On the afternoon of Friday, March 27, as several news outlets reported at the time, somebody apparently made $2.4 million from a tweet. That tweet was a bit of breaking news from Wall Street Journal writer Dana Mattioli: Quicker than any human seemingly could have done it, someone—or rather something—bought…
The Wolf of Wall Tweet
A Web-reading bot made millions on the options market. It also ate this guy’s lunch.
Illustration by Robert Neubecker
On the afternoon of Friday, March 27, as several news outlets reported at the time, somebody apparently made $2.4 million [≈ Most expensive production car in 2011]from a tweet. That tweet was a bit of breaking news from Wall Street Journalwriter Dana Mattioli:
Quicker than any human seemingly could have done it, someone—or rather something—bought$110,530 [≈ Small rural house, 2011] worth of cheap options on Altera, a company that makes digital circuits.* Over the next several minutes and until the end of the day, as humans digested Mattioli’s takeover rumor at human speed, Altera’s stock price rose. When all was said and done, those cheap options had resulted in a $2.4 million [≈ Most expensive production car in 2011]profit. Speculation immediately centered on the idea that an automated program (a “bot”) had scanned the tweet, interpreted its meaning, and instantly bought those options based on an algorithm. The robot had read the tweet and made a killing on it before anyone knew what wasgoing on.
Seth Stevenson is a frequent contributor to Slate. He is the author of Grounded: A Down to Earth Journey Around the World.
On April 6, a Reuters report disproved the initial hypothesis. In fact, Reuters reported, the trade occurred 19 seconds before thetweet, and one second after a headline appeared on the Dow Jones Newswire.
I know a guy—a human guy—who was on the other side of that trade. And he says this wasn’t the first time it happened to him. He’s convinced someone’s figured out an algorithm that’s faster than anything he’s ever seen before. So fast, he fears, that it might eventually put him out of a job.
My friend is a stock options market maker on Wall Street. You can buy an option from him that gives you the right to purchase a stock at some point in the future for a price you agree upon now. Let’s say a stock is right this second selling at $30 per share. You buy a truckload of options that grant you the right to buy the stock at $35 per share any time within the next hour. Those options are worth peanuts at the moment—they’re “out of the money” and thus cost very little to buy—but if the stock somehow zoomed up past $40 in the next 10 minutes, they’d suddenly be worth a fortune.
Explaining exactly how my friend’s job works in the real world gets really complicated really quickly, but you can think of him as a bookie. He makes it possible for you to place bets that a stock will go up or down. Like a bookie, he’s essentially playing defense while the bettors are playing offense. He wants to set a betting line that reflects realistic odds. But if one bettor knows something everybody else doesn’t (say, that the team’s star quarterback won’t be playing on Sunday, or that Intel is about to buy the team), then my friend can get slammed.
On the afternoon of Friday, March 13, my friend noticed something strange. A rumor exploded that (as news outlets later reported) Exxon might buy a company called Whiting Petroleum, and in an instant—before any human could have possibly acted on it—someone had “lit up the options market.” Trading was halted, but by the time it reopened, the damage had been done. “I personally lost $100,000 [≈ cost of Porsche 911] in one second,” says my friend. His firm lost more. As for whoever or whatever it was that bought the options? “I’d guess they made between$1 and $2 million. Which is not bad for one second.”
Some news outlets have designed news feeds that are meant to be read by computers instead of by humans.
Next came the famous $2.4 million [≈ Most expensive production car in 2011] Altera windfallon March 27. And then on Wednesday, April 1, when the drugmaker Receptos was involved in takeover rumors, it happened again. Shares in Receptos leaped, but not before somebody had already bought a slew of options at lightning speed, banking another tidy sum. (My friend’s firm escaped dramatic damage in these instances, losing less than $30,000 [≈ Average new car] between the two. Others were surely less lucky.) In each of these cases, the buyer appears to have responded within moments to a tweet, or possibly to a phrase posted in some other online venue—nailing down the precise trigger is difficult.
Could it be a human and not a bot making these trades? My friend doesn’t think so. The complexity of the orders would slow a person down too much to be feasible. “It would be impossible for me to do. By the time you could read the news, process it, and press the ‘buy everything’ button, it would take too long. The speed is unbelievable. They’re buying everything within like 3 seconds of it coming out, which is not possible for a human.”
Could there be more than one single outfit behind these three trades? Again, my friend thinks no. He says that a firm called Lime Brokerage was named on all three trades. Lime wouldn’t have placed these trades directly; it facilitated them for someone else. But my friend is confident that whoever’s using Lime to place these trades is the same person. “My job is basically being a pattern reader,” my friend says, “and on these three trades the pattern was identical. It’s the same guy.”
