Why should we remove trend and seasonality (hence, making a series stationary) before forecasting? If time series has a particular trend/seasonality, shouldn't we incorporate that model instead because its indeed useful?
$\begingroup$
$\endgroup$
4
-
1$\begingroup$ Could you point to where you have seen this advice? Depending on the exact context, it is either wrong (as you suggest) or a different way of loosely saying the same thing (i.e. "removing" = "incorporating the effect, then looking at what's still not explained"). $\endgroup$ – Chris Haug Oct 2 '20 at 21:46
-
1$\begingroup$ Once you have made a forecast for the residuals after removing the trend and seasonality, you can construct a forecast for the original time series by adding back the seasonality and trend. $\endgroup$ – Henry Oct 3 '20 at 7:18
-
$\begingroup$ @ChrisHaug I was studying the basics of Time series and its forecasting where I came across the text from an Actuarial science textbook that when forecasting for a tome series is done it is required to first difference and remove any exponential or linear trnad or any seasonal trend before forecasting $\endgroup$ – Devansh Gandhi Oct 3 '20 at 7:46
-
$\begingroup$ @Henry thanks for some clarity $\endgroup$ – Devansh Gandhi Oct 3 '20 at 7:49
Add a comment
|
