On June 24, the OCC, Federal Reserve Board, FDIC, NCUA, CFPB, and FHFA approved a final rule to implement the quality control standards mandated by the Dodd-Frank Act to address the use of artificial intelligence in estimating home values. The Bureau noted that over the years, the real estate and mortgage industry have made use of computer models to estimate a property’s value. As these AI-based models grow in complexity and include more variables, they can produce inaccurate or discriminatory results.

The final rule introduces comprehensive regulations for Automated Valuation Models (AVMs) which are sophisticated computer programs designed to appraise property values. Commonly known as algorithmic appraisal tools, these AVMs employ statistical modeling techniques to deliver estimates for commercial and residential properties. The rule addresses several of the CFPB’s key concerns about the accuracy and fairness of AVMs as they begin to incorporate more complex algorithms and artificial intelligence. The rule requires:

• High confidence in estimates. AVMs must provide reliable property valuations.
• Data protection. Safeguards must be in place to prevent data manipulation.
• Conflict of interest avoidance. Measures to avoid conflicts of interest in AVM use must be in place.
• Random sample testing. Regular testing and review of AVM outputs must be implemented.
• Nondiscrimination compliance. AVMs must comply with existing nondiscrimination laws to prevent bias.

The rule applies to mortgage originators and secondary market issuers who use AVMs in making credit decisions. This includes decisions about originating, modifying, or terminating mortgages but excludes AVMs used solely for portfolio monitoring. The rule also covers AVMs used by servicers working on behalf of mortgage originators or secondary market issuers in loan modifications.

The rule will take effect approximately one year after all federal agencies provide their final approval.

Putting it into Practice: Appraisal bias has been a longstanding concern of federal regulators (see here and here). The expanded use of algorithmic technology in lending models has only exacerbated those concerns. Companies that use such models in originating and credit decision-making should pay close attention but also be aware that there is likely more rulemaking to come.

Brandon Mohamad & Alyanna Gallo, law clerks not yet admitted to practice, also contributed to this article.

Listen to this post