Flash Boys: A Wall Street Revolt

by Michael Lewis

  ** ING, oddly enough, managed IEX’s then thirty-person 401(k) plan. Seeing this, John Schwall returned to his side career in private investigation. After some digging, he developed the opinion that any money manager who arbitrarily denied his clients access to markets might have violated his fiduciary responsibility. On those grounds, Schwall pulled the company’s 401(k) from ING.

  †† Sixty percent of the time that this feeding frenzy occurs on a public stock exchange, no trade is recorded. The frenzy comes in response to a trade that has occurred in some dark pool. The dark pools are not required to report their trades in real time; and so, on the official tape, the frenzy appears unprovoked. It isn’t.

  ‡‡ The Financial Industry Regulatory Authority (FINRA) publishes its own odd ranking of the public and private stock markets, based on how well they avoid breaking the law, presumably inadvertently, by trading outside the National Best Bid and Offer. In its first two months of trading, IEX ranked #1 on FINRA’s list.

  §§ The reader might question the characterization of such small-time skimming as scalping. But a penny here, a penny there adds up in the most extraordinary ways in the U.S. stock market. At IEX, the Puzzle Masters made a quick-and-dirty calculation of the likely profits made annually by HFT from dark pool arbitrage. They added up all its instances over a fifteen-day period, then came up with a number: The haul for HFT from the U.S. stock market alone came to more than a billion dollars a year. And this was just a single trading strategy. “They’ve been in business for ten weeks and they’ve now found four of these strategies,” said one big investor of IEX. “Who knows how many more they’ll find?” A billion here, a billion there: It adds up.

  CHAPTER EIGHT

  THE SPIDER AND THE FLY

  The trial of Sergey Aleynikov ran for ten days in December of 2010 and was notable for its paucity of informed outsiders. High-frequency trading was a small world, and the people who did it, or knew anything at all about it, apparently had far less interest in testifying at trials than in making their personal fortunes. The one outside expert witness on the subject called by the government was an assistant professor of finance at Illinois Institute of Technology named Benjamin Van Vliet. Van Vliet had become an expert in response to journalists’ need for one. While teaching a computer coding course, he’d cast around for something sexy for the students to program, and landed on high-frequency trading platforms. In mid-2010, Forbes magazine called him out of the blue to ask him what he thought about a fiber-optic cable that Spread Networks had strung from Chicago to New Jersey. Van Vliet had never heard of Spread Networks, and knew nothing about the cable, but wound up with his name in print—which, of course, led to more calls from journalists, who needed a high-frequency trading expert. Then came the flash crash, and Van Vliet’s phone rang off the hook. Eventually, federal prosecutors found him and asked him to serve as their expert witness in the trial of a former Goldman Sachs high-frequency programmer. Van Vliet still had never actually done any high-frequency trading himself, and had little to add on the value or the gist of what Serge Aleynikov had taken from Goldman Sachs. About the market itself he was badly misinformed. (He described Goldman Sachs as “the New York Yankees” of high-frequency trading.) He turned out to have testified as an expert witness in an earlier trial involving the theft of high-frequency trading code, after which the judge in the case said that the idea that a high-frequency trading program was some kind of science was “utter baloney.”

  The jury in Sergey Aleynikov’s trial consisted mainly of high school graduates; all of the jurors lacked experience programming computers. “They would bring my computer into the courtroom,” recalled Serge incredulously. “They would pull out the hard drive and show it to the jury. As evidence!” Save for Misha Malyshev, Serge’s onetime employer, the people who took the stand had no credible knowledge of high-frequency trading: how the money got made, what sort of computer code was valuable, and so on. Malyshev testified as a witness for the prosecution that Goldman’s code was of no use whatsoever in the system he’d hired Serge to build—Goldman’s code was written in a different programming language, it was slow and clunky, it had been designed for a firm that was trading with its own customers, and Teza, Malyshev’s firm, didn’t have customers, and so on—but when he looked over, he saw that half the jury appeared to be sleeping. “If I were a juror, and I wasn’t a programmer,” said Serge, “it would be very difficult for me to understand why I did what I did.”

