A credit model is an analytical method that models the credit default process based on available information used to determine the probability of default on a credit arrangement (loan, lease), with a high score indicating a low probability and a low score indicating a high probability of default.  Credit models are judgmental, statistical or a combination of the two (hybrid) and are developed solely using an institution’s internal data or rely on historical credit data from third-party providers and market variables.

The judgmental credit model relies on human expertise and quantifies an institution’s credit policy, market knowledge, risk preferences and segment strategy and is used when little or no historical data are available, such as when entering a completely new sector.  For a judgmental credit model, credit risk factors are selected, combined and weighted and the model is tested on historic, hypothetical or new cases –  this generally requires a significant investment of time and input from senior management.  Judgmental credit models combine and evaluate all key underwriting criteria such that the credit decisions generally agree with those of the credit officers.

A statistical credit model analyses extensive data on a credit applicant (borrower, lessee) in a target sector and scores the degree of the probability of default on the transaction or concludes that the transaction is not likely to be sufficiently profitable to be acceptable.  A statistical model can be developed only when an institution has collected a substantial amount of historical credit data on both the product and problematic clients.  Where there is sufficient historical credit data, use of the statistical credit model is preferable because it can also quantify the unacceptability of a credit exposure.  Popular statistical techniques include decision trees, artificial neural networks and logistic regression.

A statistically-derived credit model augmented by judgmentally weighted variables for data that are not available or not collected historically is a hybrid credit model.  A hybrid credit model allows incorporation of new risk factors and is used when not quite enough historical credit data are available for a statistical credit model or additional factors should be inCommerciald into a scorecard for a new target client or segment.

Credit Models
Judgemental Hybrid Statistical
Select Risk Factors Define Bad Loans
Weigh Risk Factors Identify Significant Risk Factors from Portfolio Data
Test Scorecard Determine Optimal Model
Perform Business Check of Significant Risk Factors
Identify Additional Factors
Weigh Additional Factors
Test and Validate Model on New Cases

 

Related Items