Can two weak stationary time series $X_t, Y_t$ have covariance $Cov(X_t,Y_t)$ that changes over time? No solution is needed, please give me some hints. I think it can but I find it is difficult to support my idea.
-
If you cannot prove your idea, just sharing your idea is also good. I think Cov(Xt, Yt) may change over time. But I can not find the way to support my idea. – Wayne W Nov 04 '20 at 12:13
-
Is that a homework exercise? If so, please add the [tag:self-study] tag and read its [Wiki](https://stats.stackexchange.com/tags/self-study/info). – Richard Hardy Nov 04 '20 at 12:27
-
It a question that teacher left us, not a necessary homework. Thanks your comment ! – Wayne W Nov 04 '20 at 12:35
-
Hint: let each series consist of *iid* variables. Does this imply the variables in one series must be independent of those in the other? – whuber Nov 05 '20 at 15:05
-
Thanks your hint. Yes! It must be independent. But because of independence, the covariance must be 0 which will not change with t. – Wayne W Nov 06 '20 at 08:28
1 Answers
Let $(X_t),$ with "times" $t=0,1,2,\ldots,$ be variables with independent and identical distributions having finite variance $\sigma^2$ -- which implies they form a stationary (whence weakly stationary) stochastic process. Suppose further that this common distribution is symmetric about $0$: this means the $-X_t$ all have the same distribution as the $X_t.$ For instance, the standard Normal distribution is symmetric about $0.$
Define
$$Y_t = (-1)^t X_t$$
and notice that
$$\operatorname{Cov}(X_t,Y_t) = \operatorname{Cov}(X_t,(-1)^tX_t) = (-1)^t \operatorname{Cov}(X_t,X_t) = (-1)^t\sigma^2$$
alternates between $\sigma^2$ and $-\sigma^2:$ that is, it changes over time. Nevertheless, the symmetry of the distribution implies the $(Y_t)$ are identically distributed and, since the $(X_t)$ are independent, the $(Y_t)$ are independent too. Thus $(Y_t)$ is a (weakly) stationary process, too. In this example the covariance changes over time.

- 281,159
- 54
- 637
- 1,101