258. VARMA
Vector autoregressive moving average
Vector analog of ARMA. Identifiability requires care (e.g., echelon forms).
Parameters: , , ,
Orders: , ,
Example:
Given
- Orders: , ,
-
Coefficient matrices:
- Intercept:
- Initial conditions: ,
- Two series stacked into :
| 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 |
| 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | |
| 8 | 9 | 7 | 6 | 10 | 11 | 9 | 8 | 12 | 13 | 11 | 10 | 14 | 15 | 13 | 12 |
Step 1 — formula
Substitute , into the VARMA recursion:
Forecast (set ):
Componentwise:
Innovation:
Step 2 — apply at
With , , every product is zero:
Step 2b — apply at
Plug in , :
Step 3 — iterate
Two contributions per row: AR part and MA part . Values rounded to 4 decimal places.
| 1 | ||||
| 2 | ||||
| 3 | ||||
| 4 | ||||
| 5 | ||||
| 6 | ||||
| 7 | ||||
| 8 | ||||
| 9 | ||||
| 10 | ||||
| 11 | ||||
| 12 | ||||
| 13 | ||||
| 14 | ||||
| 15 | ||||
| 16 |