Timeline for Intuition behind cross-correlation function interpretation vs. correlation of lagged time series
Current License: CC BY-SA 3.0
8 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jul 12, 2021 at 18:50 | comment | added | DMR | @mpiktas For the autocorrelation, lag zero should equal 1, but not necessarily for the cross-correlation between two time series. | |
| Apr 10, 2015 at 5:58 | comment | added | mpiktas | At lag zero correlation should be 1. | |
| Apr 9, 2015 at 16:27 | comment | added | Bakaburg | I have study where I correlate many pairs of time series. it happens often that the r at lag 0 taken by ccf() and the r from the simple cor() are different. any ideas why? the time series has NAs and ccf() is used with na.action=na.pass and the cor() use "pairwise.complete.obs". could this be the cue? | |
| Oct 3, 2013 at 14:05 | vote | accept | JasonAizkalns | ||
| Oct 3, 2013 at 13:06 | comment | added | mpiktas | If the series are non-stationary then sample autocorrelations are meaningless, i.e. they are spurious (it is probably possible to construct non-stationary processes for which they are not spurious). So there is no advantage in not holding means and variances fixed. | |
| Oct 3, 2013 at 12:56 | comment | added | JasonAizkalns |
This is helpful and does an excellent job illustrating the differences between the methods; however, can you elaborate on your comment? "...perfectly valid oepration if the series are considered stationary" Specifically, when would it be advantageous to NOT holding the means and variances fixed? Similarly, is it always best practice to use CCF() when the series are in fact stationary or is there a case to be made for NOT FIXING the means and variances in certain situations?
|
|
| Oct 3, 2013 at 10:40 | history | edited | mpiktas | CC BY-SA 3.0 |
added 15 characters in body
|
| Oct 3, 2013 at 9:55 | history | answered | mpiktas | CC BY-SA 3.0 |