January 22, 2012
Credit Scoring: The Next Frontier
Posted by Kathleen Engel

Credit scoring has greatly reduced the cost of credit to the benefit of industry and borrowers, and has minimized concerns about intentionally discriminatory underwriting. Despite these gains, there remain questions about the integrity of the data used to determine borrowers’ scores and the fairness of the models used by credit reporting agencies (CRAs).  These concerns are amplified as credit scores take on increasingly important roles in the society.  Indeed, they have become a form of collateral.  In this post, we muse about areas in which credit scoring needs further investigation.

Credit scoring unquestionably predicts borrower creditworthiness; however, scores could be more accurate and, thus, more fair.  In particular, there is evidence that: (1) there are errors in the inputs on individual consumers; (2) some of the variables and the weights given to them are not predictive; and (3) models omit variables that would help predict borrower creditworthiness. For example, medical debt is often treated the same as credit card debt in scoring models.  As a result, borrowers with unexpected, delinquent medical debt will be “dinged” on their credit reports just as people who took on debt buying discretionary consumer goods.

The Consumer Financial Protection Bureau’s bailiwick includes the authority to write rules that would further the purposes of the Fair Credit Reporting Act. The CFPB is already collecting credit report information on 200,000 individuals from each of the three major CRAs for the purpose of analyzing variations between the scores sold to consumers and those sold to creditors (http://www.consumerfinance.gov/wp-content/uploads/2011/07/Report_20110719_CreditScores.pdf). These efforts could expand to include requiring that CRAs and entities, like FICO, that develop scoring models provide the CFPB with their algorithms, including the inputs and the weights given each variable.  This would enable the CFPB to test how well the CRAs predict default risk and the accuracy of their inputs.

The three national CRAs are not the only entities that collect and sell data on consumers.  Smaller enterprises collect discrete data on individual borrowers that are not necessarily captured in traditional credit scores.  Another role of the CFPB should be to identify these providers, evaluate their methods, and subject them to regulatory oversight.

There is a need to understand the market for the provision of accurate credit scores.  In a well-functioning market, you would expect that competition among CRAs would lead to ever more accurate credit scoring models.  However, if the marginal gains from: (1) including omitted, predictive variables, (2) insuring the accuracy of data with precision, and (3) scrutinizing weights, is small relative to the efficiency of slightly more crude scoring, CRAs and their clients might prefer the latter course.  The result would have a potentially adverse impact on borrowers who are at the cusp of creditworthiness, which would implicate fairness concerns.   

With lenders increasingly cautious about granting credit to people with less than pristine credit scores, there is a need to survey and evaluate models other than traditional scoring.  This should include approaches taken in other countries, with an emphasis on programs that help low-income borrowers build credit and demonstrate creditworthiness.

I am sure there are other areas in which more understanding is needed and hope people will comment on this post so I can expand my catalog.

Stay tuned: Suffolk Law School Law Review will have a special issue on credit scoring and reporting later this year. (http://www.law.suffolk.edu/highlights/stuorgs/lawreview/index.cfm)


Finance, Financial Institutions, Masters: Dodd-Frank, Roundtable: Banking | Bookmark

TrackBacks (0)

TrackBack URL for this entry:

Links to weblogs that reference Credit Scoring: The Next Frontier:

Recent Comments
Popular Threads
Search The Glom
The Glom on Twitter
Archives by Topic
Archives by Date
January 2019
Sun Mon Tue Wed Thu Fri Sat
    1 2 3 4 5
6 7 8 9 10 11 12
13 14 15 16 17 18 19
20 21 22 23 24 25 26
27 28 29 30 31    
Miscellaneous Links