OCC Issues Guidance On Credit Scoring

The use of credit scoring is growing rapidly. Not only is credit scoring becoming more common, the ways in which lenders can use credit scoring are increasing. This expansion has prompted the OCC to issue guidance on the use of credit scoring. See OCC Bulletin 97-24.

Advantages
The bulletin acknowledges the many ways that credit scoring can be used to reduce risk, manage credit losses, improve application processing time, and ensure compliance with Regulation B. However, the OCC is calling attention to risks that can result from the way in which credit scoring systems are developed, used, and evaluated.

Risks
Examiners have reported problems with credit scoring systems and their use that have caused the development of this bulletin. The problems are not specific to national banks - all banks using credit scoring systems or credit scores should be aware of the problems and develop procedures to prevent or minimize the problems. Problems identified include:
Inadequate training for the use and monitoring of the system,
Deficiencies in management information systems (MIS),
Improper use of credit scoring systems, and
Use of credit scoring systems that may contain factors and weights that violate Regulation B.

Role Of Management
The bulletin specifically places responsibility for the effective and proper use of credit scoring on bank management. The bulletin states that bank management must be an active participant in the development and use of credit scoring models. Specifically, management should:
Understand the credit scoring models thoroughly,
Use credit scoring models for their intended purpose,
Validate or revalidate the systems regularly,
Take appropriate action when the model's performance deteriorates, and
Ensure compliance with ECOA and other fair lending laws.
Evaluate The System

Perhaps the most valuable advice in the bulletin are the recommendations for tracking the function and impact of the system. A fundamental procedure is to regularly test the model's performance and compare this to the expected performance. Differences probably mean that something isn't working. This testing should include the initial use of the system, and use in special promotions or unique circumstances.

Testing means identifying the applicants that the system accepts - and tracking them to determine how they perform.

The purpose of a credit scoring system is to use statistics to measure the similarity of applicants to previous good borrowers or previous poor credit customers. The system should help the lender distinguish between applicants likely to become good customers and those likely to default. The better a system does this, the better the system is performing. Measuring of the effectiveness of the system uses credit performance information on those approved by the system. Unfortunately, this measurement is an after-the-fact analysis, for example, after the system has accepted problem borrowers. For safety and soundness reasons, the bank should perform this results analysis with fair frequency. The OCC suggests that this be done at least annually and that for high volume lenders, the analysis should be more frequent.

Internal Support
The bulletin also calls banks' attention to the need for strong MIS support, not only for the system itself, but for obtaining and using sufficient information about the applicants and borrowers to enable the bank to evaluate and redevelop the system.

These concerns apply to the use of scores developed by third parties, such as a consumer reporting bureau. These scores are developed without attention to the lender's specific applicant base. Moreover, most of these scores are based on regional or national data. When using credit scores developed using the applicant's credit history, the lender should track the system's decisions to measure the utility of the score.

Monitoring
OCC recommends a series of monitoring reports. These reports should be designed to measure the decisions of the system and look for changes in the bank's market and products. Market shifts may include the arrival or exit of a significant employer, crop successes or failures. Other economic shifts, such as inflation trends and interest rate trends, are not market specific but can affect the working of the credit scoring system. Finally, the model may deteriorate over time or be used for credit products that were not part of the model's development.

Stop Overrides
Finally, watch any overrides. If the system is properly developed, it should be more effective than judgment. Loan officers are often reluctant to rely on a credit scoring system. They are used to trusting their instincts. However, overrides can lead to problems in two important areas: model validity, and discrimination. Overrides either invalidate the model or correct for a model's weakness. If the system works, there should be no need for overrides. But even riskier is the possibility that the "instincts" of the loan officer may involve discrimination. The result may be disparate even though not intended. For example, the loan officer may know or feel he knows the 37-year-old white male applicant and therefore override the system's denial. But will that loan officer do the same for a 64-year old Hispanic female applicant with the same score?

ACTION STEPS

While working on Year 2000, consider how to capture and store information about applicants that is useful or necessary in developing, evaluating, and validating credit scoring systems.
If you use credit scoring - including those scores you buy from the consumer reporting agencies - look at the customer information you will need to evaluate the system.
Also review the volume of applications and credit approvals to decide how frequently to review the system's effectiveness. The higher the application and decision rate, the more frequently the system should be reviewed.
Meet with the operations staff to work out a plan for identifying credit problems that were applications approved by the model. Reports on delinquent borrowers may be the most important information you review.
Remind loan officers about the risks of overrides.
Set up a system for tracking all overrides by department and loan officer. Watch for a pattern of loan officers that don't "trust" the system. If they really do know better than the system, the system doesn't work!