0
$\begingroup$

The ACF and PACF of a stationary timeseries "myts" is as given below:

enter image description here

> myts
      [,1]  [,2]  [,3] [,4]  [,5]  [,6] [,7] [,8]  [,9] [,10] [,11] [,12] [,13] [,14] [,15]
ACF  -0.18 -0.11 -0.15 0.11 -0.31  0.03 0.54 0.00 -0.20 -0.12  0.07 -0.22 -0.01  0.38  0.14
PACF -0.18 -0.15 -0.21 0.02 -0.37 -0.16 0.53 0.14 -0.04 -0.05 -0.09  0.00 -0.11  0.02  0.20
     [,16] [,17] [,18] [,19] [,20] [,21] [,22] [,23] [,24] [,25] [,26] [,27] [,28] [,29] [,30]
ACF  -0.31  0.06 -0.03 -0.07 -0.15  0.40 -0.06 -0.15 -0.06  0.04 -0.09 -0.06  0.15  0.00 -0.03
PACF -0.18  0.24 -0.01 -0.01  0.02 -0.09 -0.18 -0.05 -0.12 -0.05  0.02  0.04 -0.18 -0.07  0.18
     [,31] [,32] [,33] [,34] [,35] [,36] [,37] [,38] [,39] [,40] [,41] [,42] [,43] [,44] [,45]
ACF  -0.07  0.06 -0.07  0.01  0.02  0.05 -0.09  0.04 -0.01 -0.01 -0.05  0.05 -0.04 -0.01 -0.01
PACF -0.03 -0.06  0.04  0.01  0.03 -0.03 -0.02 -0.02  0.02  0.01 -0.09  0.05  0.04 -0.11 -0.09
     [,46] [,47] [,48] [,49] [,50] [,51] [,52]
ACF   0.01  0.00  0.01  0.00 -0.01  0.00  0.00
PACF -0.02 -0.08 -0.03  0.02  0.06 -0.08  0.09

I read this and auto.arima(rawts) suggests ARIMA(0,1,0)(1,0,1)[7]. But I would like to know from ACF/PACF plots.

I found weekly season pattern in ACF. There are significant positive spikes in the ACF plot at lag 7,14 and 21. I assume ACF tails off and PACF Cuts off at lag 7. But I am unable to choose the parameters. Is it ARIMA(p=0,d=1,q=1)(P=1,D=0,Q=0)[7]....?

May I know what SARIMA parameters (p,d,q and P,D,Q)can be chosen based on the these ACF and PACF plots?

$\endgroup$

1 Answer 1

1
$\begingroup$

On one hand the ACF plot lets you pick the possible values for $q$ and $Q$ of the $SARIMA$ model, on the other hand the PACF plot lets you pick the possible values for $p$ and $P$.

I think its better to show in an example, I have the ACF/PACF plots for a monthly time series. I applied seasonal differencing on this time series to remove the seasonality and a first diffrence to make the series stationnary.

ACF PACF to detect parameters of SARIMA

First lets consider the ACF:

  • We have a spike at lag 1 so $q=1$.
  • Also we have a spike at lag 12 wich is the frequency of the series so $Q=1$.

Let's take a look at the PACF:

  • We have spikes at lags 1, 2 and 3 so the possible values are $p=\left\{1,2,3\right\}$
  • Also we have a spike at lag 12 so $P=1$

In conclusion, we have $p=\left\{1,2,3\right\}$, $q=1$, $P=1$ and $Q=1$ with $d=1$ and $D=1$. The different combination of the parameters gives 32 models to estimate and then choose the one that minimize an information criteria i typically choose $AIC$.

In your case you don't have significant spikes in the first lags in both ACF and PACF, we can assume that $p=0$ and $q=0$. In both plots you got a significant spike at lag 7 which is the frequency of your time series so we can conclude that $P=1$.

Finaly, in the ACF shows significant spikes at lags 7, 14 and 21 so the possible values for $Q$ are $\left\{1,2,3\right\}$.

So identifying using ACF/PACF plots gives three possible models:

  • $SARIMA(0,1,0)(1,0,1)_{7}$
  • $SARIMA(0,1,0)(1,0,2)_{7}$
  • $SARIMA(0,1,0)(1,0,3)_{7}$

Hope this was of help to you.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.