If you’re building a time series model and you want any hope of statistical significance, do your analysis on non-overlapping periods.https://twitter.com/macrocephalopod/status/1460302495507525635 …
-
Technically you can get loads more data with overlapping intervals. Adjusting hypothesis tests for the autocorrelation structure is a bit of a mess. (not impossible though). Most people would never do this and conclude things entireyl wrongly.
If you only have the overlapping intervals that’s true for some use cases. But if you have the sub-periods too then the overlapping intervals can’t buy you any more power (for obvious info theory reasons).
-
-
This paper uses overlapping intervals, with adjusted stderrs (not the usual HAC ones but small sample versions) to estimate biases in long-horizon effects.https://twitter.com/firoozye/status/1460366581054251016 …
-
Some older references on overlapping intervals from an excellent Stack Exchange answerhttps://stats.stackexchange.com/questions/154979/covariance-estimation-of-overlapping-time-series/155039#155039 …
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.