The Problem:

Suppose I have a simple jump model for an asset price

$$ dS = S(t-)[\mu dt + YdN(t)] $$

where $N(t)$ is a Poisson process and $Y_i$ are the jump sizes (assume independece of $N(t)$ and $Y_i$). Also for simplicity assume that $Y_i$ can only take two values, lets say $a$ and $b$. What constraints are there for $\mu$, $a$ and $b$?

I should also add that we can assume the existence of a money market account with constant short rate: $$ B(t) = B_0e^{rt} $$

My initial thoughts:

In order to preserve positivity of stock prices I need both $a$ and $b$ to be greater than $-1$. I also need to consider when there might be arbitrage in my model. However I can't work out what additional constraints I need to impose. Can anyone help?

Thanks in advance for any responses.


1 Answer 1


Note that, \begin{align*} d\big( e^{-\mu t}S_t \big) &= -\mu e^{-\mu t}S_t dt + e^{-\mu t}S_{t-}(\mu dt + Y_t dN_t)\\ &=e^{-\mu t}S_{t-}Y_t dN_t. \end{align*} From the Doleans-Dade exponential formula, \begin{align*} e^{-\mu t}S_t = \prod_{0<s\le t}(1+Y_s)\Delta N_s. \end{align*} That is, \begin{align*} S_t = e^{\mu t} \prod_{0<s\le t}(1+Y_s)\Delta N_s. \end{align*} Therefore, both $a$ and $b$ are greater than $-1$.

Let $\lambda$ be the intensity of the Poisson process $N$. Moreover, for ease of exposition, we assume that $E(Y_s)=L$, for $s\ge 0$, is a constant. Let \begin{align*} M_t = N_t-\lambda t \end{align*} and \begin{align*} X_t = \int_0^t Y_s dN_s -\lambda L t. \end{align*} Then, $\{M_t, t \ge 0 \}$ is a martingale. Moreover, for $u\ge v \ge 0$ and the information set $\mathscr{F}_v$ at time $v$, \begin{align*} E(X_u-X_v\mid \mathscr{F}_v) &= E\bigg(\int_v^u Y_s dN_s \mid \mathscr{F}_v\bigg) - \lambda L (u-v)\\ &= E\bigg(\int_v^u Y_s dM_s \mid \mathscr{F}_v\bigg) + \lambda E\bigg(\int_v^u Y_s ds \mid \mathscr{F}_v\bigg) - \lambda L (u-v)\\ &=0. \end{align*} That is, $\{X_t, t \ge 0\}$ is a martingale.

To rule out any arbitrage opportunity, $e^{-rt}S_t$ has to be a martingale. Note that \begin{align*} d\big( e^{-r t}S_t \big) &= -r e^{-r t}S_t dt + e^{-r t}S_{t-}(\mu dt + Y_t dN_t)\\ &= e^{-r t}S_{t-}\big((\mu-r+\lambda L)dt + dX_t \big). \end{align*} Then, $\mu = r - \lambda L$.


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.