Y_t\text{d}W_t$$ Hence for $Y_t$ to be a (local) martingale we need: $$\lambda\mu+\lambda(\lambda-1)\frac{\beta^2}{2}=r\lambda+\lambda(\lambda-1)\frac{\sigma^2}{2}$$ This is true if: $$\begin{align} \text{C.1}\quad\mu&=r\\ \text{C.2}\quad\beta&=\sigma \end{align}$$

Method 2

Note also that, if the above conditions $\text{C.1}$ and $\text{C.2}$ hold: $$\begin{align} Y_t&=\left(\frac{S_t}{S_0}\right)^\lambda \exp\left\{-\left(r\lambda-\lambda(1-\lambda)\frac{\sigma^2}{2}\right)t\right\} \\ &=\exp\left\{\left(r\lambda-\lambda\frac{\sigma^2}{2}\right)t+\lambda\sigma W_t\right\}\exp\left\{-\left(r\lambda-\lambda(1-\lambda)\frac{\sigma^2}{2}\right)t\right\} \\ &=\exp\left\{-\lambda^2\frac{\sigma^2}{2}t+\lambda\sigma W_t\right\} \end{align}$$ Let $0<s<t$. Then: $$\begin{align} \mathbb E^Q\left[Y(t,S_t)|\mathcal{F}_s\right]&= \exp\left\{-\lambda^2\frac{\sigma^2}{2}s+\lambda\sigma W_s\right\}\mathbb E^Q\left[\exp\left\{-\lambda^2\frac{\sigma^2}{2}(t-s)+\lambda\sigma (W_t-W_s)\right\}|\mathcal F_s\right] \\ &= \exp\left\{-\lambda^2\frac{\sigma^2}{2}s+\lambda\sigma W_s\right\}\mathbb E^Q\left[\exp\left\{-\lambda^2\frac{\sigma^2}{2}(t-s)+\lambda\sigma (W_t-W_s)\right\}\right] \\ &= \exp\left\{-\lambda^2\frac{\sigma^2}{2}s+\lambda\sigma W_s\right\} \\[7pt] &= Y_s \end{align}$$

  • $\begingroup$ Do we assume the expected value under Q of the line just before the last one is 0 because expected value of $W_t - W_s$ is 0 ? If it the case, how does it cancel the all exponential ? $\endgroup$
    – TmSmth
    Nov 9 '19 at 11:22
  • 1
    $\begingroup$ @TmSmth $W_t-W_s$ is a normally-distributed random variable with $0$ expectation and variance $\lambda^2\sigma^2(t-s)$ thus its Laplace transform is $\exp\{(\lambda^2\sigma^2(t-s))/2\}$, hence terms cancel and we get $e^0=1$. $\endgroup$ Nov 19 '19 at 18:28

$E_0[Y_{\lambda,t}] = 1\,\, \forall t$, hence $Y_t$ is a martingale.

Hint: Look at the arithmetic moments section of this wiki page on lognormal distribution


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.