If a model with large AR and/or MA orders is specified on the ESTIMATE statement in PROC ARIMA, and the OUTLIER statement is used to perform outlier detection on the residuals from the resulting model, then some machines may return any one of the following errors:

ERROR: Read Access Violation in Task [ ARIMA ]
       Exception occurred at (67E2A23D)

ERROR: Segmentation Violation in Task [ ARIMA ]

ERROR: Floating Point Overflow. ERROR: Termination due to Floating Point Exception

ERROR: The OUTLIER step has been terminated due to insufficient memory.

To circumvent the problem, specify a model that is an approximation to the desired model, but does not include the large AR and/or MA orders. For example, the model defined by the following PROC ARIMA statements:

proc arima data=a;
   identify var=y(7) nlag=380 noprint;
   estimate p=(1,2,3,4,5,6,7)(365) q=(7)(365);
   outlier maxnum=10 id=date;
   run;

could be approximated for outlier detection purposes with the following:

proc arima data=a;
   identify var=y(1,7,365) nlag=380 noprint;
   estimate p=7 q=7;
   outlier maxnum=10 id=date;
   run;

An approximation such as this uses differencing in place of the large AR and MA orders, but still retains some of the flavor of the original model.