My take….
The two primary methods of consumer default/delinquency risk analysis are to produce a numerical score indicative of risk or to designate a category of risk. While FICO output produces a score, it is actually a mixture of these two methodologies.
There are models that do not even produce a numerical score, but rather divide the consumer base into categories, of which the consumer either is or is not a member.
The result could be a simply yes/no decision.
A simple example is a tree that first divides consumers into two groups… those with at least three active trade lines, and those with less than three. So-called thin or thick files.
Branching off of each would then traditionally be sub-categories of consumers with clean or dirty files, defined, for example, by the presence or absence of major derogs.
Further branching could then include those who have % util of revolving accounts less or greater than, for example, 40%.
And on and on, using each of the overall scoring categories, such as credit mix, length of credit, new credit, with the categories considered most significant being at the higher branches.
Once a consumer is traced through all of the branches in the model, a final determination of risk is made at the end twig. It could simply be a yes/no, lend or don’t lend, decision, or it could be a number designating degree rather than a simple yes/no. A FICO score, for example.
A category, or “bucket,” exists at each branching. Thus, these models don’t use one categorization, but rather a multitude of cascading categories, each with its own level of importance. I have yet to see a model where new credit is at the top of the tree.
The higher the category is on the tree, the greater is generally its importance in the final decision. Regardless of the major branch one is on, flowing into a dirty branch usually results in an unfavourable end number or decision.
