specifying transfer function in proc arima

[prooney2]

I'd like to learn and understand the ETS Transfer Function Model specification syntax.

I want to specify a transfer function that corresponds to the JMP model specifications that are in the attached Word document.

I have an output series (Traffic_In) with which I am trying to predict with two input series: Total_Wi_Fi_Users  and NumTrans_rstrnt3.

Would the corresponding proc arima be:

proc arima data=d.Stor ;

IDENTIFY VAR=Traffic_In(1,7) CROSSCORR= (Total_Wi_Fi_Users (1,7) NumTrans_rstrnt3 (1,7)) nlag=28 ;

ESTIMATE p=(1)(7) q=(1)(7)

INPUT= (  (1)Total_Wi_Fi_Users (1)NumTrans_rstrnt3 / (1)Total_Wi_Fi_Users (1)NumTrans_rstrnt3 ) noprint ; run;

thank you in advance for any help or resources!

[prooney2]

My attached JMP model specification was wrong, here is the updated specification and output

[ShelleySessoms]

JMP has their own community. It is best to post this question there: https://community.jmp.com/

Best wishes,

Shelley

[prooney2]

HI Shelley,
Just want to make sure I made it clear that I'm looking for help in learning proc arima, not JMP

[dw_sas]

Hi,

Based on the updated information, if you let Y=traffic_In, x1=Total_Wifi_Users and x2=Num_Trans_Rstrnt3, then you can specify your model in PROC ARIMA as:

proc arima data=dsname;
  identify var=y(1,7) crosscorr=( x1(1,7) x2(1,7));
  estimate p=(1) q=(1)(7) 
           input=( (1)(7)/(1) x1 (1)/(1)(7) x2) method=ml;
run;
quit;

I believe JMP uses maximum likelihood estimation by default, which is why I added the METHOD=ML option in the ESTIMATE statement.  For more details on the syntax for the INPUT= option when fitting a transfer function model in PROC ARIMA, please see the following documentation link:

http://support.sas.com/documentation/cdl/en/etsug/68148/HTML/default/viewer.htm#etsug_arima_details2...

It is possible that the parameter estimates computed by PROC ARIMA might differ from those obtained in JMP due to differences in the optimization algorithm, starting values, convergence criteria, etc.  However, for a model that fits the data well, the estimates should be relatively close.

I hope this helps!

DW