I don't think it's learning COBOL language that's the hard part. An inexperienced programmer who hasn't worked on financial systems is going to make security mistakes and write buggy code for awhile, both of which can be deadly to a bank. But it would be better to have that person rewrite the infrastructure than try to fix bugs. The problem with these ancient systems is trying to understand how bugs arise from subtle unintended interactions within millions of lines of code. I've been coding many languages for 25 years and I would consider it extremely daunting to try to tackle a bank bug knowing that one wrong assumption about one line of code could crash whole systems or cost millions of dollars.