Flash Boys: A Wall Street Revolt (9 page)

IN THE FALL
of 2009, an article in a trade magazine caught Brad Katsuyama’s eye. He’d spent the better part of a year trying and failing to find anyone who actually worked in what was now regularly referred to as high-frequency trading who was willing to explain to him how he made his money. The article claimed that HFT technologists were unhappy with the widening gulf in pay between themselves and the senior trading strategists of their firms, some of whom were rumored to be taking home hundreds of millions of dollars a year. He went looking for one of these unhappy technologists. The very first call he made, to a guy at Deutsche Bank who dealt often with HFT, gave him two names. Ronan’s was the first.

In his interview, Ronan described to Brad what he’d witnessed inside the exchanges: the frantic competition for nanoseconds, the Toys “R” Us cage, the wire gauze, the war for space within the exchanges, the tens of millions being spent by high-frequency traders for tiny increments of speed. As he spoke, he filled huge empty tracts on Brad’s mental map of the financial markets. “What he said told me that we needed to care about microseconds and nanoseconds,” said Brad. The U.S. stock market was now a class system, rooted in speed, of haves and have-nots. The haves paid for nanoseconds; the have-nots had no idea that a nanosecond had value. The haves enjoyed a perfect view of the market; the have-nots
never saw the market at all.
What had once been the world’s most public, most democratic, financial market had become, in spirit, something more like a private viewing of a stolen work of art. “I learned more from talking to him in an hour than I learned from six months of reading about HFT,” said Brad. “The second I met him I wanted to hire him.”

He wanted to hire him without being able to fully explain, to his bosses or even to Ronan, what he wanted to hire him for. He couldn’t very well call him Vice President in Charge of Explaining to My Clueless Superiors Why High-Frequency Trading Is a Travesty. So he called him Head of High-Frequency Trading Strategies. “I felt he needed a ‘Head of’ title,” said Brad, “to get more respect from people.” That was Brad’s main concern: that people on the trading floor, even at RBC, would take one look at Ronan and see a guy in a yellow jumpsuit who’d just emerged from some manhole. Ronan didn’t even pretend to know what happened on a trading floor. “He had questions that were unbelievably rudimentary but that were necessary,” said Brad. “He didn’t know what ‘bid’ and ‘offer’ was. He didn’t know what it meant to ‘cross the spread.’ ”

On the side, without making a big deal of it, Brad started to teach Ronan the language of trading. A “bid” was an attempt to buy stock, an “offer” an attempt to sell it. To cross the spread, if you were selling, meant to accept the bidder’s price, or, if you were buying, the offering price. “This fucking guy didn’t laugh at me,” said Ronan. “He sat down and explained it.” That was their private deal: Brad would teach Ronan about trading, and Ronan would teach Brad about technology.

Right away there was something to teach. Brad and his team were having trouble turning Thor into a product they could sell to investors. The investors they’d told about their discovery were clearly eager to buy Thor and use it for themselves—T. Rowe Price’s Gitlin had more or less tried to buy it on the spot—but Thor now had its problems. The experiment of arriving at the exchanges at the same time had worked perfectly—the first time. It proved hard to repeat, because it was difficult to coax thirteen light signals to arrive in thirteen different stock exchanges spread across northern New Jersey within 350 microseconds of each other—or roughly 100 microseconds less than the time they had calculated it would take some high-speed trader to front-run their order. They’d succeeded the first time by estimating the differences in travel time it took to send the messages to the various exchanges, and by building the equivalent delays into their software. But the travel times were never the same. They had no control over the path the signals took to get to the exchanges, or how much traffic was on the network. Sometimes it took 4 milliseconds for their stock market orders to arrive at the New York Stock Exchange; other times, it took 7 milliseconds. When the travel time differed from their guesses of what it would be, the market, once again, vanished.

