Dear All,
I am very new to SAS, and I am currently running a dynamic regression model without transfer functions, where the noise term follows a multiplicative seasonal and nonseasonal ARIMA model of the form: ARIMA(1,1,1)(1,1,2,12). (My reference is this source: https://documentation.sas.com/doc/en/etsug/15.2/etsug_arima_gettingstarted25.htm). My dataset is monthly and is attached to this message as an Excel file.
In SAS, I am using the following code:
proc arima data=a;
identify var=Yt(1,12) crosscorr=(X1t(1,12) X2t(1,12));
estimate p=(1)(12) q=(1)(12 24) input=(X1t, X2t) method=ml noconstant;
run;
When performing the same procedure in STATA, I use:
gen Date = _n
tsset Date
arima Yt X1t X2t, arima(1,1,1) sarima(1,1,2,12) nocons, if Date<=168
Both SAS and STATA yield nearly identical results for the estimated coefficients, except for the seasonal MA terms. Specifically, STATA reports absolute values of "0.3621697" and "0.780338", whereas SAS computes "0.20275" and "0.66553". This discrepancy suggests a potential error in my SAS code or differences in optimization convergence.
Could you please assist in identifying any errors in my SAS code, or suggest additional options to potentially enhance result accuracy?
Many thanks.
