I have a process $S_{t}=S_{0}e^{\left(r-q\right)t+mt+X_{t}}$, where $X_t$ is a Levy process and I want to check for which $m$ the process $e^{-(r-q)t}S_t$ is a martingale. The third condition of a martingale states that for $s\leq t$ $$E(e^{-(r-q)t}S_t|F_s)=e^{-(r-q)s}S_s,$$ where $F_s$ is the filtration generated by the process $S_t$.

Many authors write that this process is a martingale when $E(e^{-(r-q)t}S_t)=S_0$ i.e. when $m=-\frac{1}{t}\ln\left(\phi_{X_{t}}\left(-i\right)\right)$, where $\phi_{X_t}$ is the characteristic function of $X_t$.

Why don't they condition on $F_s$ when they verify that the process is a martingale?

  • $\begingroup$ Yes $X_t$ is a levy process. Why they don't check if $E(e^{-(r-q)t}S_t|F_s)=e^{-(r-q)s}S_s$ but only write that if $m=-\frac{1}{t}\ln\left(\phi_{X_{t}}\left(-i\right)\right)$ then discounted process is a martingale $\endgroup$
    – Math122
    Nov 10 '20 at 18:57
  • $\begingroup$ Ok but technically they dont show that this is a martingale when $m=-\frac{1}{t}\ln\left(\phi_{X_{t}}\left(-i\right)\right)$, they only check special case when $s=0$. Is it possible to show that this process is a martingale for every $s$ with this $m$? Or for every $s$ we have to choose another $m$? $\endgroup$
    – Math122
    Nov 11 '20 at 11:05
  • $\begingroup$ @Kevin So in conclusion, there is no such $m\in\mathbb{R}$ that the discounted process is a martingale, but for any fixed moment in time $s$ in which we want to price options, we can find such $m$ and can we value option only for that one time instant? Why it works? How to find general form of $m$ for which this process is a martingale? $\endgroup$
    – Math122
    Nov 12 '20 at 12:11
  • $\begingroup$ @Kevin So if we price Euopean options we can check only condition $\mathbb{E}^\mathbb{Q}[S_t]=S_0e^{(r-q)t}$ i.e with $s=0$, but for American options we need to have a martingale. Am I right? $\endgroup$
    – Math122
    Nov 12 '20 at 18:45

An exponential Lévy process is typically modelled via $$ S_t = S_0\exp\left(\left(r-q+\omega\right)t+X_t\right),$$ where $X_t$ is a Lévy process with $X_0=0$. A Lévy process includes three model features: a linear drift, diffusive shocks and jumps (which may be large and rare or small and frequent). The number $\omega$ is called martingale correction or Jensen's correction and ensures the martingale property.

For our standard finance theory to work, the reinvested and discounted stock price, $S_te^{-(r-q)t}$, needs to be a martingale under $\mathbb{Q}$ (assuming constant interest rates and dividend yields). Let $(\mathcal{F}_t)$ denote the natural filtration of $X_t$. Then, for any $s\leq t$, \begin{align*} \mathbb{E}^\mathbb{Q}[S_t|\mathcal{F}_s] &= \mathbb{E}^\mathbb{Q}[S_0e^{(r-q+\omega)t+X_s+(X_t-X_s)}|\mathcal{F}_s] \\ &= S_0e^{(r-q+\omega)t} e^{X_s} \mathbb{E}^\mathbb{Q}[e^{X_t-X_s}] \\ &= S_s e^{(r-q+\omega)(t-s)} \mathbb{E}^\mathbb{Q}[e^{X_{t-s}}], \end{align*} where we used that $X_s$ is $\mathcal{F}_s$-measurable, and $X_t-X_s\overset{d}{=} X_{t-s}$ is independent of $\mathcal{F}_s$, see here.

Let $\varphi_{X_t}(u)=\mathbb{E}[e^{iuX_t}]$ be the characteristic function of the Lévy process $X_t$. The Lévy-Khintchine formula states that $\varphi_{X_t}(u)=e^{t\Psi(u)}$ which follows from the infinite divisibility of a Lévy process. The function $\Psi$ is called the characteristic exponent and captures the drift, diffusion and jump components of $X_t$.

Then, \begin{align*} \mathbb{E}^\mathbb{Q}[S_t|\mathcal{F}_s] &= S_s e^{(r-q+\omega)(t-s)} \varphi_{X_{t-s}}(-i) \\ &= S_s e^{(r-q+\omega)(t-s)} e^{(t-s)\Psi(-i)}. \end{align*} Hence, setting $\omega=-\Psi(-i)$ yields \begin{align*} \mathbb{E}^\mathbb{Q}[S_t|\mathcal{F}_s] &= S_s e^{(r-q)(t-s)}, \end{align*} which in turn implies that the discounted reinvested stock price is indeed a $\mathbb{Q}$-martingale.

Note that $$\omega=-\Psi(-i)=-\frac{1}{t}\ln\left(\varphi_{X_t}(-i)\right)$$ is independent of time. Thus, for an exponential Lévy process, the martingale property is ensured to hold if you verify that $\mathbb{E}^\mathbb{Q}[S_t]=S_0e^{(r-q)t}$.

  • $\begingroup$ Thanks! You lost $\ln$ in $\omega$ :) $\endgroup$
    – Math122
    Nov 15 '20 at 15:04
  • $\begingroup$ I have one question: why can we write that $\mathbb{E}^\mathbb{Q}[e^{X_{t-s}}]= \varphi_{X_{t-s}}(-i)$? The domain of characteristic function is real, so why we can take value at point $-i$? $\endgroup$
    – Math122
    Mar 20 at 20:57
  • 1
    $\begingroup$ @Math122 Characteristic functions can actually be evaluated in an open strip in the complex plane. So there exists a set $\mathcal{S}=\{z\in\mathbb{C}:a<\text{Im}(z)<b\}$ in which we can evaluate $\varphi$. For a ``sensible stock price model'', we have $a\leq-1$ and $b\geq0$. In these cases, $\varphi_{\ln(S_T)}(-i)=\mathbb{E}[S_T]$ is fine. More details are in Lewis' (2001) seminal paper and Schmelzle's (2010) survey. $\endgroup$
    – Kevin
    Mar 21 at 10:24

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.