New Federal Guidelines for AVMs Elevate Bias Testing, Give Consumers a Stronger Voice–ICE’s Damien Weldon and John Holbrook

(From left to right: John Holbrook and Damien Weldon, courtesy of ICE)

Damien Weldon is the Vice President of Collateral and Valuation Solutions at ICE where he oversees their product management and data science for U.S. mortgage and real estate. A seasoned practitioner, Damien has over 25 years’ experience in quantitative solution development for risk management and asset valuation.

John Holbrook is the Vice President of Digital Valuation Solutions at ICE. He started his career in the valuation and mortgage industry as a real estate appraiser in 1995 and is a seasoned leader and strategist, developing and implementing innovative products and strategies for property valuation and data. He is passionate about using digital twins and property data to enhance the transparency and credibility of valuation processes and products.

Decades after the first fair lending laws started to address discrimination in residential lending, regulators are now laser-focused on eliminating bias in the real estate valuation process.

In today’s environment, it’s important to have a transparent, objective and credible valuation process to establish a value for a property that all parties can agree on–but also a way to challenge and test those value assessments when warranted.

The Evolution of AVMs and Regulatory Oversight

With advancements in technology and greater access to deeper insights about the condition of a property and its surrounding community, the traditionally manual and time-consuming property appraisal process has seen many improvements. As automated valuation models (AVMs) became available and more widely accepted over the years, federal regulators began to scrutinize their use. As a result, the Dodd-Frank Act (Dodd-Frank Wall Street Reform and Consumer Protection Act) of 2010, which amended the Financial Institutions Reform, Recovery and Enforcement Act of 1989 (FIRREA), codified four AVM quality control standards:

• Accuracy
• Data integrity
• Objectivity
• Random testing and review

In 2021, an interagency task force created by the Biden Administration released the Action Plan to Advance Property Appraisal and Valuation Equity (PAVE). Following that effort, the OCC, the Federal Reserve Board of Governors, FDIC, NCUA, CFPB and FHFA published a proposed rule to implement the quality control standards mandated by Dodd-Frank for the use of AVMs by mortgage originators and secondary market issuers. The proposed rule scopes using AVMs to determine the value of the real property being utilized as the collateral for a mortgage transaction.

The New Rule and Its Implications

On June 30, the final rule was published by the individual federal agencies, and its subsequent publication in the Federal Register started the clock on the rule’s implementation. The rule directs institutions that engage in certain credit decisions or securitization determinations to:

• Adopt policies, practices, procedures, and control systems to ensure that AVMs used in these transactions to determine the value of mortgage collateral adhere to quality control standards designed to ensure a high level of confidence in the estimates produced by AVM
• Protect against the manipulation of data
• Seek to avoid conflicts of interest
• Require random sample testing and reviews
• Comply with applicable nondiscrimination laws

The rule goes into effect on Oct. 1, 2025. Here are key takeaways on what the new standards could mean for lenders, service providers and consumers.

The Core Four

Although recent regulatory attention has been focused on preventing and removing bias, the new rule underlines the continuing importance of credibility and transparency in AVMs.

Credibility is achieved when the valuation process is independent, impartial, objective and sufficient. There has to be enough data to make a valuation valid.

Transparency is crucial when it comes to sharing and explaining valuation methodologies. While the underlying statistical algorithms are the intellectual property of AVM providers, this should not be a black box process. AVM modeling processes, sample testing and lending applications needs to be understandable, well-documented, explainable, consistent and robust.

Above all, AVMs must continue to adhere to the four core quality control standards outlined in Dodd-Frank:

• Confidence (accuracy measured quantitatively)–A good AVM includes a methodology for conveying confidence in estimates of value. This refers to statistical confidence as expressed in confidence intervals and forecast standard deviations (FSDs).
• No manipulation of data (data integrity)–The whole point of an AVM is to ward off the inclusion or induction of human bias. The data upon which valuations are based should be kept clean and protected as a single source of truth.
• No conflicts of interest (objectivity)–Although less of a concern for computers than with humans, integrity and transparency require an objective process free of bias or any interests running counter to the establishment of a fair and unbiased estimate of value.
• QC sampling should be random (random testing and review)–Developing and using AVMs require continuous testing and calibration to ensure that AVM models function as intended.

Addressing Bias: the Fifth Factor

In addition to increased accuracy and efficiency, one of the major value-adds for AVMs, initially, was the idea that machine-based valuation algorithms could not be biased because machines are not subject to human emotions or preconceptions.

However, research by the Brookings Institute implies that while machines may not be biased, the geospatial or defined neighborhood boundaries used to filter sales data could conceivably produce biased conclusions as a result of historical redlining. 

To avoid perpetuating historical biases, AVM users need to understand how AVM algorithms determine values. Particular attention should be paid to the selection of comparables, with an eye toward unintentional redlining. A good AVM tool should be able to quantify the risk of bias and offer an array of comparables that help mitigate that risk.

This can be accomplished by looking at subject properties by census tract and testing for differences from benchmarks, isolating value characteristics and cross-referencing against race and ethnicity census data to identify value variances not directly attributable to other valuation factors. Another option is the use of computer vision to objectively score a property’s condition and quality based on objective analysis of the photographs of the subject and comparable properties. This would provide additional granularity to comparables selection based on condition and quality variances between properties.

Giving Homebuyers a Voice With ROVs

In July 2024, the federal regulatory agencies issued final guidance on the Reconsideration of Value (ROV) of residential real estate valuations.

There are many reasons why a home may be worth more than the value reflected in an appraisal report. Sometimes an appraiser may not be aware of a transaction, or another factor that could affect the price. However, the difference is often a matter of opinion, in which a seller might attribute a higher value to a particular feature or improvement that is not supported by other home sales transactions in their local market.

In such cases, it can be beneficial to let the buyer or seller tell their side of the story. You might initiate a dialogue by providing the public record to the buyer and/or seller so they can determine if you have the facts correct. Ask what they think the property is worth but be sure to also ask why they feel that way. There are all kinds of reasons their perception of value may vary from what was in an appraisal, so nailing down specific differences is important.

Appraisers are charged with being independent, impartial and objective. Homeowners tend to be more vested in their personal, subjective views of a property’s finishes and decorative details that may not translate to a higher market value. Thanks to the increased use of computer vision, it’s now possible to get an objective third assessment by having homeowners submit pictures of key areas – kitchen, living room, primary bedroom and primary bathroom – as well as any recent updates and upgrades. Computer vision matches the pictures to similar features in previously benchmarked homes and provides condition and quality scores that can be helpful in resolving valuation differences.

Giving the buyer or seller a stronger voice using photographs, automation and collaborative tools to resolve differences is more thorough and more efficient than traditional processes. It is also a less expensive alternative to ordering a second appraisal or incidentally violating appraisal independence requirements. Under the new federal rule, it’s also imperative.

What’s Next

Lenders should work with their AVM providers to ensure their tools are up to the task and able to evaluate properties with an eye towards intentional or unintentional bias. Before the new AVM standards are implemented in 2025, lenders will need a way to demonstrate that their AVMs are monitoring for potential bias and producing valuations that are not only justifiable but also fair and equitable.

(Views expressed in this article do not necessarily reflect policies of the Mortgage Bankers Association, nor do they connote an MBA endorsement of a specific company, product or service. MBA NewsLink welcomes submissions from member firms. Inquiries can be sent to Editor Michael Tucker or Editorial Manager Anneliese Mahoney.)