PIC S9(16) is for sure COBOL, and defines a signed numeric data item that holds 16 digits. The presence or absence of a BINARY, COMP-n, or PACKED DECIMAL modifier denotes how this data is stored on disk. The different storage options are described here.
I suspect your manager doesn't know SQL and is just trying to communicate the requirements to you in "Cobolese". If so, he/she is really saying this needs to be a signed integer of up to 16 digits, and you would use the most natural Oracle type, which is NUMBER(16) or just INTEGER.
The OVERPUNCH reference @OMG Ponies found is in Pro*Cobol, which is a tool that lets you embed Oracle SQL statements in COBOL programs. For this to work, you need to map the Oracle internal types like NUMBER with the Cobol types, like PIC S9(16) PACKED-DECIMAL. In this case, you define the Cobol type as an OVERPUNCH LEADING (not TRAILING).
The term "overpunch" comes from the zoned decimal representation on early mainframes. To save space on punch cards the sign would be "overpunched" over the leading or trailing digit.