Finding the Levy measure

883 Views Asked by At

I am struggling with the derivation of the Lévy-measure of a Gamma-process $X_t$ with law $p_t(x)= \frac{\lambda^{ct}}{\Gamma(ct)}x^{ct-1}e^{-\lambda x}1_{\lbrace x>0 \rbrace }$. The paper I am reading says it is shown "easily" via the characteristic function and using the Levy-Khintchine formula:

$\Phi(s)=E[e^{isX_1}] = exp( ias - \frac{1}{2}\sigma^2s^2 + \int_{\mathbb{R}}(1-e^{isx} - is\mathbb{1}_{|x|<1})d\nu(x))$

I computed via the pdf of the Gamma-distribution $\Phi(s) = E[e^{isX_t}]= (1-\frac{is}{\lambda})^{-ct}$

But how is the Lévy-measure then derived to be $\nu(x) = \frac{ce^{-\lambda s}}{x}1_{\lbrace 1 > 0\rbrace}$?

2

There are 2 best solutions below

0
On

I know this answer comes quite late. A nice explanation is given in these great course notes.

You can show this result by using the Frullani integral, which under some conditions tells you that $$ \int_0^\infty \frac{f(ax)-f(bx)}{x}dx = (f(0)-f(\infty))\log\left(\frac{b}{a}\right), \quad b > a > 0. $$ Here, $f(\infty)$ of course represents the limit $\underset{x \rightarrow \infty}\lim f(x)$.

Applying this result on the function $e^{-x}$ for the values $a=\lambda, b=\lambda-z$ where $z<0$, we obtain \begin{align*} \int_0^\infty \frac{e^{-\lambda x}-e^{-(\lambda - z)x}}{x}dx &= \log\left(\frac{\lambda - z}{\lambda}\right), \end{align*} which can be rewritten as $$ \int_0^\infty (1-e^{z x})c\frac{e^{-\lambda x}}{x}dx = - \int_{-\infty}^\infty (e^{z x}-1)c\frac{e^{-\lambda x}}{x}1_{x>0}dx = -c \log\left(1-\frac{\lambda}{z}\right).$$ Using analytic continuation, you can show that this result holds for any $z$ such that $\text{Re}(z) \leq 0$ from which the result quite easily follows!

0
On

Hint: Perhaps first switch to the MGF (replacing i*s by s for sufficiently small s). Then take the logarithms of your candidate MGF (first Phi), and that of the Gamma process (second Phi). Differentiate the two expressions, knowing that you can replace integration and differentiation; then integral with respect to the Levy measure is very easy...Compare the two expressions and you are done.