The Microsecond Market

Sophisticated technology now drives global financial trading to extremes of time and space

11 min read
Photo: Levi Brown; Prop Stylist: Ariana Salvato
Photo: Levi Brown; Prop Stylist: Ariana Salvato

Since money first came into existence, some people have made gobs of it by having particularly timely access to important news. Perhaps the most notorious examples of this phenomenon took place during the first half of the last century in many U.S. cities. Here it was organized crime that profited immensely, and the news of interest was about horse races.

Initially, horse-race results were sent out over Western Union’s telegraph network, but when that company cut off this service to what it deemed shady customers, others with fewer scruples stepped in. On their private wires, race results were sent from the tracks to illegal bookmakers before the public at large learned of them, allowing bookies to accept bets on horses that had already lost and turn down wagers on horses that had already won.

For decades, bookmakers paid handsomely for those wire services, helping to support such notable Chicago mobsters as Mont Tennes and Al Capone. Access to a fast wire carrying breaking results from the track was an offer that bookies couldn’t refuse. A 1951 report of the United States Senate Special Committee to Investigate Crime in Interstate Commerce aptly notes, “The wire service is as essential to a book-maker as the stock ticker to a stockbroker.”

In today’s age of live sports broadcasts, bookmakers can no longer profit this way. But big financial companies that buy and sell stocks and other financial instruments with automated split-second transactions can—and do. The companies engaged in this relatively new practice, called high-frequency trading, are keenly aware of the importance of timely information about markets. And they use enormously sophisticated technology to wring out every last bit of delay—down to the microsecond level or even less—in getting that information and in executing their trades.

A few years ago, hundreds of millions of dollars were spent on a project to connect traders in New York and Chicago with an especially direct data link, and similar amounts are being invested now to hook up New York and London in the fastest possible way. These are just of couple of the most obvious investments in a multibillion-dollar game where advantages measured in millionths of a second can mean millions in profits.

“In the age of high-frequency trading, technological speed itself is a strategy,” says Benjamin Van Vliet, who teaches quantitative finance at the Illinois Institute of Technology’s Stuart School of Business. He likens high-frequency trading to picking up gold coins dropped on the ground—not much analysis or insight is required. Speed, however, is. “Whoever is fastest is going to win every time,” he says.

The buying and selling of stocks, commodity-futures contracts, and other financial instruments was traditionally a noisy affair, done by people calling out trades they wanted to make in the “pits” of various exchanges. Some of that still goes on, but most trading these days is done through the financial industry’s various electronic communication networks. These first sprouted up in the late 1960s and bloomed in the decades that followed. But until fairly recently, people largely remained in the loop, watching how market conditions were changing on their computer screens and pointing and clicking to execute their trades.

Increasingly, though, more sophisticated market participants have been using preprogrammed strategies to execute their trades, often splitting up their purchase or sales orders and submitting them at odd times so that nobody else can easily discern their overall intentions. That strategy helps to avoid driving up the price while you’re buying a lot of something or depressing it when you’re selling.

But such “algorithmic trading” is less useful lately, because the automated platforms that high-frequency traders have put in place over the last few years react so swiftly. “Whenever I wanted to trade some stocks, it seemed someone was looking over my shoulder,” says X. Frank Zhang of the Yale School of Management, who managed a US $500 million investment portfolio on Wall Street between 2008 and 2010 before taking his current academic post. “High-frequency traders could detect my patterns.”

These traders were taking information about what orders were in play in the market and using it to predict how prices would most likely shift. Armed with that knowledge, they would buy and sell stocks or other instruments and sell them again within minutes or even seconds, most likely accruing only a tiny amount on each share traded. But the tiny profits can quickly add up, given the enormous number of transactions. People like Zhang, on the other hand, have to manage long-term investment portfolios. So even if they studied the high-frequency traders’ patterns and strategies, it wouldn’t help them do their jobs.

A study of U.S. financial markets that Zhang conducted in 2010 showed that high-frequency trading was responsible for 78 percent of the dollar trading volume in 2009, up from near zero in 1995. Other estimates are somewhat lower, but most are still well over 50 percent. Not surprisingly, the value of high-frequency trading to the overall economy is controversial. Those engaged in the practice argue that, like the exchanges’ “official market makers,” they provide a valuable source of liquidity: With market makers buying and selling (as the official ones are contractually obligated to do, albeit at slightly different prices so that they can profit from their trades), there’s always somebody to trade with when you want to exchange some stock.

