iBuyers, AI, and Real Estate

The residential real estate market in the U.S. has been beset for decades with official and de facto examples of discrimination aimed at non-white potential homeowners and tenants: the effects of policies and customs such as deed restrictions, housing covenants, redlining, and exclusionary zoning, as well as urban renewal projects that tore the fabric of largely minority neighborhoods asunder, are still manifest throughout the nation’s cities.

Despite the official outlawing of explicit forms of racial discrimination in housing policy, the legacy of old laws and customs continues to affect those who live, or want to live, in certain neighborhoods or under certain circumstances. These effects often manifest themselves in cases where a homeowner needs to sell a house quickly due to dire financial need, or perhaps does not have time to thoroughly prepare a house for a sale. These properties are often traded quickly via a cash offer that ends up being quite a bit lower than a traditional brokered sale might bring.

Surprisingly, though, recent research into improving equitability in these markets has produced some counterintuitive results.

A team of researchers from the University of Washington’s information school, for example, discovered that instant buyers, or iBuyers—companies that rapidly buy and sell homes using computational models called automated valuation models (AVMs) to set prices—did indeed create more equitable markets for white and Black home sellers. However, they did not create those conditions by offering Black owners more; they offered white owners substantially less than they would have received in a private transaction.

“I wasn’t totally sure what to expect going into the project,” said University of Washington study lead author and Ph.D. student Isaac Slaughter. “There is a hypothesis on one side that racial discrimination in real estate markets in the U.S. has been so pervasive, and the consequences still very present today, that maybe you would expect that these models that are trained on a bunch of historical data would learn that discrimination as well.”

Slaughter and his colleagues found that, in their control group (individuals selling to other individuals), Black sellers were expected to earn $36,000 less on a home than a white seller. “So there’s that discrimination in the market in general,” Slaughter said.

The team analyzed 50,000 publicly available property transfer records from 2018 to 2023 for Mecklenburg County, NC, controlling for 50 factors including home size and neighborhood crime rate, and discovered an unexpected result: near-equity in sales price was achieved by iBuyers paying Black homeowners $4,376 more on average, while paying white homeowners $27,239 less than they would have gotten from a private sale through a traditional broker.

The team presented its findings in June at the ACM Conference on Fairness, Accountability, and Transparency (FAccT), held in Rio de Janeiro.

A reduction in sale price such as Slaughter and his colleagues found is a significant drag on sellers’ ability to create a more secure financial future, not just for themselves, but for their children.

“In the absence of the very wealthy, the family house is the single largest investment and the single largest transfer of generational wealth,” said Romi Mahajan, president of boutique marketing consultancy KKM Group and a regular contributor to several real estate technology content platforms. 

Flipping to corporate ownership

The researchers also discovered that iBuyers were selling the homes they purchased to institutional investors, who then rent those dwellings rather than selling them to private individuals, at a substantially higher percentage than similar transactions between individual sellers and institutional buyers. When iBuyers sold homes in Mecklenburg, institutions—frequently real estate investment trusts—bought 25% of the homes. Yet when an individual sold homes, institutions bought just 15% of them.

“These real estate investment trusts tend to look for cheap homes that they can buy and convert to rentals so that they can profit over decades,” the study’s senior author, Nic Weber, said in a release announcing the study’s publication.

Just as Mahajan observed lower sales prices reduce a family’s generational wealth, Weber said turning a home into a trust-controlled rental is just as troubling, “because in the U.S. one of the substantial ways that people gain wealth and transfer it between generations is through home ownership.”

On the surface, then, one might expect that the story behind iBuyers, institutional investors who ultimately buy houses to rent them out, and the automated valuation models upon which purchase and sale prices are based are a prime example of concentration of resources and a generational “rich getting richer” scenario.

As the University of Washington researchers said, “Ultimately, our analysis suggests that iBuyers are equalizing housing outcomes by extending real estate harms typically isolated to Black homeowners to white homeowners as well.”

Conversely, one might also think that access to the inner workings of a real estate AVM could help researchers and policymakers discover where existing biases skew value downward and encourage market players to craft new algorithms to reduce them. That knowledge shared among the community could encourage policymakers to create laws and regulations that, while maybe not absolutely ensuring fairness, could certainly help level the playing field.

