Expected value of a Gamma RV to the power of a Poisson RV

276 Views Asked by At

$\mathit{W}$ is a $\bigl(\alpha = 3, \beta = \frac 12 \bigr)$ -Gamma random variable, and $\mathit{N}$ is a $\mu$ = $\frac 13$ -Poisson random variable, independent from $\mathit{W}$.

What is $\mathbb{E}$$\bigl[\mathit{W^N} \bigr]$?

Note: $\mathit{p_N}$ = P$\bigl(N=n \bigr)$ = $\ e^\mu$ $\frac {\mu^n}{n!}$ for $\mu$- Poisson RV.

Also, the pmf for $\bigl(r,p \bigr)$- Negative Binomial RV is $\mathit{p_T (n)}$ = $\begin{pmatrix}n-1\\r-1\end{pmatrix}$ $\mathit{p^r}$$\mathit{(1-p)^{n-r}}$, for $\mathit{n= r, r+1, ...}$

So far, I have that I need to make a negative binomial variable out of $\mathit{W^N}$, I'm assuming by manipulating their pmfs and pdfs.

I have that P$\bigl(W^N \leq t \bigr)$ = P$\bigl(W \leq t^{1/N} \bigr)$, and from there I tried to find their joint cdf, which I think is just replacing t in the gamma pdf by $t^{1/n}$, and then use the usual rule to find the conditional cdf, which should be something like a negative binomial cdf.

However, when I do that, I just get stuck and can't really move forward. Am I heading in the right direction? Are there other ways to solve this?

Thanks!

2

There are 2 best solutions below

3
On BEST ANSWER

We often don't need to compute the full pmf if we only want the expected value.

Here, we can use the law of total expectation, as in $E[E[g(X,Y) \mid Y]]=E[g(X,Y)]$, to write

$$ E[W^N] = E [E[W^N | N]] \tag1$$

But $E[W^n]$, for any fixed positive integer $n$, is the $n-$th moment (non centered) moment of $W$, a $(\alpha,\beta)$ Gamma rv.

Taking advantage of this question, we can write

$$ E[W^n] = \frac{\beta^n \Gamma(\alpha+n)}{\Gamma(\alpha)}\tag2$$

which is also valid for $n=0$. In our case this gives

$$ E[W^n] = \frac{ \Gamma(3+n)}{2^n \Gamma(3)}=\frac{(n+2)!}{2^n 2!}=\frac{(n+2)!}{2^{n+1}} \tag3$$

Then $$E[W^N] = E [E[W^N | N]] = E\left[\frac{(N+2)!}{2^{N+1}} \right] \tag4$$

where $N$ follows a Poisson distribution with mean $1/3$. Can you go on from here?

0
On

It is even easier to compute if you condition on W, as the generating function of the Poisson distribution is a relatively simple exponential term and then you have to compute a relatively simple integral.