3
$\begingroup$

in my probability class I was given this question on which I am stuck concerning a sum of random number of Poisson random variables:

Let us define the countable set of independent random variables $ X_i \sim \mathrm{Pois}(\lambda _i) $ and the random variable $N$ independent from the rest of the $ X_i $ which also has a Poisson distribution $ N \sim \mathrm{Pois}(\lambda) $. We are asked to check if the following sum $ Y = \sum_{i=1}^N X_i $ also has a Poisson distribution and if so, with what parameter? As a hint we are asked to look at the characteristic function of the variable to check.

I know the deterministic finite sum of Poisson random variables is again a Poisson random variable with the sum of parameters, but I cannot solve it for a random number of summands that is also Poisson-ly distributed, I know the characteristic function of a Poisson distributed random variable $ \Phi(t) = e^{\lambda (e^{it}-1)} $. I thank all helpers who can show me a way out of this.

$\endgroup$
3
$\begingroup$

In the case where all $\lambda_j$ are equal, let's say to $\mu$, you can compute

$$ \mathbb E\left[ \exp(itY) \mid N \right] = \exp(\mu (e^{it}-1) N)$$ so that $$ \mathbb E[\exp(itY)] = \mathbb E [\exp(\mu(e^{it}-1)N] = \exp(\lambda ( \exp(\mu (e^{it}-1))-1))$$ which is not the characteristic function of a Poisson distribution. However, things are more complicated if the $\lambda_j$ are not all equal.

EDIT: In general, let $s_n = \sum_{j=1}^n \mu_j$. In particular $s_0 = 0$. Then $$ \mathbb E \left[\exp(itY) \mid N \right] = \mathbb E \left[\exp(s_N (e^{it}-1))\right] = \exp(-\lambda) \sum_{n=0}^\infty \dfrac{\lambda^n}{n!} \exp(s_n (e^{it}-1)) $$ Letting $z = 1 - e^{it}$, we want to know if there is a formula $$ \exp(-\lambda) \sum_{n=0}^\infty \dfrac{\lambda^n}{n!} \exp(-s_n z) = \exp(-\rho z)$$ where $\rho > 0$. If this is valid for $z = 1 - e^{it}$, by analyticity it must be valid in the right half plane. Now as $z \to +\infty$, the limit of the left side is $\exp(-\lambda)$ (since $s_0 = 0$), while the limit of the right side is $0$. So this is impossible: the distribution is not Poisson.

$\endgroup$
  • $\begingroup$ Thank you Robert the case that all parameters are equal is now clear to me but is there a simple method to determine the distribution if the $ \lambda $s are not equal? Could it be a Poisson distribution $\endgroup$ – Don John Prep Mar 21 '16 at 20:04
  • $\begingroup$ I figured since N is independent from all other Xs, maybe this can be useful? $\endgroup$ – Don John Prep Mar 21 '16 at 20:05

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.