4.6 Use of Models

Effective from Jun 12 2006 - Dec 31 2007
To view other versions open the versions tab on the right

	Risk assessment techniques
4.6.1	There are generally two basic methods by which ratings are assigned: (i) a modelbased process; and (ii) an expert judgement-based process. The former is a mechanical process, relying primarily on quantitative techniques such as credit scoring/default probability models or specified objective financial analysis. The latter relies primarily on personal experience and subjective judgment of credit officers¹.
4.6.2	For IRB purposes, credit scoring models and other mechanical procedures are permissible as the primary or partial basis of rating assignments, and may play a role in the estimation of loss characteristics.
	Nevertheless, sufficient human judgment and oversight is necessary to ensure that all relevant and material information is taken into consideration and that the model is used appropriately.

	Requirements for using models
4.6.3	Banks should meet the following requirements for use of statistical models and other mechanical methods in rating assignments or in the estimation of PD, LGD or EAD:
	•	Banks should demonstrate that a model or procedure has good predictive power and its use will not result in distortion in regulatory capital requirements. The model should not have material biases. Its input variables should form a reasonable set of predictors and have explanatory capability.
	•	Banks should have in place a process for vetting data inputs into a statistical default or loss prediction model. This should include an assessment of data accuracy, completeness and appropriateness.
	•	The data used to build the model should be representative of the population of the bank’s actual borrowers or facilities.
	•	When model results are combined with human judgment, the judgment should take into account all relevant information not considered by the model. Banks should have written guidance describing how human judgment and model results are to be combined.
	•	Banks should have procedures for human review of model-based rating assignments. Such procedures should focus on finding and limiting errors associated with model weaknesses. Banks should have a regular cycle of model validation that includes monitoring of model performance and stability, review of model relationships, and testing of model outputs against outcomes (see section 5 of the ”Minimum Requirements for Risk Quantification under IRB Approach”).

¹ In practice, the distinction between the two is not precise. In many model-based processes, personal experience and subjective judgment play a role, at least in developing and implementing models, and in constructing their inputs. In some cases, models are used to provide a baseline rating that serves as the starting point in judgment-based processes.

Soft Launch