Credit Data and Scoring

Chapter Thirteen

Regressions

Abstract

This is a practical chapter of how modelers pick variables and combine them to make a credit score. Usually, variables with high information value end up in a model, but often one variable might be so similar to a second variable that while each is predictive, neither adds any extra oomph when one is already in the model. In the end, the combination that best predicts delinquency or default or whatever bad thing there is becomes the method of credit scoring. Therefore, if for whatever reason the borrower would look bad on the variable retained but good on the variable that is dropped, they will appear to have a bad credit score. This is one way that arbitrary things result from regressions, though ...