Flip Flop: Why Zillow’s Algorithmic Home Buying Venture Imploded
Months before the company quit flipping houses, Stanford GSB researchers spelled out the risks of iBuying.
iBuyers’ need for speed may hinder their ability to evaluate homes carefully. | Illustration by Khyati Trehan
It wasn’t that long ago that iBuying seemed to have the potential to disrupt the residential real estate market and put fear into the hearts of flesh-and-blood realtors. Instead of enlisting an agent to put up yard signs and lead house tours, a homeowner could sell to an intermediating company that uses algorithms to determine the market value of homes. After answering a few questions on a website or smartphone app, the seller then waits for a cash offer to pop up in about 24 hours. Then the iBuyer lightly renovates the house and quickly flips it.
The new approach depends on scale and speed. iBuyers “argue they have a balance sheet and then they use technology to quickly acquire your home and then offer the home to owners who want to move,” explains Amit Seru, a professor of finance at Stanford Graduate School of Business. “The balance sheet allows them to own homes so that you can sell quickly and then they can wait and find a buyer.”
The iBuying trend got plenty of great press — a 2019 Wall Street Journal article glowingly described the potential of “high-tech flippers” to reshape the housing market with Wall-Street-style efficiency and Silicon Valley tech savvy. Rich Barton, the CEO of Zillow, one of the biggest iBuyers, predicted the “dawn of e-commerce for real estate.”
So it came as a shock to some when Zillow announced in early November that it was shutting down its iBuyer unit, Zillow Offers, and laying off a quarter of its employees. “We’ve determined the unpredictability in forecasting home prices far exceeds what we anticipated and continuing to scale Zillow Offers would result in too much earnings and balance-sheet volatility,” Barton explained in a news release.
Seru, however, wasn’t surprised at Zillow’s iBuying bust. Along with his Stanford GSB colleague, assistant professor of finance Greg Buchak, and colleagues Gregor Matvos of Northwestern University and Tomasz Piskorski of Columbia University, in December 2020 he’d co-authored a working paper for the National Bureau of Economic Research that explores the pitfalls of residential real estate intermediation and iBuying.
Looking at data from 2013 to 2018, the researchers describe a key trade-off that makes residential real estate intermediation difficult: To entice homeowners in need of liquidity to sell, companies that buy homes must move quickly while still accurately predicting the market value of a property so that they can make an adequate return when they sell. But that need for speed hinders their ability to evaluate homes carefully with traditional information-gathering techniques such as slow inspections and walkthroughs.
The risk with this strategy is that homes bought quickly by an iBuyer are likely to face problems such as major repairs while sitting on the intermediary’s balance sheet. This risk is exacerbated in markets with low liquidity, in which it’s difficult to sell a home quickly unless the price is significantly discounted.
Even for iBuyers, who rely upon technology to do quick transactions while limiting information loss, “intermediation is only profitable in the most liquid and easy to value houses,” the paper notes. “Therefore, iBuyers’ technology allows them to supply liquidity, but only in pockets where it is least valuable.”
As Seru explains, one of the big flaws of iBuying is that while buyers’ algorithms are adept at crunching numbers, it’s difficult for them to gather all the data that often influences what human buyers are willing to pay for a house. “There are a lot of features and attributes of a home that the model doesn’t capture,” he notes. Those 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 — the stuff that live buyers and agents look for when they tour houses.
As a result, Seru says, iBuyers are vulnerable to what’s known in economics as the Lemons Problem. “Suppose as a buyer you don’t really know the quality of my home — it could be a peach or a lemon. If you offer me a price that in an average price across lemons and peaches and I’ve got a peach home, I may think, ‘Why aren’t you offering me more? My home is amazing.’ The problem is that an iBuyer can’t see the quality because you’re doing the valuation so quickly.”
But while owners of peachier homes may not be so eager to sell to an iBuyer, someone with a lemon home — one with flaws that detract from its value — may be thrilled to get the “average price across peaches and lemons” generated by the algorithm, Seru explains. An iBuyer may therefore find that it faces adverse selection: Owners with lemons are more eager to sell to them, increasing its risk of overpaying for houses that it’s going to have trouble reselling without heavy discounts.
One way to avoid that problem, Seru says, is for iBuyers to stick to handling “cookie-cutter” houses in neighborhoods where it isn’t hard to figure out home prices using observable hard information. This would allow iBuyers to “avoid buying a home where the algorithm that uses this data to generate a home price would have a lot of error,” Seru says. While all algorithms and statistical models have errors when predicting future values, he explains that “such errors, on average, would wash out across a portfolio of cookie-cutter homes.”
In the five-year period that Seru and his colleagues examined, most iBuyer firms stuck to a one-size-fits-all formula. “It seemed that they understood and decided to stick with this strategy,” he says. Successful players such as Opendoor and Redfin, he says, were able to create a market, gain some market share, and turn a modest gross profit.
In Seru’s view, one of Zillow’s major mistakes was trying to capture market share by being more aggressive. “They were late entrants into this market and decided, among other things, to go into non-cookie-cutter homes, hoping their algorithmic valuation model was accurate,” he says. During the pandemic, traditional home listings increased, but Zillow found that speed wasn’t enough to beat them. “To compete, Zillow started bidding more for cookie-cutter homes than what their algorithmic model predicted for such homes.” All of this turned out to be a mistake. In early November, Zillow said it would be taking $569 million in write-downs — about $30,000 per home in its inventory.
The study also finds that, at present, iBuyers do not add much value when it comes to matching the homes bought by them to new owners. Seru says that iBuyers eventually may develop more sophisticated algorithms that can understand the psychology of different types of consumers and what features of homes appeal to them. “I think there’s a future where this can happen, where we’ll be able to do a good match,” he says. “But we’re not there yet.”
The challenges facing iBuyers provide a larger lesson, not just for real estate but other industries as well. Despite technological advances, Seru notes, it’s still important to appreciate how incentives in the marketplace work at a human level and how they impact models based on artificial intelligence and machine learning. “There’s a danger,” he says, in “getting too carried away by artificial intelligence and machine learning without understanding the underlying economics of the marketplace.”
For media inquiries, visit the Newsroom.