1
$\begingroup$

Lets say I am given time series plot.

enter image description here

How can I estimate $r_1$ and $r_2$ which are sample correlation? I don't understand how to see correlation from data plot.
I know how to get and use ACF and stuff from the data itself but I am having problem with understanding time series plot.

$\endgroup$
  • $\begingroup$ Why would you try to eyeball a correlation? What's AFC? $\endgroup$ – Glen_b Mar 24 '15 at 5:48
  • $\begingroup$ It is ACF, AFC was typo. Well, I was told to do eyeball it for problem sets and I was really wondered how it works. I have quite amount of stat courses but I always have hard time understanding the plot and eyeball it. $\endgroup$ – Kane Mar 24 '15 at 22:51
3
$\begingroup$

The time series itself is probably not the best way to "eyeball" autocorrelations. A standardized scatter plot with the desired lag between the abcissa (x) and ordinate (y) would do a better job. However, you can imagine the time series plotted with lag=1, lag=2 etc. on the same axes. Then the question is how much the original and the lagged series overlap (for lack of a better term). Your series would probably have slowly decaying autocorrelation since there are relatively flat intervals which, after a shifting, still overlap as long as the shift isn't too large. Then it would spike at lag=7 and lag=14 since there is a clear cyclical pattern of 7 days.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ +1. Agreed with the main idea. The overall weekly pattern, although detectable, is also smudged out a bit: the vertical lines are for Thursdays and each was a weekly peak in only about half of the weeks. The overall trend will reduce the damping a bit. $\endgroup$ – Nick Cox Mar 25 '15 at 19:42

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.