Let $\mathbf{X}$ be an $N$-vector of independent random variables drawn from some distribution with mean $\mu$ and variance $\sigma^2$: $$ \mathrm{E}\left[x_1\right]=\cdots=\mathrm{E}\left[x_N\right]=\mu, \mathrm{Var}\left[x_1\right]=\cdots\mathrm{Var}\left[x_N\right]=\sigma^2 $$ Let $\mathbf{Y}$ be the vectors of $n$-consecutive sums of $\mathbf{X}$: $$y_i=\sum_{j=i}^{i+n-1}x_j,i=1,\cdots,N-n+1\tag{1}\label{1}$$ or $$\mathbf{Y}=\mathbf{X}\mathbf{M}\tag{2}\label{2}$$ where $$\mathbf{M}=\begin{bmatrix} \left.\begin{matrix} 1& 0&\cdots&0\\ 1& 1&\cdots&0\\ \vdots & \vdots&\ddots&\vdots\\ \end{matrix}\right\}&N-n\\ \left.\begin{matrix} 1& 1 &\cdots&1\\ 0&1&\cdots&1\\ \vdots&\vdots&\ddots&\vdots\\ 0&0&\cdots&1\end{matrix}\right\}&n \end{bmatrix}$$ I have worked out the results of $\mathrm{E}\left[\overline{y_i^r}\right]$, $\mathrm{Var}\left[\overline{y_i^r}\right]$, $\mathrm{E}\left[\overline{\left(y_i-\overline{y_i}\right)^r}\right]$ and $\mathrm{Var}\left[\overline{\left(y_i-\overline{y_i}\right)^r}\right]$ for $r=1,2$, and also $\mathrm{E}\left[\overline {y_i}^2\right]$ and $\mathrm{Var}\left[\overline {y_i}^2\right]$, if useful. I am hindered by the cases of $r=3,4$. Can they be expressed by the known results above?

For example, I see no hope in expanding $\overline{y_i^4}$ by eq. \eqref{1}, neither can I express $\overline{y_i^4}$ into some quadratic forms by eq. \eqref{2}. I found on the Internet some discussions about the maximum of partial sum of random variables (for example, http://msp.org/pjm/1974/52-2/pjm-v52-n2-p27-p.pdf), but no discussions of the sample moments of it. One special characteristic of the current problems is $y_i$ and $y_{i+1}$ are highly overlapped and hence not independent. I know limited mathematical tools but are willing to learn.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.