This does look like an overfit:

I used 30 days in my initial model, but then I noticed that 60 days gave better results. The exciting part is that 61 days or 59 days provides the worst consequences.

It shouldn't be so sensative to such change IMHO.