Timeline for Does this ARIMA model take seasonality into account?
Current License: CC BY-SA 4.0
6 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Nov 16, 2021 at 8:30 | comment | added | Richard Hardy | @StephanKolassa, I should probably have refrained from saying a lot [like], but you got the point. Your response makes sense. Thank you! | |
| Nov 16, 2021 at 6:45 | comment | added | Stephan Kolassa | @RichardHardy: yes, you do have a point there. Then again, the OP is writing a tutorial. The intended audience may indeed profit from a little strawman beating. And yes, I agree that regularization like constraining parameters to be zero would be a worthwhile follow-on topic. Then again, if we can trust our students to understand this, we should really first be looking at seasonal models. | |
| Nov 16, 2021 at 6:34 | vote | accept | codeananda | ||
| Nov 14, 2021 at 19:54 | comment | added | Josef |
The estimated coefficient for lag 4 is close to 1, ar.L4=0.9613, but still smaller than one. So the seasonal cycle will shrink over the forecast horizon. This still gives different long run behavior than a model with seasonal differencing.
|
|
| Nov 14, 2021 at 14:16 | comment | added | Richard Hardy | This seems a lot like beating a straw man. A reasonable person would not fit a high-order AR model in an unregularized way. A more relevant example would be to show that an AR(5) model with appropriate zero restrictions (lags 1, 2 and 3 set to zero) has only one more parameter to estimate than a SARIMA(1,0,0)(1,0,0)$_4$ model. I discuss a similar case here. | |
| Nov 14, 2021 at 10:01 | history | answered | Stephan Kolassa | CC BY-SA 4.0 |