253. ARIMA
Autoregressive integrated moving average
Differencing operator removes trend.
Apply times to make non-stationary series stationary.
Parameters: , , ,
Orders: , ,
Example:
Given
- Orders: , ,
- Parameters: , ,
- Initial conditions: ,
- Data:
| 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | |
| 12 | 10 | 8 | 11 | 14 | 12 | 9 | 13 | 16 | 14 | 11 | 15 | 18 | 16 | 13 | 17 |
Step 1 — formula
Substitute , , into the ARIMA recursion:
Difference — apply to convert to a stationary series:
In the differenced space, the model is ARMA:
Forecast the difference (set ):
Undifference — convert the difference forecast back to a forecast for :
Innovation:
Step 2 — apply at
First difference: .
Plug in , , , :
Step 3 — iterate
Two-stage pipeline at each : difference ARMA-forecast the difference undifference. Values rounded to 4 decimal places.
| 2 | 10 | ||||
| 3 | 8 | ||||
| 4 | 11 | ||||
| 5 | 14 | ||||
| 6 | 12 | ||||
| 7 | 9 | ||||
| 8 | 13 | ||||
| 9 | 16 | ||||
| 10 | 14 | ||||
| 11 | 11 | ||||
| 12 | 15 | ||||
| 13 | 18 | ||||
| 14 | 16 | ||||
| 15 | 13 | ||||
| 16 | 17 | ||||
| 17 | — | — | — |