Automated valuation models (AVMs) are used to estimate the market value of homes and play a key role in mortgage lending, in some cases even replacing human appraisers. They are also a central feature of online platforms such as Zillow and Redfin. Recently, their prevalence and influence have grown as the federal agencies and the government sponsored entities Fannie Mae and Freddie Mac have allowed more mortgage lending to rely on AVM estimates.
AVMs use a vast breadth of data, including both publicly available and user-provided data. AVM developers also leverage new algorithms, such as computer vision, to improve their models. Although their accuracy is at times an improvement over human appraisers, their performance is not consistent in communities of color and low-income neighborhoods. Therefore, the ever-expanding use of AVMs raises concerns around privacy, transparency, algorithmic bias, and the perpetuation of historic redlining.
Federal agencies have provided some regulation and guidance around the use of AVMs, and a five-agency collaboration proposed a new rule in 2023. AVMs are also covered by several existing anti-discrimination laws. However, this oversight is limited and should be expanded to include:
- Expanding public transparency;
- Disclosing more information to affected individuals;
- Guaranteeing evaluations are independent;
- Encouraging the search for less discriminatory AVMs;
- Releasing more government data on AVMs;
- Regulating platform AVMs; and
- Employing new forms of AVMs to counter historic redlining.
With the use of these models poised to grow in the coming years, these policy interventions will help ensure that AVMs advance a healthy and equitable housing market, rather than jeopardize it.
Related Content
Authors
-
Acknowledgements and disclosures
The authors would like to thank Michael Akinwumi, Alexei Alexandrov, Paul Bidanset, Eric Dunn, Talia Gillis, Logan Koepke, Brandon Lwowski, Michael Neal, Christopher Odinet, Valerie Schneider, David Silberman, Alicia Solow-Niederman, Morgan Williams, and Linna Zhu for their contributions and insight. All errors are the authors’ own.
