There are occasions when you may wish to set a variable to an infinitely high or infinitely low number. For example, suppose you were merging two files on surnames as the primary key:

*in data division FILE SECTION

 FD FILE-1.
 01 RECORD-1.
     03 IN-NAME-1 PIC X(20).
     03 FILLER    PIC X(50).

 FD MERGE-FILE.
 01 RECORD-OUT    PIC X(70).

   :
   :

   PERFORM WITH TEST AFTER EOF-FLAG-1 AND EOF-FLAG-2
*loop until each file has been read to completion
*read each file
      READ FILE-1 
          AT END SET EOF-FLAG-1 TO TRUE
          MOVE HIGH-VALUES TO IN-NAME-1 
      END-READ
      READ FILE-2
          AT END SET EOF-FLAG-2 TO TRUE
          MOVE HIGH-VALUES TO IN-NAME-2 
      END-READ

*sort the records (assuming no 2 names are the same)
*on ascending surname
      IF IN-NAME-1 IS < IN-NAME-2 THEN
          WRITE RECORD-OUT FROM RECORD-1
      ELSE
          WRITE RECORD-OUT FROM RECORD-2
      END-IF

   END-PERFORM

In this example, when IN-NAME-1 is less than IN-NAME-2 (based on their ASCII values e.g. A < B etc..) then the FILE-1 record (RECORD-1) is written to the merge file (RECORD-OUT). One of FILE-1 and FILE-2 will come to an end before the other so the completed file has its IN-NAME-_ value set to constant that will ALWAYS be greater than the IN-NAME-_ value still being read, ensuring all remain files are written to the merge file. This is done with the lines: MOVE HIGH-VALUES TO IN-NAME-1 and MOVE HIGH-VALUES TO IN-NAME-2

It is important to note that HIGH-VALUES and LOW-VALUES are ALPHANUMERIC in type, so you can't set numerically defined variables to this type (you would have to implicitly redefine the variable first). This is an annoying quirk of COBOL.