Banks will "maybe" be done with COBOL in ten years

Why is it so hard to replace? Are the algorithms just that good / impenetrable? Is the language particularly well suited for the task?

Theres a whole mainframe ecosystem of schedulers, databases, disk formats, batches, 3270 screens etc that are hard to replace piecemeal

Which is what they said in 1990 as well.

Fully translated to F77 by then ...