study finds models can compensate for lost public records data

Study finds models can compensate for lost public records data

Date: June 24, 2020

The most recent VantageScore white paper touches on a topic that has received a fair amount of mainstream media attention lately: Potential increases in some consumer credit scores due to the anticipated reductions in the number of tax liens and other public records reported in consumer credit files.

The white paper, “Negative Data Suppression and the Impacts on Credit Score Models,” addresses the impact that the loss of those records may have on the accuracy of credit scoring models and demonstrates that smart model design can ameliorate any potential increases in credit scores.

The change in credit-file recordkeeping is a result of the national credit reporting companies’ National Consumer Assistance Plan (NCAP), an initiative that seeks to make credit-data reporting more accurate and understandable for consumers. Changes under NCAP include adoption of stricter requirements for verifying the accuracy of tax liens and other public records—a measure that will effectively invalidate and force the deletion of most of the records from consumer credit files.

VantageScore conducted its study to better understand how the loss of that data will impact the predictiveness of credit scoring models and develop strategies that model developers might use to effectively compensate for the loss of those records.

Using two million anonymized consumer credit files and a methodology detailed in the white paper, VantageScore developed two scoring models for comparison—one based on credit files containing all the negative data NCAP may suppress, and the other based on credit files from which all NCAP-related data had been removed.

A comparison of scores obtained by separately scoring the same population using each of those models revealed the following:

  • Despite the predictive value of the negative data that will be suppressed under NCAP, there are techniques that will enable credit score models to recapture predictive performance in their absence.
  • In fact, by adopting new variables more closely aligned to current credit-usage behavior—as opposed to the “backward-looking” historical events that NCAP would suppress—models developed in the absence of NCAP data may prove more stable than older models.
  • Consumers who scored above 660 using model designs that accommodate NCAP data suppression exhibited better credit-management habits and resulted in more stable loan portfolios than their counterparts who had been scored with models built using the negative file data. (This should reassure lenders who are concerned about “score inflation” that these higher scores do in fact accurately reflect less-risky consumer habits.)

Regulatory focus on eliminating negative data under NCAP, along with the findings outlined above, suggests lenders should evaluate incumbent scoring models to ensure those models will continue to perform well in the absence of NCAP-related data.

The white paper is available for download at