In short, Thor was inconsistent; and it was inconsistent, Ronan explained, because the paths the electronic signals took from Brad’s desk to the various exchanges were inconsistent. Ronan could see that these traders hadn’t thought much about the physical process by which their signals traveled to the New Jersey stock exchanges. “I realized very quickly,” he said, “and they’ll admit this, so I mean no disrespect, that they had no fucking clue what they were doing.” The signal sent from Brad’s desk arrived at the New Jersey exchanges at different times because some exchanges were farther from Brad’s desk than others. The fastest any high-speed trader’s signal could travel from the first exchange it reached to the next one was 465 microseconds, or one two-hundredths of the time it takes to blink your eye, if you have a talent for it. That is, for Brad’s trading orders to interact with the market as displayed on his trading screens, they needed to arrive at all the exchanges within a 465-microsecond window. The only way to do that, Ronan told his new colleagues at RBC, was to build and control your own fiber network.

To make his point, Ronan brought in oversized maps of New Jersey showing the fiber-optic networks built by telecom companies. On the maps you could see just how a signal traveled from Brad’s trading station at One Liberty Plaza to the exchanges. When he unrolled his first map, a guy who worked in RBC’s network support team burst out, “How the fuck did you get those? They’re telecom property! They’re proprietary!” Ronan explained, “When they said they wouldn’t give them to me because they were proprietary, I said, ‘Well, then, proprietarily fuck off.’ ” The high-frequency traders were paying the telecom carriers too much to be denied whatever they wanted, and Ronan had been the agent of their desires. “These maps are like fucking gold,” he said. “But I had brought them so much business that they would let me see inside their freaking wife’s underwear drawer if I asked them to.”

The maps told a story: Any trading signal that originated in lower Manhattan traveled up the West Side Highway and out the Lincoln Tunnel. Perched immediately outside the tunnel, in Weehawken, New Jersey, was the BATS exchange. From BATS the routes became more complicated, as they had to find their way through the clutter of the Jersey suburbs. “New Jersey is now carved up like a Thanksgiving turkey,” said Ronan. One way or another, they traveled east to Secaucus, the location of the Direct Edge family of exchanges founded by Goldman Sachs and Citadel, and south to the Nasdaq family of exchanges in Carteret. The New York Stock Exchange further complicated the story. In early 2010, NYSE still had its computer servers in lower Manhattan, at 55 Water Street. (They moved them to distant Mahwah, New Jersey, that August.) As it was less than a mile from Brad’s desk, NYSE appeared to be the stock market closest to him; but Ronan’s maps showed the incredible indirection of optic fiber in Manhattan. “To get from Liberty Plaza to Fifty-five Water Street, you might go through Brooklyn,” he explained. “You can go fifty miles to get from Midtown to downtown. To get from a building to a building across the street you could travel fifteen miles.” It was a ten-minute walk from RBC’s office at Liberty Plaza to the New York Stock Exchange. But from a computer’s point of view, the New York Stock Exchange was further from RBC’s offices than Carteret.

To Brad the maps explained, among other things, why the market on BATS had proved so accurate. The reason they were always able to buy or sell 100 percent of the shares listed on BATS was that BATS was always the first stock market to receive their orders. News of their buying and selling hadn’t had time to spread throughout the marketplace. “I was like, ‘Holy shit, BATS is just closest to us.’ It’s right outside the freaking tunnel.” Inside BATS, high-frequency trading firms were waiting for news that they could use to trade on the other exchanges. They obtained that news by placing very small bids and offers, typically for 100 shares, for every listed stock. Having gleaned that there was a buyer or seller of Company X’s shares, they would race ahead to the other exchanges and buy or sell accordingly. (The race they needed to win was not a race against the ordinary investor, who had no clue what was happening to him, but against other high-speed traders.) The orders resting on BATS were typically just the 100-share minimum required for an order to be at the front of any price queue, as their only purpose was to tease information out of investors. The HFT firms posted these tiny orders on BATS—orders to buy or sell 100 shares of basically every stock traded in the U.S. market—not because they actually wanted to buy and sell the stocks but because they wanted to find out what investors wanted to buy and sell before they did it. BATS, unsurprisingly, had been created by high-frequency traders.