When I spoke to Lime’s chief operating officer, Tony Huck, he said he thought it was unlikely one of Lime’s clients had made the trades. While he acknowledged it was possible, he said that options trading is a small part of Lime’s business and that with regard to these incidents, “It doesn’t fit the profile for how our clients trade and for the size that they trade.” I checked back with my friend. I then got in touch with Lime once more to tell the firm I’d seen a trade ticket suggesting it was the brokerage of record on one of these trades, made on the Miami Options Exchange (where Lime is one of 41 registered members). Lime asked for time to respond, but given several days and several more requests from me, the company did not comment further.
If you’ve read the Michael Lewis book Flash Boys, you know about the high-frequency trading wars. But the story here was a little different. Those HFT guys were detecting that someone had interest in buying a stock at $5 a share, and then, using technological hocus-pocus, jumping in to buy it first before immediately reselling it to the person at $5.01 a share—over and over, in tons of different stocks, making tiny gains at massive volume.
What we’re talking about here are options trades based on breaking rumors. And because options are derivatives—you’re buying the right to buy shares, not the shares themselves—it’s possible to achieve larger wins for a smaller outlay of cash. What makes these particular trades so striking is that they were made at the very tail end of the day, when the bought options were all only minutes from expiring. The odds that any given stock will suddenly rocket in the next few minutes are extremely low, which makes buying expiring options cheap and the bet very lucrative if it pays off. Consider that if the purchaser of those Altera options had taken his $110,530 [≈ Small rural house, 2011] and simply bought regular stock in Altera with it, he would have cleared about a $34,000 [≈ cost of Honda CR-V] profit by the end of the day. Instead, using options, he made$2.4 million [≈ Most expensive production car in 2011].
Bots that make trades based on news content have been around for years. Some news outlets, such as Bloomberg and Dow Jones, have even designed news feeds that are meant to be read by computers instead of humans. They send information directly to the robots in more easily machine-interpretable formats. There have also been reports about hedge funds that trade based on sentiments expressed in tweets. In the case of the Altera incident, though, a bot appeared to read a rumor, understand it, and instantly execute an options strategy based on it. And for my friend—at least in his corner of the business, a corner he’s worked in for seven years—this felt like something radically new. “It used to feel like a race that we could win or lose,” he says. “But whatever algorithm they’ve developed, we are now completely helpless. Sitting ducks. This is by far the most advanced version of this we’ve ever seen. It’s at a totally different level.”
It also feels pretty far from the theoretical purpose of options trading. Options are meant to provide insurance (a “hedge”) against potential losses in a stock position. Market makers like my friend create the environment in which to buy the insurance. This bot instead treats that market like a roulette wheel—except it knows exactly where the ball will land. “If someone else has what we call the ‘future script,’ ” says my friend, referring to the crystal ball of the algorithm bot, “it really feels like they’re just robbing you. Yes, what they’re doing is legal, and you can say fair is fair, the person with the fastest computer gets all the money. But think back to 1995, when the point of options was still insurance, and imagine telling someone that there’s a firm that makes a computer that can read a tweeted rumor and buy stocks one second later to make millions of dollars. It would seem crazy, but that’s where we are.”
To be fair, there is risk for the bot-users, too. “Automated trading on Internet content is a highly competitive business,” says Paul Tetlock, a Columbia Business School professor. “Many firms have started and failed.” Things can go horribly awry for the bots. Tetlock pointed me to an example in which a 6-year-old news story about United Airlines’ 2002 bankruptcy somehow reappeared online in 2008. “As algorithms traded on this stale news,” Tetlock explained to me in an email, “United’s stock price plummeted by 76 percent within minutes.” But the price almost fully rebounded within the day.
And bot errors aside, Tetlock sees little reason to be morally concerned about these sorts of developments—even if it means we end up with a market that’s just bot versus bot. “So the smartest geeks are reaping more of the gains from trading relative to people with the quickest fingers or the best personal connections,” Tetlock wrote me. “If humans’ programs are better at trading on news than humans themselves, it’s not clear why markets would be harmed. In such a world, stock prices would react quickly and accurately to new information.”
There’s another risk, as well, In addition to the bots-gone-wild hazard: A few amazingly well-designed bots might drive everyone else out of the game. That could shrink the market and make it far less useful. “If everyone gets the same information but they analyze it in different ways, or believe different things,” says Kenneth Ahern, a professor at the University of Southern California’s business school, “the market should provide the best price. But if you have a barrier to entry where some are so much faster than others, prices will be biased. Still, it might be that they’ll only be biased for 15 minutes. I haven’t seen enough evidence yet to say we should regulate it. My concern is always over whether the regulators will do better than the market will.”
You needn’t weep over my friend or his firm losing money. You might even cheer the ingenuity of a person who programs an algorithm to read tweets and profit off them. My friend, of course, has a different perspective. For him, it feels like someone is reaching into his pocket and taking money out: “It’s like they’re insider trading on the news.”
*Correction, April 21, 2015: This article originally misstated that a purchase of options on March 27 immediately followed a tweet by journalist Dana Mattioli. It occurred19 seconds before the tweet and followed a newswire post by one second. The article and its headlines have been updated to reflect this. (Return.)