They have no reason to do so, that project fails a risk / reward analysis hard.
Old does not mean bad, there are many more recent but much worse languages which have not stood the test of time. COBOL works well for what they do, the existing code is written in it, and it is well tested.
Designing a new language requires that you have staff who can design, implement, and test that compiler / interpreter. You then need to task that staff with doing so instead of something else which could make the company money, and retrain the programmers who would be using it.
If you try to pitch a project to an insurance or banking VP which involves a lot of risk, a lot of cost, and has no real benefit... you will at a minimum get a polite no as an answer.
C is pretty old as well (40 years compared to COBOLs 55), but most really important modern code is still written in it. I would not be surprised if it was still a popular language at the century mark.
The syntax for general mathematics is about 400 years old, and still going strong.