  Goldman Sachs’s role in the trial was to make genuine understanding even more difficult. Its employees, on the witness stand, behaved more like salesmen for the prosecution than citizens of the state. “It’s not that they lied,” said Serge. “But they told things that were not in their expertise.” When his former boss, Adam Schlesinger, was asked about the code, he said that everything at Goldman was proprietary. “I wouldn’t say he lied, but he was talking about stuff that he did not understand, and so he was misunderstood,” said Serge.

  Our system of justice is a poor tool for digging out a rich truth. What was really needed, it seemed to me, was for Serge Aleynikov to be forced to explain what he had done, and why, to people able to understand the explanation and judge it. Goldman Sachs had never asked him to explain himself, and the FBI had not sought help from anyone who actually knew anything at all about computers or the high-frequency trading business. And so over two nights, in a private room of a Wall Street restaurant, I convened a kind of second trial. To serve as both jury and prosecution, I invited half a dozen people intimately familiar with Goldman Sachs, high-frequency trading, and computer programming. All were authorities on our abstruse new stock market; several had written high-frequency code; one had actually developed software for Goldman’s high-frequency traders. All were men. They’d grown up in four different countries between them, but all now lived in the United States. All of them worked on Wall Street, and so, to express themselves freely, they needed to remain anonymous. Among them were employees of IEX.

  All were naturally skeptical—of both Goldman Sachs and Serge Aleynikov. They assumed that if Serge had been sentenced to eight years in jail he must have done something wrong. They just hadn’t bothered to figure out what that was. All of them had followed the case in the newspapers and noted the shiver it had sent through the spines of Wall Street’s software developers. Until Serge was sent to jail for doing it, it was common practice for Wall Street programmers to take code they had worked on when they left for new jobs. “A guy got put in jail for taking something no one understood,” as one of Serge’s new jurors put it. “Every tech programmer out there got the message: Take code and you could go to jail. It was huge.” The arrest of Serge Aleynikov had also caused a lot of people, for the first time, to begin to use the phrase “high-frequency trading.” Another new juror, who in 2009 had worked for a big Wall Street bank, said, “When he was arrested, we had a meeting for all the electronic trading personnel, to talk about a one-pager they’d drafted to be discussed with their clients around this new topic called ‘high-frequency trading.’ ”

  The restaurant was one of those old-school Wall Street places that charge you a thousand bucks for a private room and then more or less challenge you to eat your way back to even. Food and drink arrived in massive quantities: vast platters of lobster and crab, steaks the size of desktop computer screens, smoking mountains of potatoes and spinach. It was the sort of meal cooked decades ago, for traders who spent their days trusting their gut and their nights rewarding it; but this monstrous feast was now being served to a collection of weedy technologists, the people who controlled the machines that now controlled the markets, and who had, in the bargain, put the old school out of business. They sat around the table staring at the piles of food, like a conquering army of eunuchs who had stumbled into the harem of their enemy. At any rate, they made hardly a dent. Serge, for his part, ate so little, and with such disinterest, that I half expected him to lift off his chair and float up to the ceiling.

  His n
ew jurors began, interestingly, by asking him lots of personal questions. They wanted to figure out what kind of guy he was. They took an interest, for example, in his job-market history, and noted that his behavior was pretty consistently that of a geek who had more interest in his work than in the money the work generated. They established fairly quickly—how, I do not know—that he was not just smart but seriously gifted. “These guys are usually smart in one small area,” one of them later explained to me. “For a technologist to be so totally dominant in so many areas is just really, really unusual.”

