Fair Lending Implications of Credit Scoring Systems

PDF version of this article

During the past decade, lenders' use of credit scoring systems has increased significantly, and examiners routinely consider the role of credit scores in lending decisions. The comparative analysis portion of a fair lending examination includes an interview to determine the criteria the lender considered in the decision point (underwriting, pricing, etc.) selected for review.¹ This interview frequently reveals that a credit score was one of the criteria. At this point, examiners can determine how to proceed by consulting the Interagency Fair Lending Examination Procedures.² However, examiners must synthesize information from several sections of the Procedures and the appendixes.

This article gives examiners the tools they will need to navigate this situation. It provides an overview of credit scoring systems, analyzes why the use of credit scores has proliferated, and explains how their use is considered as part of a fair lending examination. It then recommends a concise conceptual framework for proceeding with a fair lending examination when a credit score is one of the criteria considered by the lender.

An Overview of Credit Scoring Systems

A credit scoring system mechanically evaluates creditworthiness on the basis of key attributes of the applicant and aspects of the transaction.³ A system can be as simple as a form the loan officer completes by hand that assigns points to particular attributes, or as complex as an artificial intelligence-based neural network with a continuous feedback loop that adjusts the weighting coefficients and the cutoff score. A credit scoring system can be the only factor considered in making the credit decision, or the lender may combine a credit score with other criteria.⁴

Two types of credit scores exist—bureau scores and custom scores. A bureau score considers only the information on an individual's credit report and is generated by a consumer reporting agency. The largest three consumer reporting agencies are Experian, Equifax, and TransUnion. A lender pays the consumer reporting agency an additional fee to obtain the score at the time it obtains a copy of the credit report. An "acceptable" score varies with the lender's appetite for risk; however, an acceptable score usually falls around 600.

A custom score (sometimes referred to as an application score) is generated by the lender from a scoring system either developed by the lender or purchased from a vendor. A custom score usually considers the information on the applicant's credit report, selected information about the applicant, and characteristics of the credit transaction. Examples of commonly considered applicant information are type of residence, length of time at current residence, type of employment, length of time in current employment, and income. Typically, eight to ten variables are considered in a custom scorecard. Many custom scoring systems are scaled so that an acceptable score will be around 200, again depending on the lender's risk appetite. The scaling of custom credit scoring systems varies considerably among vendors and lenders. Some lenders blend an applicant's bureau score and a custom score in making a credit decision.

The Use of Credit Scoring Systems Has Increased Considerably

The accuracy and cost of credit scoring systems have benefited substantially from technological advances in automated data processing and improvements in statistical methodologies. Many lenders have found that credit scoring systems are cutting the time and administrative costs of making credit decisions, as well as improving the consistency of the decisions within their organizations. As a result:

• More lenders are using credit scoring systems.
• Lenders are applying credit scoring systems to more credit products.
• Lenders are using credit scoring systems in additional aspects of credit transactions, such as pricing and account administration.
• Lenders are using multiple systems in a single credit product.⁵

The increased use of credit scoring systems has implications for examiners as they conduct fair lending examinations.

The Role of Credit Scoring Systems in a Fair Lending Examination

A fair lending examination attempts to detect either overt discrimination or disparate treatment on a prohibited basis. Examiners select a focal point based on the risk that discrimination may be occurring, determine the criteria the lender considers in making the credit decision, evaluate the criteria and procedures for overt discrimination, and compare how the criteria are applied to a selected prohibited basis group with how they are applied to an appropriate control group. For example, the treatment of Hispanic applicants may be compared with the treatment of non-Hispanic whites.⁶

The use of a fairly developed and applied credit scoring system can reduce the possibility of unlawful discrimination by helping to ensure consistency and uniformity and minimizing individual judgment and discretion. However, a credit scoring system is not a panacea, and in certain circumstances, it can even be the source of fair lending violations.

Disparate treatment can occur at three stages in the use of a custom credit scoring system:

• Data development and input: For example, a lender credits white applicants with the length of time they have worked in the same field but credits Hispanic applicants only with the length of time they have worked for their present employer. Or, a lender credits white applicants with secondary income (such as bonuses, overtime, or commissions) but credits Hispanic applicants only with base salary. In either example, because discriminatory data are input into the system, the system will produce a discriminatory result.
• Within the credit scoring system: The system could include a prohibited basis as one of the variables, or, if not a prohibited basis itself, a factor that is so highly correlated with a prohibited basis that it serves as a proxy for the basis. (As discussed later in this article, in certain circumstances age can be considered in a credit scoring system.) A variable that considers the geographic area in which an applicant lives should be carefully scrutinized to determine if the geographic distinctions are so highly correlated with a prohibited basis that they serve as a proxy for that basis. In 2001, the Department of Justice (DOJ) settled a case against Associates National Bank in which the bank required a higher cutoff score for applicants who applied on Spanish-language applications than it required of applicants who applied on English-language applications.⁷ DOJ treated the Spanish-language application as a proxy for ethnicity.⁸
• Discretionary overrides: The more discretion bank staff is permitted in overriding a credit scoring system, and the greater the number of staff with override authority, the greater the risk that the discretion will be exercised discriminatorily. Discretionary overrides fall into two categories. Low-side overrides are decisions to approve an applicant whose credit score falls below the cutoff score, and high-side overrides are decisions to deny an applicant whose credit score exceeds the cutoff score. The two types of overrides should be independently analyzed to detect an overall pattern of disparate treatment. This type of violation is illustrated by a settlement agreement between DOJ and Deposit Guaranty National Bank in 1999. The bank used a custom scorecard to underwrite applications for home improvement loans, but gave broad discretion to loan officers to override the credit scoring system. The pattern of overrides showed that white applicants were significantly more likely than black applicants to be approved with a credit score below the cutoff, and black applicants were significantly more likely than white applicants to be denied with a credit score above the cutoff.⁹

