ARIMA-transfer function

[uribem]

In order to prewhiten and filter my variables for a cross correlation analysis using the arima procedure:

proc arima data=in;
      identify var=x;
      estimate p=1 q=1;
      identify var=y crosscorr=x;
   run;

how do I specify a seasonal ARIMA model in the ESTIMATE statement? What syntax should I use? 

My model is a seasonal ARIMA (1,0,0) (1,0,1) perioid=6.

Thanks,

Maria

[chwell]

Hi Maria.

It looks like you have taken the appropriate steps to pre-whiten the x for transfer function identification. If you send details of how the x variable enters the model, the pre-whitened CCF will work, I'll be happy to help with that.

As far as your ARIMA specification, the syntax below should specify the model. Note, I'm assuming that the paretheses indicate a factored specification, and that the second set of parentheses indicate seasonal factors.

proc  arima  data=in;

   identify var=x;

   estimate p=1 q=1;

   identify var=y crosscorr=(x);

   estimate input=(<transfer fnt      

     for x goes here>)

     p=(1)(6) q=(6) ml;

run;

Also, you should check out the University Edition of SAS Studio. We have created some forecasting tasks that allow you to create ARIMA specifications in a point and click environment, and then see the corresponding model syntax. See, https://www.sas.com/en_us/software/university-edition/download-software.html

Hope this helps, and feel free to follow up. Best, Chip

[uribem]

Hi Chip, 

I'm actually not interested in using the transfer function for prediction. The CCF is as far as I need to go in my analysis. but I wanted to make sure that the syntax for specifying the seasonal ARIMA to be fitted is correct. 

I don't really understand why you are using two IDENTIFY and two ESTIMATE statements, is that necessary? All I want to do is to run CCF with pre-whitened series, using a seasonal ARIMA for the pre-whitening.

Thanks!

Maria

[chwell]

Hi Maria.

Sorry if I miss-understood you question.

The way to produce pre-whitened CCF plots in Proc ARIMA is to:

1) identify the input or X variable.

2) Estimate a model for X that results in white noise residuals. This model is the pre-whitening filter.

3) Identify the Y variable and list the X variable in a crosscorr=(X) option.

Modifying the code sent earlier;  

proc  arima  data=in;

   identify var=x;

   estimate p=(1)(6) q=(6) ml;;

   identify var=y crosscorr=(x);

run;

Hope this helps. Chip