  They then began to probe his career at Goldman Sachs. They were surprised to learn that he had “super-user status” inside Goldman, which is to say he was one of a handful of people (roughly 35, in a firm that then had more than 31,000 employees) who could log onto the system as an administrator. Such privileged access would have enabled him, at any time, to buy a cheap USB flash drive, plug it into his terminal, and take all of Goldman’s computer code without anyone having any idea that he had done it. That fact alone didn’t prove anything to them. As one pointed out to Serge directly, lots of thieves are sloppy and careless; just because he was sloppy and careless didn’t mean he was not a thief. On the other hand, they all agreed, there wasn’t anything the least bit suspicious, much less nefarious, about the manner in which he had taken what he had taken. Using a subversion repository to store code and deleting one’s bash history were common practices. The latter made a great deal of sense if you typed your passwords into command lines. In short, Serge had not behaved like a man trying to cover his tracks. One of his new jurors stated the obvious: “If deleting the bash history was so clever and devious, why had Goldman ever found out he’d taken anything?”

  To these new jurors, the story that the FBI found so unconvincing—that Serge had taken the files because he thought he might later like to parse the open source code contained within—made a lot of sense. As Goldman hadn’t permitted him to release his debugged or improved code back to the public—even though the original free license often stated that improvements must be publicly shared—the only way for him to get his hands on these files was to take the Goldman code. That he had also taken some code that wasn’t open source, which happened to be in the same files as the open source code, surprised no one. Grabbing a bunch of files that contained both open source and non–open source code was an efficient way for him to collect the open source code, even if the open source code was the only code that interested him. It would have made far less sense for him to hunt around the Internet for the open source code he wanted, as it was scattered all over cyberspace. It was also entirely plausible to them that Serge’s interest was confined to the open source code, because that was the general-purpose code that might be repurposed later. The Goldman proprietary code was written specifically for Goldman’s platform; it would have been of little use in any new system he wished to build. (The two small pieces of code Serge had sent into Teza’s computers before his arrest both came with open source licenses.) “Even if he had taken Goldman’s whole platform, it would have been faster and better for him to write the new platform himself,” said one juror.

  Several times Serge surprised the jurors with his answers. They were all shocked, for instance, that from the day Serge first arrived at Goldman, he had been able to send Goldman’s source code to himself weekly, without anyone at Goldman saying a word to him about it. “At Citadel, if you stick a USB drive into your work station, someone is standing next to you within five minutes, asking you what the hell you are doing,” said a juror who had worked there. Most were surprised by how little Serge had taken in relation to the whole: eight megabytes, in a platform that consisted of nearly fifteen hundred megabytes of code. The most cynical among them were surprised mostly by what he had not taken.

  “Did you take the strats?” asked one, referring to Goldman’s high-frequency trading strategies.

  “No,” said Serge. That was one thing the prosecutors hadn’t accused him of.

  “But that’s the secret sauce, if there is one,” said the juror. “If you’re going to take something, take the strats.”

  “I wasn’t interested in the strats,” said Serge.

  “But that’s like stealing the jewelry box without the jewels,” said another juror.

  “You had super-user status!” said the first. “You could easily have taken the strats. Why didn’t you?”

  “To me, the technology really is more interesting than the strats,” said Serge.

  “You weren’t interested in how they made hundreds of millions of dollars?” asked someone else.

  “Not really,” said Serge. “It’s all one big gamble, one way or another.”

  Because they had seen it before in other programmer types, they were not totally shocked by his indifference to Goldman’s trading, or by how far Goldman had kept him from the action. Talking to a programmer type about the trading business was a bit like talking to the house plumber at work in the basement about the card game the Mafia don was running upstairs. “He knew so little about the business context,” one of the jurors said, after attending both dinners. “You’d have to try to know as little as he did.” Another said, “He knew as much as they wanted him to know about how they made money, which was virtually nothing. He wasn’t there for very long. He came in with no context. And he spent all of his time troubleshooting.” Another said he had found Serge to be the epitome of the programmer whose value the big Wall Street banks tried to minimize—by using their skills without fully admitting them into the business. “You see two résumés from the banks,” he said. “You line them up on paper and say maybe there’s a ten percent difference between them. But one guy is getting paid three hundred grand and the other is getting one point five million. The difference is one guy has been given the big picture, and the other hasn’t.” Serge had never been shown the big picture. Still, it was obvious to the jurors—even if it wasn’t to Serge—why Goldman had hired him when it had. With the introduction of Reg NMS in 2007, the speed of any financial intermediary’s trading system became its most important attribute: the speed with which it took in market data and the speed with which it responded to that data. “Whether he knew it or not,” said one juror, “he was hired to build Goldman’s view of the market. No Reg NMS, no Serge in finance.”

