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