What is an Automated Valuation Model (“AVM”) and do we have to validate it?
Posted in Credit Risk, Insights, Model Validations, MRM
AVMs are a regulatory hot topic at the moment. It likely these are considered “models” per regulatory guidance meaning users will have to validate any AVM in use.
To provide some background, an AVM is an Automated Valuation Model, and per Fannie Mae, AVMs are “statistically based computer programs that use real estate information, such as comparable sales, property characteristics, tax assessments, and price trends, to provide an estimate of value for a specific property.” (Fannie Mae Sellers Guide, 8/2/23) AVMs are getting a lot of attention lately because of dialog and guidance from regulators and GSEs about appraisal bias (fair lending) and valuation modernization. (see example: https://capitalmarkets.fanniemae.com/mortgage-backed-securities/single-family-mbs/updates-regarding-valuation-modernization)
Given the attention on AVMs in the mortgage industry, the joint regulators (OCC, FDIC, Federal Reserve, NCUA, CFPB, and FHFA) issued a Notice of Proposed Rulemaking (NPR) and request for comment on June 21, 2023. The document presents the history of the disparate expectations among regulators regarding validations for AVMs. Historically, the NCUA hasn’t been a party to the Model Risk Management guidance issued by the OCC, FDIC, and FRB in 2011/2017. (OCC Bulletin 2011–12 4/4/2011; FRB SR Letter 11–7 4/4/2011; FDIC FIL–22–2017 6/7/2017). The NCUA does, however, join its peers in this NPR. Comments on the NPR were due by August 21, 2023.
Proposed Requirements
What this NPR requires of mortgage originators and secondary market participants (covered institutions) is a set of Quality Controls described as policies, practices, procedures, and control systems to ensure that AVMs used in connection with making credit decisions or covered securitization determinations regarding a mortgage. The intent is to ensure a high level of confidence in the estimates produced by AVMs; protect against the manipulation of data; seek to avoid conflicts of interest; require random sample testing and reviews; and comply with applicable nondiscrimination laws. Although institutions would have flexibility regarding which quality controls they would implement, this proposed rule has more teeth than mere guidance in that the proposed rule would be binding, and noncompliance would result in enforcement action.
Validation a requirement?
Risk Managers at covered institutions need to be cognizant of the possibility of enforcement action when designing their system of quality controls over AVMs. This means if institutions choose to not have a model validation performed, they will be expected to defend that decision. Given the focus lately on artificial intelligence reliability, appraisal bias, and discrimination, the decision to not perform a validation of the AVM might be impossible to defend. And if the reason for not performing a validation is “we really don’t know how this model works because the vendor can’t or won’t tell us,” that is akin to asking for a Cease and Desist letter.
The path forward
The professionals at VBC strongly encourage covered institutions to at least document their AVM model and prepare for a validation over the next year. While we are still waiting on a final rule, it is unlikely to read “disregard the need for a validation.”
What's the potential cost of not leveraging the experience, tools, and talent VBC brings to the table?
