I am reaching out for a problem I have seem to hit wall on. I am trying to build a rare event model:- Default event occurred in NY/NJ area.
Obviously my logistic and robust logistic models are failing Hosmer GOF tests and so I am thinking about bootstrapping data. I found a great read on R-help, but I was wondering if someone can help me with a more practical direction before I head into a road I have traveled less.
Some key statistics:
138 default aka 4bp (.04%).
I have about unique ID's from [300,2700] over time[1999-2011], duplicated to total dataset of 290K. My question is when bootstrapping data: 1. Am I bootstrapping unique ID's [eg., 2700 id's] or Observations [290k obs]? 2. How do I oversample binary event = 1 [138 defaults and built robust model]?
Thank you in advance.