Zhang, for one, is skeptical. He points out that unlike official market makers, high-frequency traders can withdraw from the market at their discretion. As an example of the kind of problem that can cause, he cites the “Flash Crash” of 6 May 2010, when the Dow Jones average mysteriously fell by about 1000 points, only to recover minutes later. That crash has been blamed, in part, on the sudden disappearance of high-frequency trading when the market became unpredictable.

Zhang also says that the volume of high-frequency trading is too large to be justified in terms of providing liquidity to others. If every single traditional market participant bought from or sold to only high-frequency traders, he explains, they would account for only 50 percent of the trading volume. That’s because the calculation of trading volume includes both purchases and sales. So with every transaction between a traditional investor and a high-frequency trader, each contributes the same amount to overall trading volume. That high-frequency traders are responsible for more than 70 percent of overall trading volume means that these firms must often be trading with one another.

The New York Stock Exchange, at 11 Wall Street in lower Manhattan, might seem the epicenter of U.S. stock trading. In fact, the real action takes place about 50 kilometers away, in a huge, windowless building in suburban Mahwah, N.J. NYSE Euronext opened a 400 000-square-foot (37 000-square-meter) data center there in 2010. This is where the New York Stock Exchange houses its “matching engines”—servers that link together a vast number of buy and sell orders coming in from traders. It’s also where the exchange leases space to companies that want their computerized trading equipment installed as close as possible to these matching engines so as to limit signaling delays, both in receiving market information and in executing trades. Every day, some $70 billion changes hands there electronically, with about 1.6 billion shares (out of a total of 6.8 billion in the United States) being bought and sold inside this one stock exchange’s machines.

This data center is thus a key hub for U.S. financial markets. No wonder it’s protected by armed guards, hydraulically operated steel barriers, and bomb-sniffing dogs. The high-frequency trading firms that colocate their equipment there enjoy a valuable speed advantage over others that are not so strategically placed. Some of those tenants had even hoped to garner a fraction of a microsecond’s edge by positioning their equipment especially close to the exchange’s matching engines. But that’s impossible, by design.

“Everyone routes through the same set of switches, the same core network, the same local area network, and then [the data] is delivered at the same speed to each colocation customer’s top of rack, wherever they are located—no one has an advantage,” says Don Brook, global head of infrastructure for the NYSE. “The last piece of secret sauce to make that happen isn’t really that secret: Every cable is the same length, whether you are 10 feet away or 500 feet away.”

Although it seems rather wasteful to have kilometers of fiber-optic networking cable looping around in circles above the server racks, fairness (and the U.S. Securities and Exchange Commission) demand such attention to detail in leveling the playing field among the exchange’s colocated customers. That doesn’t mean, though, that mere chance determines which of those customers will be quickest to grab those gold coins off the ground (to use Van Vliet’s analogy). Technology determines that. “Customers do everything they can to reduce latency inside their boxes,” says Brook.

One of the companies they turn to for help in that race is Solace Systems of Ottawa, Canada. A hardware-based networking appliance that Solace sells is used to connect high-frequency traders’ servers with special-purpose “feed handlers,” which format the raw data feeds that such exchanges provide. “You can use software for this,” says Shawn McAllister, chief technology officer of Solace Systems. “That’s traditionally done, but it’s slower.” And that makes Solace’s ultrafast hardware an easy sell.

The microseconds you save by going to such extremes are well worth the effort, because high-frequency traders often seek to profit using a relatively simple strategy: examining the current set of orders and predicting how prices will shift in response in the next instant. Designing algorithms to make such predictions isn’t all that challenging, and running the code needed to carry them out isn’t all that computationally taxing—what’s hard is coming up with those predictions and acting on them faster than anyone else can.

Wall Street might seem the epicenter of U.S. stock trading. In fact, the real action takes place about 50 kilometers away, in a huge, windowless building in suburban Mahwah, N.J.

Another general strategy that high-frequency traders use is arbitrage. The basic idea behind arbitrage is that the prices of certain financial instruments are fundamentally linked. So if you find two things that ought to have the same value but are temporarily showing different prices, you should buy the cheaper one, or borrow and then sell the more expensive one. You’ll make a profit if the price of the cheaper one rises or the price of the more expensive one declines. Because it’s hard to know which will happen (or if the price of both will shift together because of other factors), the best strategy is to buy and sell the two things simultaneously. You’ll then profit in any event.

