I do not want to say that losing a lot of money on a trade is as good as making a lot of money on a trade. It is not. For one thing, if you make a lot of money on a trade, then you have money, which can be used to buy goods and services. For another thing, if you make a lot of money on a trade, then you were at least in some rough sense correct about the trade, and being correct often about big questions is a valuable skill in finance and life.
If you lose a lot of money on a trade, neither of those things are true. And yet there is a certain prestige to it? Being correct is not the whole ballgame. Being important is important. “Important people like to deal with important people,” the Goldman Sachs commandment goes; “are you one?” The most famous JPMorgan Chase & Co. trader of the last decade, the one JPMorgan trader whom many people know by name or at least by nickname, is Bruno Iksil, the London Whale, who is famous for losing $6.2 billion of the bank’s money. Did that work out great for him? No, not really, not at all. But I suspect a lot of people in finance would buy him a drink if they ran into him. He is a celebrity in a way that thousands of traders who have never lost $6 billion are not.
Also, as I have written before, it is a rule of thumb in high finance that losing a lot of money on a trade can be good for your career: It shows that someone has trusted you to take risk, and that you took risk, and while it didn’t work out you probably learned from your mistakes. That may not work at the celebrity level, but if you were just around the London Whale losses you probably have some good war stories that any interviewer will appreciate.
Anyway I predict a bright future for the house traders who lost $300 million trading houses for Zillow and got not only themselves but 25% of the company fired:
Zillow Group Inc. is pulling the plug on its tech-powered home-flipping operation, after an ambitious effort to transform the company collapsed when its vaunted pricing algorithms proved unequal to the task.
The company plans to take writedowns of as much as $569 million and reduce its workforce by 25% as it winds down the business in coming months, according to a statement Tuesday. Zillow shares plunged as much as 11% to $76.22 in late trading.
The decision to abandon home flipping comes as the company’s third-quarter results showed it lost more than $380 million in the operation, called Zillow Offers. The business hit a major snag in recent months as Zillow tweaked its algorithms to make more aggressive offers, causing it to overpay for houses just as the heated U.S. market began to cool slightly.
With the company’s losses mounting, Chief Executive Officer Rich Barton said it had become too risky to scale the business in a U.S. housing market that has been running hot for well over a year during the pandemic.
“Fundamentally, we have been unable to predict future pricing of homes to a level of accuracy that makes this a safe business to be in,” Barton said on an earnings call.
I say “traders” because that is sort of the conventional way to describe people who lose hundreds of millions of dollars of their employers’ capital doing trades, but I suppose the more correct description is, like, “algorithm designers.” Zillow did not hire a team of house traders to make gut-instinct-based calls on housing valuation; it hired a team of programmers to build a house-buying algorithm. And they did, and it was cool, but it lost $300 million last quarter and is now getting shut down.
Also I should say that Zillow’s traders (or rather its algorithm) are not the sorts of traders who are supposed to lose hundreds of millions of dollars. They were not entrusted to take big risks, to swing for the fences; they did not aim for a massive profit, miss, and land on a massive loss instead. They were market makers; they were supposed to buy at the bid, sell at the offer, collect a spread for providing liquidity and make a steady profit. Here’s how Barton put it on the earnings call yesterday[1]:
When we decided to take a big swing on Zillow offers 3.5 years ago, our aim was to become a market maker not a market risk taker. And this was underpinned by the need to forecast the price of homes accurately three to six months in the future.
We used historical data and countless simulations to test this belief. We set unit economics targets that required us to stay within plus or minus 200 basis points in breakeven, holding ourselves accountable to these levels publicly with you all.
The central problem is that those first two sentences sort of contradict each other. A market maker is someone who buys and sells an asset in order to profit from the spread, not someone who accurately forecasts the price of an asset six months from now. End users want to buy or sell stocks or bonds or houses, they want to do it quickly at a predictable price, so they go to a market maker who will provide that service. The market maker buys from sellers and sells from buyers and does its best to match them up; ideally it buys an asset from a seller and resells it to a buyer within a fairly short time. It collects a “spread” from the buyer and seller: It buys from the buyer at a bit less than the fair market price, and sells to the seller at a bit more than the fair market price, because it is providing them a valuable service, the service of “immediacy” or “liquidity,” the service of always being available to buy or sell.
In pure theory the market maker makes all of its money from the spread; it is so perfectly hedged, or turns over its inventory so frequently, that it doesn’t care if the prices of stocks or bonds or houses go up or down. In practice this is impossible and all market makers have some amount of inventory risk; at any moment they are long or short some amount of assets, and if prices move they will make or lose money. Still, to the extent you are a pure market maker, you try to minimize that.[2] In the stock market, high-frequency electronic market makers really do turn over their inventory so often that they can reliably collect spreads without worrying too much about price movements. Famously Virtu Financial Inc., one of those market makers, can go years without a down day: It makes money (from spreads) every single day, whether the stock market goes up or down. It is trading stocks all day, but in some important sense it is not betting on stock prices.
But in the house business you can’t generally buy a house in the morning and sell it in the afternoon. You sign a contract to buy a house in the morning, then you do an inspection and title search and stuff, then a few weeks later you close on the house and deliver the money, then you spruce up the house a bit, then you wait for a buyer to come in — which takes, not seconds as it does in the stock market, but days or weeks or months — then you show the house to the buyer, then you sign a contract to sell it, then they do an inspection and title search and stuff, then you wait around for them to get a mortgage, then a few months later you close on the sale.
