The median sale price of an existing home rose to a record $350,300 last month, according to the latest report from the National Association of Realtors. Sales of existing homes fell in May for the fourth straight month as competition for the few homes available continued to push up prices.
Last week, Zillow announced an update to its tool for estimating home values, saying the changes allow its algorithm to “react more quickly to market trends.”
Norm Miller is a professor of real estate finance at the University of San Diego and a consultant to companies that create these kinds of automated valuation models, or AVMs, for banks and the government. He said a lot of the data comes from the multiple listing services that brokers use to share information about homes for sale. The following is an edited transcript of our conversation.
Norm Miller: It used to be that the multiple listing service data was reported monthly and then weekly. And right now, it’s reported daily. So, we are getting more towards real-time data, and that’s probably been the biggest improvement in evaluations and in the models, especially in a quickly changing marketplace. In terms of machine learning, we have a long ways to go. It turns out to be labor-intensive. It turns out to be hard to do it well. I don’t want to give the impression that the research is done yet, but the data just keeps improving.
Amy Scott: So a lot of our listeners are probably familiar with the Zestimate, which is Zillow’s proprietary model, but I understand it’s not just real estate companies that use them, but banks and regulators too. Who is using these?
Miller: Pretty much all your banks and thrift institutions that make mortgage loans will use AVM. Sometimes it’s the only value estimate. The Treasury uses them. Fannie Mae and Freddie Mac, [which] buy most of the mortgages in the country, use AVMs, and mortgage investors, investment bankers use them, so they’re quite heavily used in the business-to-business context as well as business-to-consumer context.
Scott: How does the lack of supply and the rapidly increasing prices in many markets affect the accuracy of these models and the stakes for getting it right?
Miller: When prices change rapidly, whether it’s up or down, appraisal models, both manual appraisals and AVMs, tend to be less accurate, and it is harder to appraise. Now, having said that, AVMs do a pretty good job because they can consider not just past sales, but current listing prices, search activity and other factors like that. But it is harder in a rapidly changing market, and you’re going to have more error in a market like today than you would in a market that’s sort of steady clipping along at a 2% or 3% appreciation rate, which might be historically more typical than today.
Scott: There’s been a lot of concern recently about the appraisal gap in terms of Black-owned homes appraising at lower rates than white-owned homes were in predominantly white neighborhoods. And I’m wondering if that shows up in these automated valuation models as well. Or is this potentially a way that we could address those gaps?
Miller: Well, there’s two things to keep in mind. One is that AVMs are agnostic with respect to race, ethnicity. They don’t know it. They don’t care. It’s not in the models at all. Even unconsciously, they can’t consider it. And so, there’s less likely to be any unconscious bias, as you might with an appraiser. We can also test for disparate impact on different racial groups, and we have done that. I’ve been involved in research on that topic, and we can look to see if there’s any bias and sometimes there is, but again, generally the AVMs are not biased. And it’s one of the reasons why some of the regulators are thinking about relying more on AVMs because of the fact that they’re racially and ethnically neutral with respect to who’s in the household.
Scott: But they can perpetuate bias in terms of the baked-in bias that’s in some of the data that goes into these models, right? Comparable sales, for example. If you’re comparing homes in a Black, disinvested neighborhood, it’s going to be lower than a white neighborhood that’s more affluent.
Miller: Well, you’re right that if the comparable sales are lower-priced, and if that was baked in because of past market behavior, then the AVM will interpret those as comparable properties without regard to why they’re lower. And so you’re right, it might be baked in.
Scott: Do you see a future in which we won’t have manual appraisals?
Miller: Yes. I tell my real estate students to go into commercial appraisal, if they want to go into appraisal, or if they go into residential, go into unique properties. Because the run-of-the-mill, typical tract home or condo can be appraised so well by an AVM, there’s no reason to compete with it. Is this displacing appraisers? Yes it is, and we are not seeing a lot of people enter residential appraisal as a result of it, or if they do it, they have to enter on the technical side and understand statistics and AVM models as well. So yes, it’s causing displacement just like, you know, so many other technologies have displaced people.
Related links: More insight from Marketplace’s Amy Scott
Speaking of racial disparities in home appraisals, reporter Natalie Moore did a deep dive for WBEZ. A version of her story also aired on Marketplace. In Chicago, researchers compared houses with similar features in neighborhoods with similar socioeconomic characteristics. They found that houses in predominantly Black and Latinx areas appraised for $324,000 less, on average, those those in white neighborhoods. That gap had increased more than sixfold since 1980.
Earlier this month, the Joe Biden administration announced the broad strokes of a plan to address inequity in home appraisals through potential enforcement of fair-housing laws and regulatory action.
A couple of years ago, I reported on the obsession among some homeowners, and aspiring homeowners, with online valuation tools. As prices have marched mostly up over the past decade, it’s been tempting to check your own Zestimate like a 401(k) — as a measure of wealth. But owner beware. After its recent update, Zillow said its median error rate for homes that aren’t on the market is about 7%. Meaning that half the time, the actual sale price falls within plus or minus 7% of the Zestimate. For the other half, the gap is larger.
As I found out, real estate agents have sort of a love-hate relationship with these online estimates because their clients sometimes end up with unrealistic expectations about what their homes are worth.
Just for fun, I checked my house on Redfin, Zillow, Chase and Realtor.com and got wildly different estimates — as far as $40,000 apart.
The future of this podcast starts with you.
Every day, Molly Wood and the “Tech” team demystify the digital economy with stories that explore more than just “Big Tech.” We’re committed to covering topics that matter to you and the world around us, diving deep into how technology intersects with climate change, inequity, and disinformation.
As part of a nonprofit newsroom, we’re counting on listeners like you to keep this public service paywall-free and available to all.