6
$\begingroup$

Assume those $N$ Bernoulli random variables are i.i.d. with probability $p$, where $N \sim \operatorname{Poisson}(\lambda)$. The finite sum of them is Poisson distributed with $p\lambda$. I read a book claims that but without proof.

$\endgroup$
1
  • $\begingroup$ Alternatively to the MGF proof, you can calculate the distribution directly from the law of total probability: $P(Y=j)=\sum_{n=0}^\infty P(Y=j|N=n)P(N=n)$, where $Y=\sum_1^N X_i$, $N$ is Poisson $\lambda$ and $X_i$ are iidrv Bernoulli. All one must know is the conditional distribution—which should be an extremely familiar distribution of a discrete RV related to sums of Bernoulli variables... $\endgroup$ Oct 30, 2017 at 14:53

2 Answers 2

5
$\begingroup$

Here's a pedestrian way: Let $j\in \mathbb N$, $$\begin{align} P\left( \sum_{k=1}^N X_k = j \right) &= \sum_{l=j}^\infty P\left((N=l)\;\cap \left(\sum_{k=1}^l X_k = j\right)\right)\\ &=\sum_{l=j}^\infty P\left(N=l\right)P\left(\sum_{k=1}^l X_k = j\right) \quad \text{by independence of $N$ and $X_k$}\\ &=\sum_{l=j}^\infty e^{-\lambda}\frac{\lambda^l}{l!} \binom lj p^j(1-p)^{l-j}\\ &=\frac{e^{-\lambda}p^j\lambda^j}{j!} \sum_{l=j}^\infty \frac{\lambda^{l-j}(1-p)^{l-j}}{(l-j)!}\\ &= \frac{e^{-\lambda p}(\lambda p)^j}{j!} \end{align}$$

Hence $\sum_{k=1}^N X_k$ follows a Poisson distribution with parameter $\lambda p$.

$\endgroup$
4
  • $\begingroup$ Could you please provide explanation to the last rewrite of equation? Where did the summation go? And how did u get p to the exponent of lambda? $\endgroup$
    – Smarty77
    Apr 28, 2018 at 11:29
  • 1
    $\begingroup$ @Smarty77 $$\sum_{l=j}^\infty \frac{\lambda^{l-j}(1-p)^{l-j}}{(l-j)!} = \sum_{l=0}^\infty \frac{\lambda^l(1-p)^l}{l!} = e^{\lambda (1-p)} $$ $\endgroup$ Apr 28, 2018 at 11:37
  • $\begingroup$ ah, the Taylor Series! Thank you! And can we rewrite any j to 0 like that, because it is constant and it gets lost in infinite series, or is there some other reason? $\endgroup$
    – Smarty77
    Apr 28, 2018 at 11:47
  • 1
    $\begingroup$ @Smarty77 To be precise this is more related to the power series expansion of $\exp$: $e^z = \sum_{k=0}^\infty \frac{z^k}{k!}$. I just reindexed the sum. If you wanted to be formal you could write $$\sum_{l=j}^\infty \frac{\lambda^{l-j}(1-p)^{l-j}}{(l-j)!} = \lim_{N\to \infty} \sum_{l=j}^{N+j} \frac{\lambda^{l-j}(1-p)^{l-j}}{(l-j)!} = \lim_{N\to \infty} \sum_{l=0}^N \frac{\lambda^l(1-p)^l}{l!}=\sum_{l=0}^\infty \frac{\lambda^l(1-p)^l}{l!}$$ where the reindexing is done on finite sums. $\endgroup$ Apr 28, 2018 at 12:03
2
$\begingroup$

You can prove by showing that the moment generating function of $\sum_{i=1}^{N}{X_i}$ is that of a Poisson. \begin{align*} E\left[\exp\left(t \sum_{i=1}^{N}{X_i}\right)\right] &= E\left[E\left[\exp\left(t \sum_{i=1}^{N}{X_i}\right)\bigg|N\right]\right] \\ &= E\left[\prod_{i=1}^{N}E[\exp(t X_i)]\right] \\ &= E\left[(p\exp(t)+(1-p))^{N}\right] \\ &= \sum_{i=0}^{\infty} \frac{(p\exp(t)+(1-p))^{i} \exp(-\lambda)\lambda^{i}}{i!} \\ &= \exp(-\lambda) \sum_{i=0}^{\infty}\frac{(\lambda(p\exp(t)+(1-p)))^{i}}{i!} \\ &= \exp(\lambda(p\exp(t)+(1-p)-1)) \\ &= \exp(\lambda p(\exp(t)-1)) \end{align*} Note that $\exp(\lambda p(\exp(t)-1))$ is the moment generation function of a Poisson($\lambda p$).

$\endgroup$
1
  • 3
    $\begingroup$ This works only if you have a theorem that says a distribution with the same moment-generating function as a Poisson distribution has a Poisson distribution. $\endgroup$ Oct 30, 2017 at 16:15

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.