Credit scores are based on statistical analysis of account and payment data. Generally they focus on determining the likelihood of making a bad decision when underwriting a loan. Higher risks are associated with higher returns for the lender - or larger payments for us. This is all then translated into a score -- FICO or the new VantageScore. What is clearly not discussed is how the score is actually created nor the accuracy of the underlying model -- those are trade secrets.
A quick look at the real estate or automotive section of a newspaper shows that whatever statistical tools are used and their validity, the results have real economic impact.
A conference on the Validation of Consumer Credit Models made the point.
Rather than establishing some arbitrary statistical criteria for a model's performance, the central question for validation is whether the model is working as intended and producing results that are at least as good as alternative approaches.The above is good advice for propensity models - be they acquisition or churn. We shouldn't talk about r-square, K-S, or goodness of fit but is our customer base growing the way we'd like?