A concrete example here helps. Chicago has long been a hub for trading in futures contracts, which are agreements that give you the right to buy or sell something—corn or pork bellies, say—for a given price at some date in the future. On some Chicago exchanges, you can also buy and sell futures contracts on the stock of companies—often companies being traded on the New York Stock Exchange. The prices of those futures contracts are, of course, closely linked to the prices of the underlying stocks.

Now suppose the price of such a stock changes in New York—or rather, inside the servers in Mahwah, N.J. You can be pretty sure that the price of related futures contracts will shift accordingly in Chicago. But that won’t happen instantaneously: It will take 7 milliseconds or more for a report of the stock’s price change to reach Chicago through the fiber-optic cables linking these two financial centers. If you got wind of the news faster than that, your lightning-fast computerized trading machines could use “latency arbitrage” to profit. The TABB Group estimates that $21 billion is made with latency arbitrage every year. No wonder considerable engineering effort has gone into finding ways to transmit such news as fast as is technically possible.

In June 2010, Spread Networks, of Ridgeland, Miss., announced that it had installed a fiber-optic communications cable between New York and Chicago that followed an especially direct route. That required blasting though mountains at a cost of perhaps several hundred million dollars. Lease bandwidth on that line, the company suggested to high-frequency traders, and you’ll shave more than a millisecond off the time it takes you to send information between the two cities. Like the wire services that mobsters had once touted to bookies, Spread Network’s offer was something high-frequency traders couldn’t refuse—unless they just didn’t have the money to pay for it.

“You had to have it to do arbitrage,” says Bob Meade, who until 2010 ran a high-frequency trading group at Ronin Capital, a Chicago-based proprietary trading firm. Meade’s group couldn’t afford Spread Network’s charges, however. “A good part of my business was completely undermined by this new technology,” he says. But this unwelcome development got Meade—who had earlier earned a Ph.D. in physics from Harvard—thinking. In January of 2011, he teamed up with Stéphane Tyc, a buddy from his Harvard days who had also become a financial “quant,” and they applied some obvious physics to the problem at hand.

Meade and Tyc knew that electromagnetic radiation travels only about two-thirds as fast in glass fiber as it does in air, and they wanted to make use of that fact to create an especially fast communication link between New York and Chicago. They considered using shortwave radio signals, which follow Earth’s curved surface and can thus travel long distances. But they found that the only frequencies available for that were in the amateur bands, where business communication is forbidden. They also thought about relaying radio signals with high-altitude balloons, but they decided that was too crazy, even for them.

Eventually Meade and Tyc settled on a more straightforward solution: a chain of fixed microwave towers. They calculated that a properly engineered microwave connection could handily beat Spread Network’s fiber-optic offering. Then they formed a company, McKay Brothers, named after the Gordon McKay Laboratory for Applied Sciences at Harvard, to build it. “We thought we were the only geniuses in the world to think of this,” says Meade. But as soon as they examined the Federal Communication Commission’s database, they discovered otherwise. Meade says that there are now about a dozen companies working on microwave links that will compete with the one they expect to complete within a few months. (Meade declined to give a total cost for the project, but he did say that he and Tyc were funding the project themselves, without outside investors.)

Meade believes that his and Tyc’s backgrounds in both physics and finance give them an edge here. Others, he says, are designing their microwave systems very conservatively, as electrical engineers are apt to do: They site their towers within 50 kilometers or so of one another and shun over-water stretches, which sometimes prove a challenge for these radios. The McKay Brothers route goes over both Lake Michigan and Lake Erie and includes hops between towers that are often twice the normal distance limit. It also follows the shortest path possible over Earth’s surface (a great circle) or very nearly so. “It’s a 740-mile [1191-km] route, and we’re 4 miles from perfect,” says Meade. “We considered latency at every decision point.”

Won’t those latency optimizations compromise reliability? Perhaps. But Meade argues that high-frequency traders would much rather have access to a communications channel that’s faster than every one else’s, even if it gets flaky every now and then. A link that’s second or third fastest isn’t of much use to them, even if it’s always available. That’s a very different calculus than the one most engineers use—but it’s clearly the one you want to follow if you’re trying to get ahead of the pack.

