There are three types of number formats (that I'm aware of): DISPLAY, PACKED-DECIMAL (or COMPUTATIONAL-3 aka COMP-3), and BINARY. These are defined in the data division after the PIC clause (although DISPLAY format is default so you don't really have to define it). In DISPLAY format (aka character format) a single digit (i.e. PIC 9) would use 1 byte (8 bits) of memory. In order to save space and processing time, calculation can be performed in a more economical way using COMP-3 or BINARY formats. This is because they use less bytes:

e.g. 

01 ITEM-1  PIC 9(3) VALUE 123 USAGE IS DISPLAY.

This uses 4 bytes of memory: one for each digit plus one for the sign (+ or -).

01 ITEM-1  PIC 9(3) VALUE 123 USAGE IS PACKED-DECIMAL.

This uses 2 bytes: each individual digit is converted to binary -
                        1    2    3   sign
                      0001 0010 0011  1111 (unsigned)
                                      1101 (negative)
                                      1100 (positve)


01 ITEM-1  PIC 9(3) VALUE 123 USAGE IS BINARY.

This uses 1 byte: 123 (one-hundred and twenty-three) is converted into binary: 01111011

        :
     000100 01 NUMBER-1    PIC 9(3) USAGE IS BINARY.
     000110 01 NUMBER-2    PIC 9(3) USAGE IS BINARY.
     000120 01 ANSWER-OUT  PIC 9(4) USAGE IS DISPLAY.
        :
     000200       MULTIPLY NUMBER-1 BY NUMBER-2 GIVING ANSWER-OUT
     000210       DISPLAY ANSWER-OUT

Both 'USAGE' and 'IS' are optional (as is 'DISPLAY').