Zillow, the American real estate marketplace, said last Wednesday that it will pay $212.5 million (USD) to settle a class action lawsuit brought by its own shareholders over Zillow Offers, the home-flipping business it shut down in November 2021. Zillow Offers never operated in Canada, but I am writing about it because it is the clearest recent example of what happens when a very large, well-funded company decides it can run a local business from a head office, and because the way it failed says something useful about how any house gets priced.
Zillow App Icon
What Zillow Offers was
Zillow started the program in Phoenix in 2018, using the data behind its Zestimate to make cash offers, close within weeks, tidy the homes up and resell them. Sellers got certainty, investors were told that more housing data than anyone else could be turned into a margin on volume, and by the summer of 2021 Zillow was buying thousands of homes a month across the Sun Belt.
In early November 2021 word got out that most of the homes in Zillow's inventory were worth less than the company had paid and that it was trying to sell about 7,000 of them. Within days Zillow announced it was closing the division, writing down $304 million on homes bought that summer and letting go a quarter of its staff. The stock fell 32 per cent in two days, shareholders sued in Seattle that same month, and after the Supreme Court declined to hear Zillow's appeal in early October the company settled, admitting no wrongdoing, with insurance covering about $48 million and a judge still to approve the deal.
Zillow was not undone by a crash or by rates, because American prices were still rising when it folded. It was undone by being wrong on each house by a few percent, with the best-funded pricing model in the industry behind it. The company's own explanation at the time was "unpredictability in forecasting home prices," which is a careful way of saying that a model built in Seattle could not tell which house on a Phoenix street was worth more than its neighbour.
“unpredictability in forecasting home prices”
Carvana ran the same play with cars
Carvana, the American online used-car retailer, spent 2021 buying cars near the top of the market on the theory that a national algorithm and a large balance sheet could replace the local dealer, and when prices turned it was holding inventory it had overpaid for and about $5.7 billion in debt. The stock fell roughly 99 per cent from its August 2021 peak to under $4 by December 2022, about 4,000 people lost their jobs, and the company came within sight of bankruptcy before its bondholders agreed in 2023 to forgive more than a billion dollars of principal. Carvana survived, and joined the S&P 500 last December, so the comparison is not that these companies disappear, its that the model is challenging. The pattern is that the first thing they get wrong is the price of what they are buying, and the people paying for the mistake aren’t the ones who made it.
A used car is close to a commodity, with a wholesale auction price published every week, and if a national algorithm can misjudge a three-year-old Honda Civic badly enough to nearly sink a company, a house with a ravine lot and a finished basement of uncertain vintage is a far harder problem.
Why it never came to Canada
Properly, the Toronto company that promised to buy your house if it did not sell, pulled back from that guarantee when our market turned in 2022, and the large American iBuyers never crossed the border. The GTA has nothing like the thousands of near-identical houses that make Phoenix or Atlanta easy to model, and our offer-night way of selling does not translate. Another version will arrive the next time money is cheap, and the Zillow file is worth rereading when it does. A guaranteed price is only a gift if it sits above what the open market would pay, and the company offering it is betting that it does not.
Pricing a house here
Nearly every buyer I meet now arrives with an estimate from HouseSigma or Zoocasa, and so do I. They are good at the middle of the bell curve, where a townhouse in Wismer has forty near-identical neighbours. Where they struggle is where the money changes hands: the 60-foot lots of Markham Village, the backsplit on a ravine in Thornhill, the Unionville house with an addition that may or may not have a permit. The model does not know that the neighbour's sale dragging the estimate down had a wet basement on inspection, or that the one pulling it up went to the family next door who wanted the lot, and those details are what cost Zillow a few per cent on thousands of homes.
For a seller, the estimate on your phone is where the conversation starts. What closed on your street in the last sixty days, and why each one closed where it did, is where the price comes from, and that is most of what I do before I put a number on a listing. For a buyer, the same tools that call a house overpriced by four percent, were wrong by about that much on a few thousand homes in Phoenix. An automated-estimate will get you to the front door. What the house is worth comes from someone who knows the street, has stood inside the homes that sold, and can read where this market is heading, which is the part Zillow spent $212.5 million learning about the hard way.