And meanwhile the price of houses has gone up or down, and the effect of that dwarfs the effect of your spread. This past quarter Zillow wrote down its inventory of houses by $304 million, to $3.8 billion, a loss of something like 750 basis points, way worse than the 200 basis point spread it was targeting.[3] But the quarter before that — April through June of 2021 — it had a gain of 576 basis points. “Clearly some portion of the holding costs and a smaller portion of the renovation costs likely benefited from the strong housing market,” Barton said at the time: House prices went up, so Zillow, which owned a bunch of houses, made more money than it planned to. As Barton said on yesterday’s call, both results are a problem:
We've been unable to accurately forecast future home prices at different times in both directions by much more than we modeled as possible. With Zillow Offers unit economics on a quarterly basis swinging from plus 576 basis points in Q2 to an expected minus 502 to minus 700 basis points in Q4. Put simply our observed error rate has been far more volatile than we ever expected possible. And makes us look far more like a leveraged housing trader than the market maker we set out to be.
But that’s sort of inherent in the model. The model is “predict where housing prices will be in three to six months, buy houses for prices that will be profitable if those predictions are right, and oh yeah probably get some extra juice for providing liquidity to buyers and sellers.” As opposed to the standard high-frequency market-maker model in the stock market, which is something close to “buy at the bid, sell at the ask, and do both of those things so quickly that the stock price doesn’t have time to move.”
Now this does not mean that Zillow’s business is a bad business. If you build an algorithm that is really good at predicting house prices six months out, you can probably make a lot of money buying and reselling houses. (Frankly, if you build an algorithm that is pretty good at estimating current house prices, and you have a lot of capital, you can probably do okay buying and reselling houses, because most of the time house prices seem to go up. You will have some bad quarters though.) It does mean, though, that it’s a volatile, risky, capital-intensive “leveraged housing trader” business, not a market-making one.
I have been writing about Zillow Offers a lot over the past few weeks, because (1) it is an interesting model that pushes the envelope of what sorts of assets can be bought and sold using algorithmic market makers and (2) it has been falling apart very publicly. We have talked about various boring old-fashioned objections to applying algorithmic market making to houses. For instance, I have written that houses are less legible to trading algorithms than stocks or bonds are:
Houses are like bonds — there are a lot of them, they’re all different, they each have idiosyncratic weird issues, the trade size tends to be in the six figures, no individual house trades all that often — but much more so. ... Most of the facts about a house are, like, the roof is in rough shape, or there’s radon in the basement, or the wallpaper is ugly; they are fuzzy and subjective and complicated and not all that computer-legible.
I have written that the time scale of the housing market makes this business risky and capital-intensive:
If you buy 100 houses you’re out tens of millions of dollars, and if you get the price wrong you won’t find out for the months it takes you to resell them. Meanwhile you keep buying houses using the wrong pricing algorithm.
And that those errors tend to all be in the same direction at the same time:
In particular it will sometimes get it wrong for a while, all in the same direction. It will think “house prices are going up” and pay a lot for thousands of houses, and then it won’t be able to sell them, and it will think “oops house prices are going down” and try to dump all those houses quick in a fire sale.
Again, these are boring objections, and I’d have loved it if Zillow was like “no actually, really good artificial intelligence models solve all those problems, a robot can do market-making for houses, the future is here.” That just would have been cool! I like the future! If your algorithm is really good and you have enough capital, then maybe six months of house trading for you is like six minutes of stock trading for Virtu and it all sort of works out.
But, no, in the event it’s a risky speculative business and Zillow didn’t have the stomach for it. Barton said on the call:
Fundamentally, we have been unable to predict future pricing of homes to a level of accuracy that makes this a safe business to be in. ... And we’ve seen all this volatility in both directions, right now in the wrong direction. And we're still at a scale that is small compared to what it needs to be. And so as we put our minds in the state of, all right, we've got these new assumptions that we'd be naive not to assume will happen again in the future, we pump them into the model and the model cranks out a business that has a high likelihood, at some point, of putting the whole company at risk, not just the business, but in the more normal case, just causes a ton of volatility in earnings, which is not a great look for a public company.
Also, I’ve been writing about Zillow Offers a lot because it is a big brand name and falling apart in public, but there are other companies that do this. They still do it:
It’s become clear that Zillow misjudged the housing market, making more aggressive offers just as competitors Opendoor Technologies Inc. and Offerpad Solutions Inc. were growing more cautious.
Opendoor, which went public last year through a merger with one of Chamath Palihapitiya’s blank-check companies, saw its shares drop 15% on Tuesday after the news that Zillow was selling 7,000 homes raised questions about the iBuyer business model.
The company it in a statement Tuesday it was “well-positioned to meet consumer demand.”
“We are open for business,” a spokesman for the company said.
Ben Thompson has a good column today about Zillow Offers; he writes about the competition:
Unlike Zillow, Opendoor was built from day one to turn over houses, and while the company doesn’t release earnings until next week, its public posture is that it remains open for business.
What is also worth noting is that Opendoor has two more advantages that come from being a startup laser-focused on home-buying: first, while Zillow started with a Zestimate tool that was about attracting customers to the top of the real estate funnel, and thus only ever had to be directionally correct, Opendoor knew from day one its entire fate as a business rested on its model’s accuracy; it’s definitely plausible to imagine its accuracy being much better as a result.
Second, Opendoor has a dramatically larger appetite for risk than Zillow does. Yes, it is a public company now, but it is a public company whose entire business is home-buying; investors know what they are getting into. And, by extension, Opendoor has no choice but to make the model work: they don’t have a profitable Marketplace business to fall back on.
