4
$\begingroup$

Let $N$ be a Poisson random measure in $(0,\infty)^2$ with intensity $\eta$ given by $$\eta(ds, dx) = \mathbb{I}_{\{x>0\}} \dfrac{C}{x^{\alpha + 1}} ds \, dx.$$

  • Find the values $\alpha$ for which $\eta$ is a $\sigma$-finite measure.

  • Let $X_t = N(f_t)$ with $$f_t(s,x) = \mathbb{I}_{\{s\leq t\}}\cdot x.$$

    Find the values $\alpha$ for which $X_t<\infty$ for all $t\geq 0$ a-s.

  • Compute $\mathbb{E}\left[e^{-\lambda X_t}\right]$

  • Prove $X_t \overset{d}{=} t^{1/\alpha}X_1$.

Here is the problem, I have read a bit about stable Levy $\alpha$-processes and I can realize that the $\eta$ measure is related to a $\alpha$-stable process, which means that $\alpha$ must be between 0 and 2 for the $\eta$ measure to be $\sigma$-finite.

My problem with the first question is that I don't know how to prove that alpha must take those values mentioned. I understand that for a measure to be sigma-finite we must find a succession of countable sets with finite measure covering all $S = (0,\infty)^2$ and that is easy considering the intervals $A_n = \left(\dfrac{1}{n}, n\right)$. So my reasoning (which I think is not correct) is the following

$$\eta\left(A_n\times A_n\right) = \int_{1/n}^{n} \int_{1/n}^{n}\mathbb{I}_{\{x>0\}} \dfrac{C}{x^{\alpha + 1}} \, ds \, dx$$

but here we can conclude that this measure is finite for any value $\alpha \in \mathbb{R}$, which I believe is incorrect to assume from what I have already said about $\alpha -$stable processes.

For questions 2 and 3 I am using Campbell's theorem: in question two I had no problem finding that $\alpha \in (1,2]$ (this is assuming I had concluded that $\alpha \in (0,2]$ in question 1).

For question 3, again using the result of Campbell's theorem and assuming $\alpha \in (1,2]$, I have the following calculation:

\begin{align*} \mathbb{E}\left[e^{-\lambda X_t}\right] &= \exp\{-\int_S (1-e^{-\lambda f_t(s,x)})\ \eta\left(ds,dx\right)\}\\[0.2cm] &= \exp\{-\int_0^\infty \int_0^\infty \left(1-e^{-\lambda f_t(s,x)}\right) \mathbb{I}_{\{x>0\}} \dfrac{C}{x^{\alpha+1}}\ ds\,dx\}\\[0.2cm] &= \exp\{-\int_0^\infty \int_0^\infty \left(1-e^{-\lambda \mathbb{I}_{\{s\leq t\}}x}\right) \mathbb{I}_{\{x>0\}} \dfrac{C}{x^{\alpha+1}}\ ds\,dx\}\\[0.2cm] &= \exp\{-\int_0^\infty \int_0^\infty \left(1-e^{-\lambda \mathbb{I}_{\{s\leq t\}} x}\right) \dfrac{C}{x^{\alpha+1}}\ ds\,dx\}\\[0.2cm] &= \exp\{-\int_0^\infty \dfrac{C}{x^{\alpha+1}}\left[ \int_0^\infty \left(1-e^{-\lambda \mathbb{I}_{\{s\leq t\}}x}\right)\ ds\right]\,dx\}\\[0.2cm] &= \exp\{-\int_0^\infty \dfrac{C}{x^{\alpha+1}}\left[ \int_0^t \left(1-e^{-\lambda x}\right)\ ds\right]\,dx\}\\[0.2cm] &= \exp\{-\int_0^\infty \dfrac{C}{x^{\alpha+1}}\left[ \left(1-e^{-\lambda x}\right) \int_0^t \ ds\right]\,dx\}\\[0.2cm] &= \exp\{-\int_0^\infty \dfrac{C}{x^{\alpha+1}} \left(1-e^{-\lambda x}\right) \cdot t\ dx\}\\[0.2cm] &= \exp\{-C\cdot t \int_0^\infty \left(1-e^{-\lambda x} \right) \dfrac{1}{x^{\alpha+1}} \, dx\} \end{align*}

However, once that point is reached, I am stuck, since I don't know how to calculate that remaining integral, I am almost sure that the development up to that point is correct, but I would appreciate some comment in case something is wrong or some alternative how to conclude the calculation.