How should a fair lending examination be conducted once an examiner learns that a credit score is one of the criteria used in making a credit decision? Initially, the examiner should determine if the credit score is a bureau score or a custom score. If it is a bureau score, the examiner does not need to obtain more information about the scoring system. The comparative analysis should focus on the pattern of overrides and the lender's consideration of other criteria unrelated to the system. It is rare for a bureau score to be the only criterion considered in making a credit decision.

However, if the credit score is a custom score, the examiner should obtain a list of the variables considered by the system and determine if the scoring system is split into multiple cards on the basis of age. If a prohibited basis other than age, or a possible proxy for a prohibited basis other than age, is contained in the variables, the examiner should report this information to his or her manager as soon as possible. Addressing the overt discrimination issue will consume significant resources; therefore, the examiner should also consult with the manager about whether to continue with the planned comparative analysis.

As mentioned previously, age is the only prohibited basis that legally can be considered in a credit scoring system. Age is not a prohibited basis under the Fair Housing Act, and the Equal Credit Opportunity Act and Regulation B provide a narrow exception for the consideration of age if the system meets certain requirements.

It is preferable from a risk management standpoint for a lender to validate every credit scoring system used to underwrite or price loans. However, from a compliance standpoint, a credit scoring system does not have to be validated unless it considers age. A credit scoring system can consider age in one of two ways: (1) the system can be split into different scorecards depending on the age of the applicant or (2) age may be directly scored as a variable. Some systems may consider age in both ways. Regulation B requires that all credit scoring systems that consider age be validated. The regulation uses the term "empirically derived, demonstrably and statistically sound."¹⁰ For purposes of this article, we will refer to this term as "valid."

The burden is on the lender to demonstrate that a credit scoring system that considers age is valid for each credit product for which it is being used. An initial validation and periodic revalidations must occur to allow the scoring system to consider age.¹¹ Generally, a lender must validate a credit scoring system based on data from the institution's own through-the-door applicant population. However, if the lender's data are insufficient for an initial validation, the lender is permitted to obtain a validated scoring system or the data from which to develop a validated system from another lender or lenders for use on an interim basis. A lender must validate and revalidate its system based on its own data when they become available.¹²

Age-Split Systems

The system is treated as considering, but not scoring, age if it is split into only two cards, neither of which contains age as a variable, and one card covers a wide age range that encompasses elderly applicants. (Elderly applicants are applicants 62 years of age or older.)¹³ Typically, the younger card in an age-split system is used for applicants under a specific age between 25 and 30. The younger scorecard de-emphasizes certain factors, such as the number of accounts on the applicant's credit history, the age of the oldest account on the applicant's credit history, length of employment, and length of time at present residence, but increases the negative weight of any derogatory information on the credit report. Validation is the only requirement Regulation B imposes on a system that considers, but does not score, age.¹⁴

Systems that Score Age

A system is treated as scoring age if age is directly scored as a variable, regardless of whether the system is also age-split, or if elderly applicants are included in a card with a narrow age range in an age-split system. Regulation B imposes a second requirement on scoring systems that score age-the age of an elderly applicant must not be assigned a negative factor or value.¹⁶

The next steps in the fair lending examination framework flow from these requirements. If a custom scoring system considers age, the examiner should obtain the lender's documentation on the initial validation and all periodic revalidations, including the weighting coefficients. At the FDIC, the documentation is then submitted to the Washington Office through regional management for expert review. The examiner should then complete the comparative analysis considering whether there are indications of disparate treatment in either the development and input of the applicant data, low- and high-side overrides of the system, or both.

In summary, based on an understanding of the different types of credit scoring systems and the Regulation B requirements for scoring systems that consider age, the framework in the shaded box is recommended for conducting a fair lending comparative analysis of credit decisions in which one of the criteria considered is a credit score.

Benefits of Using This Framework

This conceptual framework is recommended as an aid in conducting efficient fair lending examinations that result in correct, legally supportable conclusions. Applying the framework does the following:

• Assists in focusing the review for overt discrimination and disparate treatment only on those areas in which it possibly exists;
• Ensures that the requirements of Regulation B for validation and treatment of the elderly are considered only for the small minority of credit scoring systems to which they apply;
• Ensures that lenders that choose to use custom credit scoring systems that consider age comply with the rigorous requirements for the narrow exception to the general prohibition against age discrimination; and
• Ensures that validation documentation is reviewed by FDIC staff with the appropriate, highly specialized expertise.

R. Russell Bailey
Senior Fair Lending Specialist