Certain model specifications in the SSM procedure can lead to incorrect results, including the model parameter estimates, fit statistics, smoothed and filtered component estimates, and other output generated by the procedure.

The results returned by PROC SSM might be incorrect if either of the following conditions is true:

• The model includes multiple TREND statements, or
• A combination of TREND and STATE statements are used in the model specification.

The results returned by PROC SSM are correct if either of the following conditions is true:

• No TREND statement is used in the model specification, or
• Only one TREND statement is specified and no STATE statements are used in the model specification.

An example of a specification that returns incorrect results is Example 6: Model with Multiple ARIMA Components in the PROC SSM documentation. The code used to specify the model is shown below. Note that two TREND statements are included in the model specification.

The table below shows the expected results based on the Bell (2001) reference, along with the results currently returned by PROC SSM:

Although these results might not appear to be vastly different, a check of the difference between the dependent variable, Logy, and the sum of the smoothed components for this model shows that the differences are too large. For the above model, this check can be performed as follows using the smoothed component estimates written to the OUTFOR= data set:

For this model, the minimum and maximum DIF values are -0.0161 and 0.01372, respectively. The differences between the dependent variable and the sum of the underlying components should be on the order of 10e-8 or smaller.

This check can be difficult to perform for some model specifications, including models that contain regressors, dependent lags, or both. Technical Support can assist in defining this check for more complex models if needed.

For some models, the problem can be circumvented by replacing the component defined by a TREND statement with a comparable specification that involves the STATE and COMPONENT statements. This alternative specification can be difficult to program, depending on the trend type.

Click the Hot Fix tab in this note to access the hot fix for this issue.

Operating System and Release Information

Viya on Linux: An update for this issue is available for SAS Viya 3.4. For instructions on how to access and apply software updates, see the Updating Your SAS Viya software section in the SAS Viya 3.4 for Linux Deployment Guide at

Viya on Linux: An update for this issue is available for SAS Viya 3.5. For instructions on how to access and apply software updates, see the Updating Your SAS Viya software section in the SAS Viya 3.5 for Linux Deployment Guide at

A fix for this issue for SAS/ETS 15.2 is available at:

A fix for this issue for SAS/ETS 15.1 is available at:

A fix for this issue for SAS Threaded Kernel Extensions for Econometric Time Series 15.1 is available at:

A fix for this issue for SAS/ETS 14.3 is available at:

A fix for this issue for SAS/ETS 14.2 is available at: