The X12 and X13 procedures issue

WARNING: Extreme values will be treated as missing.

if any value of the variables listed in the VAR statement is greater than or equal to 1e13. These extreme values are set to missing, and a summary table of "Extreme or Missing values" is printed prior to Table A1: Time Series Data in the PROC X12 or PROC X13 output.

The U.S. Census Bureau programs X-12-ARIMA and X-13ARIMA-SEATS check for embedded missing values in the input data. However, they do not identify extreme values in the input series and set them to missing. The difference in the treatment of extreme values in the input data causes differences in the subsequent seasonal adjustment results between the X12 and X13 procedures and the Census Bureau programs.

To circumvent the problem, you can use a prior DATA step to rescale the analysis variables by a factor or your choice so that none of the variables has a value greater than or equal to 1e13. For example, this code rescales the variable Y using a factor of 10000:

data b;
   set a;
   y=y/10000;
   run;

When you run the analysis on the rescaled data, tables that are “expressed as percent”, such as Table D10: Final Seasonal Factors, directly match the results computed by the Census Bureau. Other tables that depend on the scale of the data, such as Table D11: Final Seasonally Adjusted Data and Table D12: Final Trend Cycle, need to be post-processed in order to match the Census Bureau results. Post-processing returns the table results back to the metric of the original data. You can do this by writing these tables to the OUTPUT OUT= data set and then using a DATA step to multiply the table variables by the same scaling factor used in the earlier DATA step.