ICYMI: At Hearing, Senator Warren Calls on Housing Regulators to Combat Algorithmic Bias That Contribute to the Racial Wealth Gap
“Without proper guardrails and oversight, AVMs may reinforce racial disparities in the appraisal process and actually worsen the racial wealth gap.”
Washington, D.C. – In case you missed it, yesterday during a hearing of the Senate Banking, Housing, and Urban Affairs (BHUA) Committee, United States Senator Elizabeth Warren (D-Mass.) highlighted the role that automated valuation models (AVMs) – computer programs that use data to estimate home values – play in perpetuating racial disparities in the home appraisal process. While these systems could be a tool to reduce bias in appraisals, they often rely on data that reflect decades of racist federal housing policies. Senator Warren is fighting to ensure proper regulation is in place to end the perpeturation of racial discrimination in the housing process and close the racial wealth gap.
In response to Senator Warren, Melody Taylor, the Executive Director of the Property Appraisal and Valuation Equity (PAVE) Interagency Task Force and the Regional Director of the Office of Fair Housing and Equal Opportunity (HUD), confirmed that characteristics of majority-Black neighborhoods brought about by decades of racist housing policies – including higher shares of distressed home sales, lower household incomes, and higher levels of gentrification relative to majority-white neighborhoods – may lead AVMs to produce larger appraisal errors in majority-Black neighborhoods and to perpetuate racial disparities in valuation.
Today’s hearing builds on Senator Warren’s concerns about algorithmic bias disproportionately affecting communities of color in the financial sector, health care systems, and education. Most recently, she demanded answers about Wells Fargo's discriminatory mortgage-refinancing practices against Black homeowners.
Transcript: Strengthening Oversight and Equity in
the Appraisal Process
U.S. Senate Banking, Housing, and Urban Affairs Committee
Thursday, March 24, 2022
Senator Elizabeth Warren: So appraisals are powerfully important. For example, they determine how much a lender will offer in financing a mortgage and that in turn affects the price that a seller can get. If homes are accurately appraised, then the process works like it should. But Black and Hispanic homeowners, however, have consistently faced discrimination in the appraisal process. According to a recent study, homes in majority of Black neighborhoods are valued 23% lower than homes with similar characteristics in White majority neighborhoods, resulting in more than $150 billion in lost wealth for Black families. Because homeownership is the number one way that middle-class families build wealth, the disparities in home valuations help perpetuate the racial wealth gap.
Now, research shows that the persistent undervaluation of Black and Hispanic-owned homes is due in part to unconscious bias among appraisers. So it’s not surprising that automated valuation models. AVMs – computer programs that use data to estimate home values – have been held up as an alternative. The argument goes that by taking the human element out of the appraisal process, we can eliminate bias and discrimination, which sounds terrific. But the reality of these algorithms is a whole lot more complicated.
Ms. Taylor, you lead the Biden administration’s interagency task force to root out racial discrimination in the appraisal process. Have AVMs produced the unbiased, error-free valuations that home lenders and homeowners were counting on?
Melody Taylor: Thank you for that question, Senator Warren. We looked at the Urban Institutes study which found that AVMs in majority-Black neighborhoods produce larger errors relative to the underlying sales price than AVMs in majority-white neighborhoods, potentially contributing to the wide housing wealth gap between Black and white homeowners.
Additionally, we believe that, you know, while AVMs have the potential if properly used to reduce human bias and improve consistency in decision-making; however, just like human biased appraisals, they’re not immune from the risk of discrimination. A phrase that, often times, many people use: data, bad data in equals bad outcomes. And we believe that we have to ensure that AVMs are responsible, or the models are responsibly created.
Senator Warren: So I want to underscore the point you’re making because I think it’s a very good point. The key issue is that AVMs are only as good as the data they use and the algorithms that interpret those data. And too often, these algorithms are just a black box, with little visibility into the data and the modeling that are going on inside. And if the underlying data itself reflect past racial discrimination, AVMs may end up compounding disparities in valuation rather than reducing them.
So let me just ask you a couple of pieces about this so we can disaggregate this a little bit. Ms. Taylor, majority-Black neighborhoods tend to have a higher share of distressed home sales than majority-white neighborhoods. Is that fact likely to contribute to AVMs systematically undervaluing Black-owned homes?
Taylor: Yes, ma’am.
Senator Warren: Majority-Black neighborhoods also have lower household incomes on average. So Ms. Taylor, is that likely to contribute to AVMs undervaluing Black-owned homes?
Taylor: Yes, ma’am.
Senator Warren: And what about if neighborhoods are gentrifying. Is that likely to contribute to the undervaluing of Black-owned homes?
Taylor: Yes, ma’am.
Senator Warren: Okay. Thank you. You know, we’ve pulled apart some of the pieces. The way I see this is that closing the racial wealth gap means that we need to confront decades of racist federal housing policies like redlining that helped create it in the first place. And part of this means ensuring that the legacies of racial discrimination aren’t coded into the algorithms that have direct and real impact on families’ wealth. Without proper guardrails and oversight, AVMs may reinforce racial disparities in the appraisal process and actually worsen the racial wealth gap. That’s why I was very glad to see the CFPB take a step toward issuing rules on combatting algorithmic bias in AVMs. Regulators should follow their lead and take the necessary steps to make sure that AVMs aren’t putting homeownership and wealth-building even further out of reach for families of color.
Thank you very much for your work. Thank you for being here today. And thank you, Mr. Chairman, for having this hearing.
###
Next Article Previous Article
