The following is a proposition which is important for deriving the distribution of the sample mean for a covariance stationary process.


Let $\{x_{t}\}$ be a mean zero covariance stationary time series with $\sum_{k=0}^{\infty}|\gamma_{x}(k)|<\infty$ ( where $\gamma_{x}(k)$ is the autocovariance function at lag $k$).

Then $\frac{1}{T} \mathbb{V} \left[ \sum_{t=1}^{T} x_{t} \right] = \gamma_{x}(0) + 2 \sum_{k=1}^{\infty}\gamma_{x}(k)$


My proof thus far is:

\begin{align*} \frac{1}{T} \mathbb{V} \left[ \sum_{t=1}^{T} x_{t} \right] &= \frac{1}{T}\left[ \sum_{t=1}^{T} \mathbb{V}[x_{t}] + \mathop{\sum^T\sum^T}_{i\neq j} Cov(x_{i}, x_{j}) \right]\\ &= \frac{1}{T}\left[ \sum_{t=1}^{T} \mathbb{V}[x_{t}] + 2 \mathop{\sum^T\sum^T}_{i < j} Cov(x_{i}, x_{j}) \right]\\ &= \frac{1}{T}\left[ T\cdot \gamma_{x}(0) + 2 (T-1) \cdot \gamma_{x}(1) + 2(T-2) \cdot \gamma_{x}(2) + \ldots + 4 \gamma_{x}(T-2) + 2 \gamma_{x}(T-1) \right]\\ &= \frac{1}{T}\left[ T\cdot \gamma_{x}(0) + 2 \sum_{i=1}^{T-1} (T-i) \cdot \gamma_{x}(i) \right]\\ &= \gamma_{x}(0) + 2 \sum_{i=1}^{T-1} \frac{(T-i)}{T} \cdot \gamma_{x}(i) \\ \end{align*} Now all I need to show is: \begin{align*} \lim_{T\to\infty} \gamma_{x}(0) + 2 \sum_{i=1}^{T-1} \frac{(T-i)}{T} \cdot \gamma_{x}(i) = \gamma_{x}(0) + 2 \sum_{i=1}^{\infty} \gamma_{x}(i) \\ \end{align*}

However I am having trouble with this limit and am worried it does not converge.


This proposition was stated by my professor in a time series class and has been used to prove many other results. However it is possible I have approached the proof in the wrong way. Any input will help.


The upper limit of your sum should be $T-1$ (since the $t=T$ term is zero). We can write

$$\sum_{i=1}^{T-1}\frac{T-i}T\gamma_x(i)=\sum_{i=1}^{T-1}\left(1-\frac iT\right)\gamma_x(i):=a_T. $$ Clearly $a_T$ need not converge as $T\to\infty$. However, $a_T$ converges if and only if $$\sum_{i=-\infty}^\infty \gamma_x(i)<\infty $$ (from my time series notes, a proof of this statement is escaping me at the moment). When that limit exists, your result holds.

  • $\begingroup$ I have incorporated this minor change to the limit of the sum. However, the issue of the convergence of $a_{T}$ is at the crux of the proof; not only must $a_{T}$ converge, but it must converge to the desired limit above. Your comment gives me hope, but the proof of your comment is needed to resolve the issue. $\endgroup$ – möbius Feb 22 '15 at 14:45
  • $\begingroup$ Well, my notes say that the proof is in this book onlinelibrary.wiley.com/book/10.1002/9781118186428 $\endgroup$ – Math1000 Feb 22 '15 at 16:23
  • $\begingroup$ Actually, elementary computations show that the result holds if and only if $$\frac1T\sum_{i=1}^Ti\gamma(i)\to0,$$ a condition which is implied by the convergence of the series $$\sum_{i}\gamma(i).$$ $\endgroup$ – Did Feb 22 '15 at 16:32
  • $\begingroup$ @Did Can you elaborate on why the convergence of your first series to zero is implied by the convergence of the second series? Even a hint would be helpful! $\endgroup$ – möbius Feb 22 '15 at 18:31
  • 1
    $\begingroup$ Rewrite the first sums in terms of the rests $\Gamma(i)=\sum\limits_{k\geqslant i}\gamma(k)$ of the second series, which, by hypothesis, are finite and converge to zero. $\endgroup$ – Did Feb 22 '15 at 19:01

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.