The Banking industry shifted dramatically in 2004, when the first Basel II accord was published. Banks changed their focus from creating more credit scores to modeling probability of default (PDs), which traditional scores were unable to accurately predict. Recently, bankers are focusing on scores that are based on duration models, survival models, Cox proportional hazard models, and age-period-cohort (APC) models. How does that affect companies that buy scores from vendors?
Vendor scores traditionally provide a risk ranking and leave pricing (and implicitly estimating probabilities) to the lender. If you’re not building scores, chances are you’re not building a probability model either and are just setting cutoff scores to manage originations.
Today, that would be called standard practice, but certainly not leading practice. To stay in the risk-based pricing game, you will need a probability model. Whether you hire a statistician to make a probability model wrapped around a vendor score or buy a probability model from a vendor, here are some questions you can ask:
- Is the model built to consider when the loan defaults? Defaults for young loans are much more costly than defaults for old loans.
- Does the model consider economic conditions when the loan defaulted? A default during a good economy is not the same as a default during a bad economy.
- How is a lifetime-loss forecast created? Which economic scenario is used? Loan pricing for lifetime losses cannot be based solely on current economic conditions.
- Is the model specific to the product being offered? A product-independent bureau score is useable as input to a probability model, but the probability of default is very specific to the product, features, and terms offered.
The above is based on an excerpt from The RMA Journal, February 2014 article “Pricing Loans with Real PD Forecasts” by Joe Breeden, Ph.D., founder and CEO of Prescient Models LLC.