2
$\begingroup$

If I am given $N$ is from a binomial distribution with the parameter $n,p$ and also $X$ is from exponential distribution with the parameter $\lambda$.

Assume that $S_N=X_1+X_2+...+X_N$.

How can I find the distribution of $S$?

$\endgroup$
2
  • 1
    $\begingroup$ The right-hand side reminds me that Sum of the exponential random variables is gamma random variables. $\endgroup$ – Daman Nov 17 '20 at 7:48
  • 1
    $\begingroup$ But I never learn the compound distribution of (Bin and exp). Just wondering how can I find the distribution of it. $\endgroup$ – Mathxx Nov 17 '20 at 8:06
5
$\begingroup$

Since $S_N$ is the sum of $N$ many i.i.d. $\exp(\lambda)$ random variables, therefore $S_N \sim \Gamma(N , \lambda)$ .

Now, $N \sim \text{Bin}(n , p)$. Therefore, $S_N~|~N = k~ \sim \Gamma(k , \lambda)$ . Thus, $S_N$ is a mixture of a Gamma r.v., and a Binomial r.v., and it's PDF will be : \begin{align*} f_{S_N} (x) &= \sum_{k = 0}^n f_{\Gamma}(k , \lambda) \cdot f_N (k)\\ &= e^{-\lambda x}\sum_{k = 0}^n \frac{\lambda^k x^{k-1}}{(k-1)!} \cdot \binom{n}{k}p^k (1 - p)^{n-k} \end{align*} I don't know if this sum can be simplified or not. Whatever be the sum, this is the density of $S_N$.

$\endgroup$
5
  • $\begingroup$ Yes. I got a similar expression for CDF. But not sure any existing distribution fits the density. $\endgroup$ – Mathxx Nov 17 '20 at 8:37
  • $\begingroup$ By the term "existing distribution", you mean commonly used distributions like Normal, Cauchy, Beta etc., right ? $\endgroup$ – Kolmogorov Nov 17 '20 at 8:38
  • $\begingroup$ Yes. I mean the common distributions. $\endgroup$ – Mathxx Nov 17 '20 at 8:39
  • $\begingroup$ Well, it doesn't matter. One should even expect that a mixture of two "well-known" distributions will give rise to another "well-known" distribution ! This "common distribution" thing is just a name, nothing else. $\endgroup$ – Kolmogorov Nov 17 '20 at 8:43
  • $\begingroup$ ibb.co/fDPQQrj Mind asking does this fit my question? Thank you. $\endgroup$ – Mathxx Nov 17 '20 at 8:48
2
$\begingroup$

Since $X_1,...,X_N \sim \text{IID Exp}(\lambda)$ you have $S_N | N \sim \text{Ga}(n, \lambda)$. Applying the law of total probability conditional on $N=k$ gives you the probability density function:

$$\begin{align} f_{S_N}(s) &= \sum_{k=0}^n \text{Ga}(s| k, \lambda) \cdot \text{Bin}(k|n,p) \\[6pt] &= \mathbb{I}(s=0) \cdot \text{Bin}(0|n,p) + \sum_{k=1}^n \text{Ga}(s| k, \lambda) \cdot \text{Bin}(k|n,p) \\[6pt] &= \mathbb{I}(s=0) \cdot (1-p)^n + \exp(-\lambda s) \sum_{k=1}^n \frac{\lambda^k s^{k-1}}{(k-1)!} \cdot \frac{n!}{k!(n-k)!} \cdot p^k (1-p)^{n-k} \\[6pt] &= \mathbb{I}(s=0) \cdot (1-p)^n + \frac{(1-p)^n}{s} \cdot \exp(-\lambda s) \sum_{k=1}^n \frac{(n)_k}{k} \cdot \bigg( \frac{\lambda s p}{1-p} \bigg)^k \\[6pt] &= \mathbb{I}(s=0) \cdot (1-p)^n + \frac{(1-p)^n}{s} \cdot \exp(-\lambda s) \cdot G \Big( \frac{\lambda s p}{1-p} \Big), \\[6pt] \end{align}$$

where $(n)_k = \prod_{i=1}^k (n-i+1)$ denotes the falling factorial, and where we have used the ordinary generating function for the series $0, (n)_1, (n)_2/2, (n)_3/3, ...$, which is given by:

$$G(t) \equiv \sum_{k=1}^n \frac{(n)_k}{k} \cdot t^k.$$

This is a mixture distribution with a discrete part at $s=0$ and continuous part over the support $s>0$. I am not aware of any further simplification for the ordinary generating function $G$, so it would seem that this is the most simplified form for the density.

$\endgroup$
1
  • $\begingroup$ Can you define the Pochhamer symbol? And, I think there is missing the mixing weight $1-(1-p)^n$, and neither can I see that the combinatorial part simplifies that much ... $\endgroup$ – kjetil b halvorsen Nov 26 '20 at 21:36
1
$\begingroup$

Building on the answer by @Kolmogorov. First, this mixture distribution does have an atom at zero, since if $N=0$ (with probability $(1-p)^n$), then $S_N=0$, as an empty sum. In the other case, conditional on $N\ge 1$, $S_N$ has a continuous distribution. The sum, given by @Kolmogorov, seems difficult, but maple is able to "simplify" it, and finds a solution with Laguerre polynomials:

sum( (lambda^k*x^(k-1) * exp(-lambda*x) / (k-1)! )*binomial(n,k)*p^k*(1-p)^(n-k), k=1..n ) assuming lambda> 0,n,posint,x,real,p>0,p<1;

with result $$ \lambda\,{{\rm e}^{-\lambda\,x}}p \left( 1-p \right) ^{n-1} \left( -{ \it LaguerreL} \left( n,{\frac {\lambda\,xp}{-1+p}} \right) +{\it LaguerreL} \left( n,1,{\frac {\lambda\,xp}{-1+p}} \right) \right) $$ See NIST Digital Library of Mathematical Functions. For instance, in R, there is an implementation in package mpoly on CRAN, and probably otherplace.

$\endgroup$

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.