The funny thing was that a lot of what Ronan had seen and heard didn’t make sense to him: He didn’t know what he knew. Brad now helped him to understand. For instance, Ronan had noticed the HFT guys creating elaborate tables of the time, measured in microseconds, it took for a stock market order to travel from any given brokerage house to each of the exchanges. “Latency tables,” these were called. The times were subtly different for every brokerage house—they depended upon where the brokerage house physically was located and which fiber networks it leased in New Jersey. These tables took trouble to create and were of obvious value to high-frequency traders, but Ronan had no idea why. This was the first Brad had heard of latency tables, but he knew exactly why they had been created: They enabled high-frequency traders to identify brokers by the time their orders took to travel from one exchange to the other. Once you had figured out which broker was behind any given stock market order, you could discern patterns in each broker’s behavior. If you knew which broker had just come into the market with an order to buy 1,000 shares of Intel, you might further guess whether those 1,000 shares were the entire order or a part of a much larger order. You might also guess how the broker might distribute the order among the various exchanges and how much above the current market price for Intel shares the broker might be willing to pay. The HFT guys didn’t need perfect information to make riskless profits; they only needed to skew the odds systematically in their favor. But, as Brad put it, “What you’re looking for ultimately is large brokers who are behaving idiotically with their customers’ orders. That’s the real gold mine.”

He also knew that Wall Street brokers had a new incentive to behave idiotically, because he had himself succumbed to the temptation. When Wall Street decided where to route their clients’ stock market orders, they were now greatly influenced by the new system of kickbacks paid and fees charged to them by the exchanges: If a big Wall Street broker stood to be paid to send an order to buy 10,000 shares of Intel to BATS but was charged to send the same order to the New York Stock Exchange, it would program its routers to send the customer’s order to BATS. The router, designed by human beings, took on a life of its own.

Along with the trading algorithms, the routers were a critical piece of technology in the automated stock markets. Both are designed and built by people who work for the Wall Street broker. Both do the thinking that people used to do, but the intellectual tasks they perform are different. The algorithm does its thinking first: It decides how to slice up any given order. Say you want to buy 100,000 shares of XYZ Company at no more than $25 a share, when the market shows a total of 2,000 shares offered at $25. To simply attempt to buy 100,000 shares all at once would create havoc in the market and drive the price higher. The algorithm decides how many shares you buy, when to buy them, and the price to pay. For example, it may instruct the router to carve the 100,000-share order into twenty pieces, and to buy 5,000 shares every five minutes, so long as the price is no higher than $25.

The router determines
where
the order is sent. For instance, a router might instruct the order to go first to a Wall Street firm’s dark pool before going to the exchanges. Or it might instruct the order to go first to any exchange that will pay the broker to trade, and only then to exchanges on which the broker will be compelled to pay to trade. (This is a so-called sequential cost-effective router.) To illustrate how stupid routing can be, say you have told your Wall Street broker—to whom you are paying a commission—that you wish to buy 100,000 shares of Company XYZ at $25 and now, conveniently, there are 100,000 shares for sale at $25, 10,000 on each of ten different exchanges, all of which will charge the broker to trade on your behalf (though far less than the commission you have paid to him). There are, however, another 100 shares for sale, also at $25, on the BATS exchange—which will pay the broker for the trade. The sequential cost-effective router will go first to BATS and buy the 100 shares—and cause the other 100,000 shares to vanish into the paws of high-frequency traders (in the bargain relieving the broker of the obligation to pay to trade). The high-frequency traders can then turn around and sell the shares of Company XYZ at a higher price, or hold onto the shares for a few seconds more, while you, the investor, chase Company XYZ’s shares even higher. In either case, the result is unappealing to the original buyer of Company XYZ’s shares.

That is but the most obvious of many examples of routing stupidity. The customer (you, or someone investing on your behalf) is typically entirely oblivious to the inner workings of both algorithms and routers: Even if he demanded to know how his order was routed, and his broker told him, he would never be sure what was said was true, as he has no sufficiently detailed record of what shares traded and when they traded.

The brokers’ routers, like bad poker players, all had a conspicuous tell. The tell might be a glitch in their machines rather than a twitch of their facial muscles, but it was just as valuable to the HFT guys on the other side of the table.