Six Agencies Finalize Rules to Protect AI Real Estate Valuation Models — AI: The Washington Report

(co-author: Matthew Tikhonovsky)

On June 24, 2024, six federal agencies finalized regulations aimed at creating protections for automated valuation models (AVMs) in the real estate industry.
AVMs are becoming an increasingly popular tool, often based on artificial intelligence, used by mortgage lenders to estimate the value of properties for mortgage and lending purposes.
The new regulations require companies using AVMS to implement five quality control factors to ensure their models provide accurate, secure data and do not discriminate.
The regulations will enter into force one year after their publication in the Journal of Laws. Federal Register.

On June 24, 2024, six federal agencies—the Department of the Treasury, the Federal Reserve System (FRS), the Federal Deposit Insurance Corporation (FDIC), the National Credit Union Administration (NCUA), the Consumer Protection Financial Bureau (CPFB), and the Federal Housing Finance Agency (FHFA)—finalized regulations on quality control standards for automated valuation models (AVMs), an increasingly popular tool, often based on artificial intelligence, used in real estate valuations and mortgage lending processes.

The final regulations outline five quality control factors that firms using AVMs must adhere to to ensure that their models — and the property valuations and credit decisions they underpin — are neither inaccurate nor discriminatory.

Below we provide an overview of the AVM and the proposed regulations governing the AVM before moving on to the final AVM regulations.

AVM boom

In recent years, AI has impacted nearly every industry, and real estate is no exception. AVMs, which often rely on artificial intelligence and machine learning, estimate the value of properties for mortgages and loans. Widely considered more reliable than human appraisals, AVMs’ valuations and forecasts often form the basis of lending decisions and securitization arrangements, providing access to credit for first-time homebuyers and businesses alike.

The growing use of AVMs has been accelerated by advances in AI technologies, combined with the COVID-19 pandemic and the resulting shortage of mortgage appraisers, according to a report by the Brookings Institution. These models are now widely used in the real estate industry, including by the Federal National Mortgage Association (Fannie Mae) and the Federal Home Loan Mortgage Corporation (Freddie Mac), because they speed up and reduce the costs of home appraisals and mortgage lending. However, there are concerns about the potential discriminatory impact of AVMs on home appraisals and lending decisions. The Fair Housing Act (FHA) prohibits discrimination in the purchase and rental of homes, which includes mortgage lenders using AVMs. Studies have shown that AVMs can perpetuate the same biases against certain protected classes of homeowners that are often found in human appraisals, because AVMs ultimately rely on human-generated data that may contain the same biases, as well as historical data that may reflect past discrimination and housing disparities. While AVMs do not completely eliminate valuation errors, they have been shown to reduce valuation errors compared to human appraisals.

Proposed regulations

On June 1, 2023, six federal agencies issued proposed regulations regarding quality control standards for AVMs. The proposed regulations were promulgated in response to Section 1473(q) of the Dodd-Frank Act of 2010, which added a new section to the Financial Institutions Reform, Recovery, and Enforcement Act of 1989 and directed the six agencies to issue guidance requiring firms using AVMs to comply with four quality control factors. In addition to the four quality control factors in the Dodd-Frank Act, which focused on data accuracy and security, the June 2023 proposed regulations also introduced a fifth quality control factor, requiring AVMs to comply with applicable anti-discrimination laws.

The proposed regulations prefigure the Biden administration’s October 2023 Executive Order (EO) on AI, which “encourages” the Directors of the FHFA and CFPB to “consider using their authority” to “evaluate automated security valuation and appraisal processes in a manner that minimizes bias” in order to “(address) discrimination and bias against protected groups in the housing and consumer financial markets.”

Final version of the rule

With relatively few changes to the proposed rule, the final rule applies to mortgage originators and secondary market issuers that use AVMs for “certain credit decisions or securitization determinations.” The rule broadly defines an AVM as “any computer model used by mortgage originators and secondary market issuers to determine the value of a consumer’s principal residence that secures a mortgage loan.” The rule requires mortgage originators and secondary market issuers to “adopt policies, practices, procedures, and control systems to ensure” that their use of AVMs meets the following five quality control factors:

Ensure a high level of confidence in the estimates generated;
Protect yourself against data manipulation;
Try to avoid conflicts of interest;
Require random sample testing and reviews; and
Comply with applicable anti-discrimination laws.

Importantly, the rule does not establish specific “policies, practices, procedures, or control systems” that firms using AVMs must implement to comply with the rule. Instead, the rule leaves it up to these firms to develop their own policies and practices to ensure compliance with the rule’s five quality control factors. As modeling and artificial intelligence technologies advance, “this flexible approach will allow institutions to refine their implementation of the rule as needed,” according to the rule. Additionally, the rule provides mortgage originators and secondary market issuers with “flexibility to establish quality control standards for covered AVMs based on the size, complexity, and risk profile of their institutions.”

The regulations will enter into force one year after their publication in the Journal of Laws. Federal Register.

(Show source.)