Regulating AI & Predictive Models : A Look at a Finalized CFPB Rule Imposing New Obligations for AVMs
DOEREN MAYHEW
Automated valuation models ( AVMs ) have become increasingly common in real estate lending . These computer models provide insight for buyers , sellers and lenders alike in situations that may not require a full appraisal . As the models underlying the AVM grow seemingly more intricate and complex with the inclusion of additional variables , the Consumer Financial Protection Bureau ( CFPB ) has notably drawn parallels between AMVs and artificial intelligence ( AI ), which has consequently attracted regulatory interest .
Like the current regulatory issues surrounding AI , the CFPB is concerned unregulated AVMs will produce property estimates reflecting discriminatory bias through replicating systematic inaccuracies and historical patterns of discrimination .
In response to these concerns , the CFPB has published a new rule coinciding with its effort to ensure the appraisal system is credible , accurate , nondiscriminatory and free of any conflicts of interest . By imposing several quality control standards AVMs must adhere to , the forthcoming rule will have pertinent implications in the real estate / mortgage industry .
Definition of AVM
The CFPB , along with other prudential regulators , amended title XI of the Financial Institutions Reform , Recovery , and Enforcement Act of 1989 ( FIRREA ), and added Section 1125 . Section 1125 directs federal agencies to implement quality control standards for AVMs used in valuing real estate collateral securing mortgage loans . Section 1125 further defines an AVM to encompass any computerized / automated model “ used by mortgage originators and secondary market issuers to determine the collateral worth of a mortgage secured by a consumer ’ s principal dwelling .” The agencies further clarified their definition of AVM by explaining a model is characterized by these three criteria from the Interagency Supervisory Guidance on Model Risk Management :
1 . An information input component , which delivers assumptions and data to the model .
2 . A processing component , which transforms inputs into estimates .
3 . A reporting component , which translates the estimates into useful business information .
Continued on page 4
VIEWPOINTS : REGULATORY COMPLIANCE EDITION | 03