As an example, on July 1 the Consumer Financial Protection Bureau, the primary federal watchdog of the U.S. consumer finance industry, approved a new rule to regulate the use of AVMs affecting the mortgage industry. The new rule, set to take effect on July 1, 2025, requires covered businesses to enact policies, procedures, practices, and control systems designed to: (1) ensure a high level of confidence in the estimates produced; (2) protect against the manipulation of data; (3) avoid conflicts of interest; (4) require random sample testing and reviews; and (5) comply with applicable nondiscrimination laws.

However, some experienced industry observers note that the proprietary nature of commercial AVMs, and their uneven performance, make any sort of universal evaluation that might enable coherent fair housing policy in quick-sell cases, difficult. And the speed of housing market dynamics, such as mortgage interest rate fluctuations, and seismic waves of activity such as the rush to buy homes outside major urban areas during the COVID pandemic, often make the data available obsolete before it can be useful to model.

Spectacular failure

Perhaps the most prominent example of a failed algorithmic approach to quickly turning over a property, or “flipping” a house, was real estate technology giant Zillow’s attempt at the practice; it exited iBuying and laid off nearly a quarter of its workforce in 2021 after an estimated $1-billion loss.

Mahajan said Zillow’s failure illustrates a larger issue with AVMs.

“I fundamentally believe that AVMs are hard to do well,” Mahajan said. “I’m not saying it’s a trivial thing to have 140 million houses for sale and to run those numbers every day. But, using my own house, for example, if you put it up on Zillow and compared it on Redfin, there’s a difference of a couple hundred thousand dollars in their estimates. I’m not saying who’s right and who’s wrong, but clearly both can’t be right.”

Amit Seru, a Stanford University finance professor, co-authored a working paper for the National Bureau of Economic Research in which the difficulties of accurately pricing homes via algorithm were outlined.

Seru told Insights from Stanford Business there are numerous features and attributes of a home the model doesn’t capture, which might range from how buyers react to a house’s architectural style to local noise levels and whether the neighbors take good care of their lawns.

Problems: Long-term intangibles and political will

The unknown, perhaps unknowable, effects of the iBuying/institutional ownership trend is how ancillary factors may or may not prove beneficial to occupants of those properties. In some instances, the timeline to gauge those factors is presently too long to study.

One example is research by Joshua Coven, a Ph.D student at New York University. In a nationwide analysis, Coven discovered that, while institutional ownership decreased the housing available for owner-occupancy by the 30% of the homes they converted, the institutional investors also increased the supply of homes available for renter occupancy by 69%, and lowered rents slightly. The increase in the supply of rental housing allowed the financially-constrained to move into neighborhoods that previously had few rental units. The new renters were also likely to be moving into a “better” neighborhood, according to data Coven analyzed.

“After controlling for new tract unobservables, we can see that those who moved into buy-to-rent homes came from tracts with 0.98% lower middle school test scores, 1.1% lower median household incomes, areas where people are historically 2.34% less likely to end up in the highest income quintile, 4.56% more likely to end up in jail, and 0.87% less likely to end up married by age 32,” Coven found.

But exactly how long a family might have to stay in that rental unit to allow for analysis as to whether or not it has long-term benefits is still unknown.

“That would be an extremely important margin to see if these families actually benefit from renting in theses areas,” Coven told Communications. “But the investors know the families want these better school areas.”

Mahajan said he is not certain even the best research can substantially change the overall dynamics of the nation’s housing market, for good or ill.

“I don’t think policy can ever keep up with such a fast-changing market,” he said. “The question really is, why is the market so fast-changing? How can you have a $40-trillion asset class that is moving so quickly? It has to be because of institutions and speculation and people buying second houses; all these things that don’t really have a lot to do with getting families in homes.”

Mahajan said he can think of 10 to 12 research teams doing what he called “amazing work” crunching housing market data, but that the political will to do anything is largely missing among policymakers.

“There has to be the will to then do what is necessary, which includes things like changing zoning laws, ensuring no redlining exists, and understanding the existing data has redlining and bias built into it. There also has to be pressure on private employers to pay more so people can buy houses. There is a lot of stuff that has to be done that is structural, things I think both major U.S. political parties give lip service to but don’t do anything about.”

Gregory Goth is an Oakville, CT-based writer who specializes in science and technology.