  At least some part of the reason he remained oblivious to the nature of Goldman Sachs’s trading business, all of the jurors noticed, was that his heart was elsewhere. “I think passion plays a big role,” said a juror who himself had spent his entire career writing code. “The moment he started talking about coding, his eyes lit up.” Another added, “The fact that he kept trying to work on open source shit even while he was at Goldman says something about the guy.”

  They didn’t all agree that what Serge had taken had no value, either to him or to Goldman. But what value it might have had in creating a new system would have been trivial and indirect. “I can guarantee you this: He did not steal code to use it on some other system,” one said, and none of the others disagreed. For my part, I didn’t fully understand why some parts of Goldman’s system might not be useful in some other system. “Goldman’s code base is like buying a really old house,” one of the jurors explained. “And you take the trouble to soup it up. But it still has the problems of a really old house. Teza was going to build a new house, on new land. Why would you take one-hundred-year-old copper pipes and put them in my new house? It isn’t that they couldn’t be used; it’s that the amount of trouble involved in making it useful is ridiculous.” A third added, “It’s way easier to start from scratch.” Their conviction that Goldman’s code was not terribly useful outside of Goldman grew even stronger when they learned—later, as Serge failed to mention it at the dinners—that the new system Serge planned to create was to be written in a different computer language than the Goldman code.

  The perplexing question, at least to me, was why Serge had taken anything. A full month after he’d left Goldman Sachs, he
still had not touched the code he had taken. If the code was so unimportant to him that he didn’t bother to open it up and study it; if most of it was either so clunky or so peculiar to Goldman’s system that it was next to useless outside Goldman—why take it? Oddly, his jurors didn’t find this hard to understand. One put it this way: “If Person A steals a bike from Person B, then Person A is riding a bike to school, and Person B is walking. Person A is better off at the expense of Person B. That is clear-cut, and most people’s view of theft.

  “In Serge’s case, think of being at a company for three years, and you carry a spiral notebook and write everything down. Everything about your meetings, your ideas, products, sales, client meetings—it’s all written down in that notebook. You leave for your new job and take the notebook with you—as most people do. The contents of your notebook relate to your history at the prior company but have very little relevance to your new job. You may never look at it again. Maybe there are some ideas, or templates, or thoughts you can draw on. But that notebook is related to your prior job, and you will start a new notebook at your new job which will make the old one irrelevant. . . . For programmers, their code is their spiral notebook. [It enables them] to remember what they worked on—but it has very little relevance to what they will build next. . . . He took a spiral notebook that had very little relevance outside of Goldman Sachs.”

  To the well-informed jury, the real mystery wasn’t why Serge had done what he had done. It was why Goldman Sachs had done what it had done. Why on earth call the FBI? Why exploit the ignorance of both the general public and the legal system about complex financial matters to punish this one little guy? Why must the spider always eat the fly?

  The financial insiders had many theories about this: that it was an accident; that Goldman had called the FBI in haste and then realized the truth, but lost control of the legal process; that in 2009 Goldman had been on hair-trigger alert to personnel losses in high-frequency trading, because they could see how much money would be made from it, and thought they could compete in the business. The jurors all had ideas about why what had happened had happened. One of the theories was more intriguing than the others. It had to do with the nature of a big Wall Street bank, and the way people who worked for it, at the intersection of technology and trading, got ahead. As one juror put it, “Every manager of a Wall Street tech group likes to have people believe that his guys are geniuses. Russians, whatever. His whole persona among his peers is that what he and his team do can’t be replicated. When people find out that ninety-five percent of their code is open source, it kills that perception. What the guy can’t say, when he gets told Serge has taken something, is ‘it doesn’t matter what he took because it’s worse than what they’ll create on their own.’ So when the security people come to him and tell him about the downloads, he can’t say, ‘No big deal.’ And he can’t say, ‘I don’t know what he took.’ ”