Construct a martingale from a Poisson Process

1.2k Views Asked by At

Let $X_t = \sum_{i=1}^{N_t} Y_i$ and $N_t$ be a Poisson process with intensity $\lambda >0$. Suppose $Y_i$ are i.i.d. (independent of $N_t$) with normal distribution $N(m,\sigma^2)$. Determine $\mu>0$ such that $\exp(X_t-\mu t)$ is a martingale.

My attempt:

I compute: $E[e^{u X_t}] = e^{(\lambda t) \exp(u m + 0.5 u \sigma ^2)}$.

Then, $$\begin{align*} E[\exp{(X_t)} \mid F_s] &= E[\exp{(X_t - X_s + X_s)} \mid F_s] = E[\exp{(X_{t-s})}]e^{X_s} \\ &= e^{(\lambda (t-s)) \exp(u m + 0.5 u \sigma ^2)}e^{X_s}\end{align*}$$

So, $E[\exp{(X_t - \mu t)} \mid F_s] = e^{(\lambda (t-s)) \exp(u m + 0.5 u \sigma ^2)}e^{X_s}e^{-\mu t}$

So martingale condition holds if and only if

$e^{(\lambda (t-s)) \exp(u m + 0.5 u \sigma ^2)}e^{X_s}e^{-\mu t} = e^{X_s - \mu s}$

But I cannot find $\mu$ from this equation.

1

There are 1 best solutions below

1
On BEST ANSWER

Note that $u=1$ in your calculation (since you need $\mathbb{E}\exp(X_{t-s})$ in your calculation), i.e.

$$\mathbb{E}(\exp(X_t) \mid \mathcal{F}_s) = e^{\lambda(t-s) \cdot \exp(m+ \frac{1}{2} \sigma^2)} e^{X_s}$$

Therefore the martingale condition boils down to

$$e^{\lambda(t-s) \cdot \exp(m+ \frac{1}{2} \sigma^2)} e^{X_s} e^{-\mu t} = e^{X_s} e^{-\mu s},$$

i.e.

$$e^{\lambda(t-s) \cdot \exp(m+ \frac{1}{2} \sigma^2)} = e^{\mu(t-s)}$$

Taking logarithm yields

$$\mu = \lambda \exp \left( m+ \frac{1}{2} \sigma^2 \right).$$