What next? A fast link for transatlantic high-frequency trading. Hibernia Atlantic of Summit, N.J., is now working on a project to connect high-frequency traders in New York and London with an especially direct undersea cable, something it calls Project Express [PDF]. Mike Saunders, vice president of business development at Hibernia, says that the route is being surveyed now and that the new fiber-optic cable will be laid next year, as soon as winter storms on the North Atlantic abate, with the link becoming operational soon after.

This $300 million project will reduce the travel time for signals between New York and London to something less than 30 ms, which is a couple of milliseconds faster than any fiber-optic connection now in place. Saunders says that Hibernia has customers already signed up for the new link—all high-frequency traders. “There’s nobody else who will pay the same price,” he says.

The rise of high-frequency trading over the last few years certainly raises the technological bar for anyone attempting to make money off fleeting inefficiencies in financial markets—say, from short delays in how fast prices are calculated or from slight lags in the flow of information from place to place. You now need state-of-the-art servers located at the exchanges’ data centers, low-latency networking hardware, and the fastest possible long-distance data connections. Things have come a long way from a few telegraph wires linking racetracks with seedy back rooms.

Many people don’t see much point to the technological arms race that now lets only the best-equipped companies profit from high-frequency trading. Michael Wellman, a professor of computer science and engineering at the University of Michigan who studies electronic commerce and automated trading, has suggested that a better solution might be to move away from today’s continuous electronic trading toward a discrete-time market mechanism. This would be what those in finance refer to as a call market, but on a very short timescale.

According to Wellman’s concept, anyone could submit electronic orders to trade anytime, but no one would have access to information about those orders until they cleared at the end of some discrete interval, say one second. That would remove the advantage high-frequency traders gain from their microsecond-level optimizations, but it wouldn’t otherwise interfere with the trading of ordinary investors or other businesses. “You could run a call market fast enough that nobody would complain that they had to wait too long,” says Wellman.

To IIT’s Van Vliet, such proposals are unappealing, even anachronistic. “You can’t stop progress,” he says. But he also recognizes that there are dangers lurking in today’s highly automated trading infrastructure. If your firm’s trading system goes haywire, “not only would you lose a lot a money, you’d also destabilize the market,” he says.

To avoid such disruptions, Van Vliet argues, the financial industry needs to develop technical standards for its equipment and software—just as, say, the airline industry has done—so that it doesn’t put the public at risk. But it’ll surely take a while yet before protecting the public becomes a high priority on Wall Street or at the world’s other financial centers. “Trading has gone from a gambling problem to an engineering problem,” says Van Vliet. “But the culture of financial markets is still a gambling culture.”

The Conversation (0)

Video Friday: DARPA Subterranean Challenge Final

1 min read

This week we have a special DARPA SubT edition of Video Friday, both because the SubT Final is happening this week and is amazing, and also because (if I'm being honest) the SubT Final is happening this week and is amazing and I've spent all week covering it mostly in a cave with zero access to Internet. Win-win, right? So today, videos to watch are DARPA's recaps of the preliminary competition days, plus (depending on when you're tuning in) a livestream of the prize round highlights, the awards ceremony, and the SubT Summit with roundtable discussions featuring both the Virtual and Systems track teams.

Keep Reading ↓ Show less

Making 3D-Printed Objects Feel

3D-printing technique lets objects sense forces applied onto them for new interactive applications

2 min read

Researchers from MIT have developed a method to integrate sensing capabilities into 3D printable structures comprised of repetitive cells, which enables designers to rapidly prototype interactive input devices.


Some varieties of 3D-printed objects can now “feel," using a new technique that builds sensors directly into their materials. This research could lead to novel interactive devices such as intelligent furniture, a new study finds.

The new technique 3D-prints objects made from metamaterials—substances made of grids of repeating cells. When force is applied to a flexible metamaterial, some of their cells may stretch or compress. Electrodes incorporated within these structures can detect the magnitude and direction of these changes in shape, as well as rotation and acceleration.

Keep Reading ↓ Show less

How to Write Exceptionally Clear Requirements: 21 Tips

Avoid bad requirements with these 21 tips

1 min read

Systems Engineers face a major dilemma: More than 50% of project defects are caused by poorly written requirements. It's important to identify problematic language early on, before it develops into late-stage rework, cost-overruns, and recalls. Learn how to identify risks, errors and ambiguities in requirements before they cripple your project.

Trending Stories

The most-read stories on IEEE Spectrum right now