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