0
$\begingroup$

Let $X_1, X_2, \cdots, X_n$ be independent and identically distributed Poisson random variables and define the partial sums $S_n = X_1 + X_2 + \cdots + X_n$.

I would like to calculate the expectation value $E(S_n)$ where I count only $X_i > A$ (A = const) and skip $k=2$ next random variables in the sum after the event $X_i > A$ occurs (no matter they are greater or smaller than A).

$\endgroup$
1
$\begingroup$

Assume $n\geq 2$. Let $p=P(X_i\leq A) = \sum_{j=0}^{A}{\dfrac{\lambda^j}{j!}e^{-\lambda}},\;$ and let $q = \sum_{j=A+1}^{\infty}{\dfrac{\lambda^j}{(j-1)!}e^{-\lambda}}$.

For each $j$, if $X_j$ "counts" then it contributes amount $q$ to the required expectation.

\begin{eqnarray*} X_1 && \text{ counts if $X_1\gt A$} \\ X_2 && \text{ counts if $X_1\leq A$ and $X_2 \gt A$} \\ X_j && \text{ counts if $X_{j-2}\leq A$ and $X_{j-1}\leq A$ and $X_j \gt A$, for $j\geq 3$.} \\ \end{eqnarray*}

So, using linearity of expectation,

\begin{eqnarray*} E(S_n) &=& q + qp + \sum_{j=3}^{n}{qp^2} \\ &=& q(1 + p + (n-2)p^2). \end{eqnarray*}

$\endgroup$
  • 1
    $\begingroup$ Thank you very much for your reply! One maybe a bit stupid question: It is not clear for me how do you pass from expectation to probability. I mean that you have only probabilities from the right side (p) and expectation from the left (E($S_n$)). Should it be E($X_i$|$X_i>A$) somewhere in the right side? $\endgroup$ – poisson_fish Apr 16 '15 at 12:08
  • $\begingroup$ @poisson_fish Sorry, you are right about that. I've revised my answer accordingly. I don't think there is any simple expression for variables $p$ or $q$ unfortunately. $\endgroup$ – Mick A Apr 16 '15 at 13:02

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.