It's a really common misconception that "most bank software is still written in Cobol".
Of course it varies from bank to bank, and of course there's still some legacy systems on Cobol. But many banks run primarily Java [1] in their backend. They have to: besides the fact that Java is a much more productive language than Cobol, it's the only way you can actually hire people nowadays! No bank running "most software" on Cobol is sustainable. The supposed fear to rewrite or touch simply isn't real either: if a bank does not understand and therefore care enough for "most" of its infrastructure, it isn't a sustainable bank either. These large corporations tend to be as conservative as you can get, but they are not totally oblivious.
That doesn't mean there are some little used systems still running on ancient stuff. I once heard an anecdote (no way to verify if true) that the Dutch revenue service rehired 80 year olds because they were the only ones that understood how to program some of their old machines that were programmed using wires and knobs.
The financial services market was the first market that was primarily an IT market, and a lot of infrastructure was built up on stuff that is totally unfamiliar nowadays. There are places that will pay you a lot of money for programming in an ancient language.
[1]: It's interesting to think that Java will be the Cobol of next few decades. Although I've seen signs that Scala and even Clojure are entering the banking world as well, which is encouraging.