The TSCSREG procedure does not provide an option to compute predicted values or residuals. In the PANEL procedure, the PREDICTED= (or P=) option and the RESIDUAL= (or R=) option in the OUTPUT statement can be used to save predicted and residual values in an output data set.

For observations in which all independent variables contain nonmissing values but the dependent variable is missing, PROC PANEL computes predicted values for the dependent variable provided that, for at least one cross section for those time periods, both the dependent variable and all independent variables are nonmissing. If for one or more time periods the dependent variable is missing in all cross sections, then PROC PANEL displays the following error in the log and stops processing.

ERROR: Each observation for time point ts=xxx has a missing value for at least one variable in MODEL statement.

PROC PANEL requires at least one cross section with observed values for all variables specified in the MODEL statement for any time period, and at least one time period with observed values for all variables specified in the MODEL statement for any cross section. If either condition is violated, then the procedure displays the above error.

To obtain out of sample predicted values for some time periods where the dependent variable is missing in all cross sections, consider one of the following alternative approaches.

• In SAS^® 9.4M5 (TS1M5) or above, use PROC HPPANEL with OUT = option and PREDICTED option in the OUTPUT statement.
• If you have SAS^® Viya^® and a SAS^® Econometrics license, use PROC CPANEL on SAS Viya with the PREDICTED= (or P=) option in the OUTPUT statement.
• Compute the predicted values and residuals based on the structural part of the model. This is done with a DATA step to process the OUTEST= data set available in PROC PANEL.
• Alternatively, rather than fit the model in PROC PANEL, use SAS/STAT^® procedure MIXED to estimate a similar panel data model. PROC MIXED estimates a broad class of linear mixed models that includes some panel data models. The OUTP= or OUTPM= option in the MODEL statement in PROC MIXED computes and saves observation-level statistics.

Also, see these notes that discuss the relationship between models estimated using PROC PANEL or PROC TSCSREG and certain models estimated using PROC MIXED:

• Using PROC MIXED to fit models similar to FIXONE, FIXTWO, RANONE, and RANTWO models in PROC PANEL or PROC TSCSREG
• Analyzing unbalanced panel data using PROC PANEL or PROC TSCSREG
• Estimating a model using the Parks method with unbalanced panel data