Finally, for question 4, I think it is enough to find the expectation of the Laplace transform of $t^{1/\alpha} X_1$, and see that it matches the previous calculation, but again I run into a problem when developing the calculation

\begin{align*} \mathbb{E}\left[e^{-\lambda t^{1/\alpha} X_1}\right] &= \exp\{-\int_S \left(1-e^{-\lambda t^{1/\alpha} f_1(s,x)}\right)\ \eta\left(ds,dx\right)\}\\[0.2cm] &= \exp\{-\int_{0}^{\infty} \int_{0}^{\infty} \left(1-e^{-\lambda t^{1/\alpha} \mathbb{I}_{\{s\leq 1\}}x}\right)\ \mathbb{I}_{\{x>0\}} \dfrac{C}{x^{\alpha + 1}}\ ds \, dx\}\\[0.2cm] &= \exp\{-\int_0^\infty \int_0^1 \left(1-e^{-\lambda t^{1/\alpha} x}\right)\ \dfrac{C}{x^{\alpha + 1}}\ ds \, dx\}\\[0.2cm] &= \exp\{-\int_0^\infty \dfrac{C}{x^{\alpha + 1}} \left[\int_0^1 \left(1-e^{-\lambda t^{1/\alpha} x}\right)\ \ ds\right] dx\} \end{align*}

Again, I would appreciate any comments or observations that may help me to proceed.

$\endgroup$

1 Answer 1

1
$\begingroup$

1. For $\alpha>0$, the integral, $$ \int_0^\infty(1-e^{-\lambda x}){dx\over x^{1+\alpha}} $$ converges if and only if $0<\alpha<1$, because the integrand is $\sim \lambda x^{-\alpha}$ for $x\to 0+$. This is precisely the range of $\alpha$s for which $X_t<\infty$ a.s.

For $\alpha\in(0,1)$, the change of variables $u=\lambda x$ results in $$ \int_0^\infty(1-e^{-\lambda x}){dx\over x^{1+\alpha}}=\lambda^\alpha\int_0^\infty(1-e^{-u}){du\over u^{1+\alpha}}. $$ Now write $1-e^{-u} =\int_0^u e^{-t} dt$ and use Fubini: $$ \eqalign{ \int_0^\infty(1-e^{-u}){du\over u^{1+\alpha}} &=\int_0^\infty\int_0^t e^{-t}dt {du\over u^{1+\alpha}} \cr &=\int_0^\infty\int_t^\infty {du\over u^{1+\alpha}} e^{-t} dt\cr &=\int_0^\infty t^{-\alpha}e^{-t} dt\cr &=\int_0^\infty t^{(1-\alpha)-1}e^{-t} dt\cr &=\Gamma(1-\alpha).\cr } $$

2. The measure $\eta$ is $\sigma$-finite for each $\alpha>0$. Indeed $$ \eta\left( (0,n)\times (1/n,n)\right)=\lambda n\cdot{C\over\alpha}(n^\alpha-n^{-\alpha})<\infty $$ for each positive integer $n$.

$\endgroup$
1
  • $\begingroup$ Thank you very much for your help! I was able to realize some mistakes in question 2 to find that $\alpha \in (0,1)$ and thanks to the development of the integral I was able to complete the whole exercise. $\endgroup$
    – DkRckr12
    May 30, 2021 at 23:32

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.