Pursuant to section 1473(q) of the Dodd-Frank Act, a group of federal agencies have proposed a quality control rule for automated valuation models (AVMs). The agencies are the Comptroller of the Currency, Consumer Financial Protection Bureau (CFPB), Federal Deposit Insurance Corporation, Federal Housing Finance Agency, Federal Reserve Board and National Credit Union Administration. Comments on the proposed rule will be due 60 days after publication in the Federal Register. The agencies propose that the final rule would be effective the first day of a calendar quarter following 12 months after publication of the rule in the Federal Register.

Under the proposed rule, an AVM would be defined as “any computerized model used by mortgage originators and secondary market issuers to determine the value of a consumer’s principal dwelling collateralizing a mortgage.” The proposed rule would apply regardless of whether the credit was for consumer purposes or for business purposes. The proposed rule would incorporate the definition of “mortgage originator” from section 103 of the Truth in Lending Act (TILA). Subject to exceptions, the section provides that the “term “mortgage originator”-

(A) means any person who, for direct or indirect compensation or gain, or in the expectation of direct or indirect compensation or gain-

(i) takes a residential mortgage loan application;

(ii) assists a consumer in obtaining or applying to obtain a residential mortgage loan; or

(iii) offers or negotiates terms of a residential mortgage loan;

(B) includes any person who represents to the public, through advertising or other means of communicating or providing information (including the use of business cards, stationery, brochures, signs, rate lists, or other promotional items), that such person can or will provide any of the services or perform any of the activities described in subparagraph (A).”

The proposed rule would define a “secondary market issuer” as “any party that creates, structures, or organizes a mortgage-backed securities transaction.”

The proposed rule would apply to the use of an AVM by a mortgage originator or secondary market issuer “in determining the value of collateral in connection with making a credit decision or covered securitization determination regarding a mortgage or mortgage-backed security.” A “credit decision” would be defined as “a decision regarding whether and under what terms to originate, modify, terminate, or make other changes to a mortgage, including a decision whether to extend new or additional credit or change the credit limit on a line of credit.” A “covered securitization determination” would be defined as “a determination regarding: (A) Whether to waive an appraisal requirement for a mortgage origination in connection with its potential sale or transfer to a secondary market issuer; or (B) Structuring, preparing disclosures for, or marketing initial offerings of mortgage-backed securitizations.”

The proposed rule would require that mortgage originators and secondary market issuers that engage in credit decisions or covered securitization determinations themselves, or through or in cooperation with a third-party or affiliate, adopt and maintain policies, practices, procedures, and control systems to ensure that AVMs used in these transactions adhere to quality control standards designed to:

(i) Ensure a high level of confidence in the estimates produced;

(ii) Protect against the manipulation of data;

(iii) Avoid conflicts of interest;

(iv) Require random sample testing and reviews; and

(v) Comply with applicable nondiscrimination laws.

For purposes of the requirements, “control systems” would be defined as “the functions (such as internal and external audits, risk review, quality control, and quality assurance) and information systems that are used to measure performance, make decisions about risk, and assess the effectiveness of processes and personnel, including with respect to compliance with statutes and regulations.” Addressing the proposed requirements in the preamble to the proposed rule, the agencies state that “[t]his approach would allow mortgage originators and secondary market issuers the flexibility to set their quality control standards for covered AVMs as appropriate based on the size of their institution and the risk and complexity of transactions for which they will use covered AVMs.”

The agencies also advise in the preamble that they “considered whether to propose more prescriptive requirements for the use of AVMs and decided not to do so. Different policies, practices, procedures, and control systems may be appropriate for institutions with different business models and risk profiles, and a more prescriptive rule could unduly restrict institutions’ efforts to set their risk management practices accordingly.”

The fifth factor for the quality control standards—compliance with applicable nondiscrimination laws—is not specified in the Dodd-Frank Act, and the agencies added the factor based on their authority under the Act to account for any other factor that the agencies determine to be appropriate. Addressing the potential for bias in AVMs, the agencies state in the preamble that “[a]s with models more generally, there are increasing concerns about the potential for AVMs to produce property estimates that reflect discriminatory bias, such as by replicating systemic inaccuracies and historical patterns of discrimination. Models could discriminate because of the data used or other aspects of a model’s development, design, implementation, or use. Attention to data is particularly important to ensure that AVMs do not rely on data that incorporate potential bias and create discrimination risks. Because AVMs arguably involve less human discretion than appraisals, AVMs have the potential to reduce human biases. Yet without adequate attention to ensuring compliance with Federal nondiscrimination laws, AVMs also have the potential to introduce discrimination risks.” (Footnote omitted.)

The proposed rule would not apply to the use of AVMs in (i) monitoring of the quality or performance of mortgages or mortgage-backed securities, (ii) reviews of the quality of already completed determinations of the value of collateral, or (iii) the development of an appraisal by a certified or licensed appraiser.

In the preamble, the agencies address the common situation of Fannie Mae or Freddie Mac granting appraisal waivers:

“[When Fannie Mae and Freddie Mac (the GSEs)] use AVMs to determine whether the mortgage originator’s estimated collateral value or the contract price meets acceptable thresholds for issuing an appraisal waiver offer, the GSEs would be making a “covered securitization determination” under the proposed rule. As a result, the proposed rule would require the GSEs, as secondary market issuers, to maintain policies, practices, procedures, and control systems designed to ensure that their use of such AVMs adheres to the rule’s quality control standards. On the other hand, when a mortgage originator submits a loan to determine whether a GSE will offer an appraisal waiver, the mortgage originator would not be making a “covered securitization determination” under the proposed rule because the GSE would be using its AVM to make the appraisal waiver decision in this context. As a result, the mortgage originator would not be responsible for ensuring that the GSEs’ AVMs comply with the proposed rule’s quality control standards.”

Addressing the proposed rule, CFPB Director Chopra continued his theme of asserting that automated systems may have baked in biases:

“Algorithmic appraisals that use so-called automated valuation models can be used as a check on a human appraiser or in place of an appraisal. Unlike an appraisal or broker price opinion, where an individual person looks at the property and assesses the comparability of other sales, automated valuations rely on mathematical formulas and number crunching machines to produce an estimate of value.

While machines crunching numbers might seem capable of taking human bias out of the equation, they can’t. Based on the data they are fed and the algorithms they use, automated models can embed the very human bias they are meant to correct. And the design and development of the models and algorithms can reflect the biases and blind spots of the developers. Indeed, automated valuation models can make bias harder to eradicate in home valuations because the algorithms used cloak the biased inputs and design in a false mantle of objectivity.”

The Director also stated that “[e]merging AI-marketed technologies can negatively impact civil rights, fair competition, and consumer protection. Because technology has the power to transform our lives, we must ensure that AI-marketed technologies do not become an excuse for evasion of the law and consumer harm. It is critical that these emerging technologies comply with the law.”

The Director did not offer any examples of situations where AVMs were found to have inherent biases.