Inspired by this question, I tried to get an expression for the third central moment of a sum of a random number of iid random variables. My question is whether it is correct and, if not, what is wrong or what additional assumptions might be missing.

Specifically, let:

$$S=\sum_1^N{X_i},$$ where $N$ is a non-negative integer-valued random variable.

Suppose that the distributions of both $N$ and $X$ are known (and $X_i$ are iid), I want to know the value of the third central moment of $S$.

Using the law of total cummulance:


but $E[S|N]=N\cdot E[X]$, $E[S|N]=N\cdot V[X]$ and, if I am right, $\mu_3(S|N)=N\cdot \mu_3[X]$. Hence:

$$\mu_3(S)=E[N\cdot \mu_3(X)]+\mu_3(N\cdot E[X])+3cov(N\cdot E[X],N\cdot V[X]),$$

and, since the moments of $X$ are supposed to be known:


Of course, $cov(N,N)=V[N]$, so:


Is it right? What is wrong? What additional assumptions am I missing?


Your steps look right to me. We need to assume that the moments exist. The only step I was unsure about was $\mu_3(S | X) = N \cdot \mu_3[X]$. But, we can prove that:

\begin{align} \mu_3(S | X) &= E\left[(S - E[S])^3 | N \right] \\ &= E\left[ \left( \sum_{i=1}^N (X_i - E[X]) \right)^3 \middle| N \right] \\ &= E \left[ \sum_{i=1}^N (X_i - E[X])^3 \middle| N\right] \end{align} where to establish the last equality, we can use the multinomial theorem. For a given $n$,

\begin{align} E\left[ \left( \sum_{i=1}^n (X_i - E[X]) \right)^3 \right] &= E \left[ \sum_{\sum_{i=1}^n k_i = 3} {3 \choose{k_1, \ldots, k_n}} (X_1 - E[X])^{k_1} \cdots (X_n - E[X])^{k_n} \right] \\ &= E\left[ \sum_{i=1}^n (X_i - E[X])^3 \right], \end{align} because when $k_i = 2$ for any $i$, there exists another $j$ where $k_j=1$ (Due to independence of $X_i$ and $X_j$ and the fact that the expectation of $X_j - E[X]$ is zero, causing that particular term to become zero). Now it should be clear that $\mu_3(S | X) = N \cdot \mu_3[X]$.


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.