The aim in credit scoring is to distinguish good clients from bad clients. This is done at the stage of the application (application scoring - AS) based on demographic and similar data. AS is used to decide whether the application is accepted or rejected.
When the credit is being repayed the bahavior of the obligor (did s/he need any reminders, ...) is used to calculate a behavior scoring (BS). The BS can be used to estimate needed reserves for the credit given and for further credit descisions
It is known (and clear) that the discriminatory power of AS deteriorates in the months after the application. Thus it is good practice to combine AS and BS.
What are best practices about when to combine AS and BS and how to do this? I am looking for references and opinions in the context of banking credit risk.
Machine learning like aspects are very important and interesting but I would be happy to have the banking context too. Thanks a lot !