At 9:30 A.M. on August 1, a software executive in a spread-collar shirt and a flashy watch pressed a button at the New York Stock Exchange, triggering a bell that signaled the start of the trading day. Milliseconds after the opening trade, buy and sell orders began zapping across the market’s servers with alarming speed. The trades were obviously unusual. They came in small batches of 100 shares that involved nearly 150 different financial products, including many stocks that normally don’t see anywhere near as much activity. Within three minutes, the trade volume had more than doubled from the previous week’s average.
Soon complex computer programs deployed by financial firms swooped in. They bought undervalued stocks as the unusual sales drove their prices down and sold overvalued ones as the purchases drove their prices up. The algorithms were making a killing, and human traders got in on the bounty too.
Within minutes, a wave of urgent email alerts deluged top officials at the Securities and Exchange Commission. On Wall Street, NYSE officials scrambled to isolate the source of the bizarre trades. Meanwhile, across the Hudson River, in the Jersey City offices of a midsize financial firm called Knight Capital, panic was setting in. A program that was supposed to have been deactivated had instead gone rogue, blasting out trade orders that were costing Knight nearly $10 million per minute. And no one knew how to shut it down. At this rate, the firm would be insolvent within an hour. Knight’s horrified employees spent an agonizing 45 minutes digging through eight sets of trading and routing software before they found the runaway code and neutralized it.
By then it was shortly after 10 a.m., and officials from the NYSE, other major exchanges, and the Financial Industry Regulatory Authority were gathering for an emergency conference call. It didn’t end until 4 p.m.
In the four years since the collapse of Lehman Brothers drove the global financial system to the brink of oblivion, new technologies have changed Wall Street beyond recognition. Despite efforts at reform, today’s markets are wilder, less transparent, and, most importantly, faster than ever before. Stock exchanges can now execute trades in less than a half a millionth of a second—more than a million times faster than the human mind can make a decision. Financial firms deploy sophisticated algorithms to battle for fractions of a cent. Designed by the physics nerds and math geniuses known as quants, these programs exploit minute movements and long-term patterns in the markets, buying a stock at $1.00 and selling it at $1.0001, for example. Do this 10,000 times a second and the proceeds add up. Constantly moving into and out of securities for those tiny slivers of profit—and ending the day owning nothing—is known as high-frequency trading.
This rapid churn has reduced the average holding period of a stock: Half a century ago it was eight years; today it is around five days. Most experts agree that high-speed trading algorithms are now responsible for more than half of US trading. Computer programs send and cancel orders tirelessly in a never-ending campaign to deceive and outrace each other, or sometimes just to slow each other down. They might also flood the market with bogus trade orders to throw off competitors, or stealthily liquidate a large stock position in a manner that doesn’t provoke a price swing. It’s a world where investing—if that’s what you call buying and selling a company’s stock within a matter of seconds—often comes down to how fast you can purchase or offload it, not how much the company is actually worth.
As technology has ushered in a brave new world on Wall Street, the nation’s watchdogs remain behind the curve, unable to effectively monitor, much less regulate, today’s markets. As in 2008, when regulators only seemed to realize after the fact the threat posed by the toxic stew of securitization, the financial whiz kids are again one step—or leap—ahead.
The Knight episode was “a canary in the mine,” says Michael Greenberger, a University of Maryland law professor and former regulator at the Commodity Futures Trading Commission (CFTC). “We’ve been lucky so far that this hasn’t been more serious.”
Knight wasn’t the worst-case scenario. Not even close. A lot of high-frequency trading is done by small proprietary trading firms, subject to less oversight than brand name financial institutions. But big banks have also tried to get in on the act. Imagine a runaway algorithm at a too-big-to-fail company like Bank of America, which manages trillions, not billions, in assets. Or, says Bill Black, a former federal regulator who helped investigate the S&L crisis of the ’80s and ’90s, imagine trading algorithms causing “a series of cascade failures”—like the domino effect that followed Lehman’s collapse. “If enough of these bad things occur at the same time,” he says, “financial institutions can begin to fail, even very large ones.” It’s not a question of whether this will happen, Black warns. “It is a question of when.”
Years of mistakes and bad decisions led to the 2008 collapse. But when the next crisis happens, it may not develop over months, weeks, or even days. It could take seconds.
Alpha, New Jersey, is a sleepy hamlet in the Lehigh Valley, near the Delaware River. Somewhere in town (the owners won’t say exactly where) is one of 10 2,000-square-foot amplifier facilities that dot the landscape every 75-or-so miles between Chicago and New York City, ensuring that fiber-optic signals travel between the two points as clearly and quickly as possible. Spread Networks, the firm that operates the facility, may have seen some poetry in the community’s name—”alpha” is the term investment managers use to describe the performance of an investment after adjusting for risk.
Spread is part of a growing industry dedicated to providing hyperspeed connections for financial firms. A faster trader can sell at a higher price and buy at a lower one because he gets there first. A connection that’s just one millisecond faster than the competition’s could boost a high-speed firm’s earnings by as much as $100 million per year, according to one estimate.
Because of this, trading firms are increasingly pushing the limits to establish the fastest connections between trading hubs like New York, Chicago, and London. Every extra foot of fiber-optic cable adds about 1.5 nanoseconds of delay; each additional mile adds 8 microseconds. That’s why companies like Spread have linked financial centers to each other by the shortest routes possible. Spread’s Alpha facility is one of more than a dozen similar centers arrayed along the path of its 825-mile-long, $300 million fiber-optic cable between Wall Street and the Chicago Mercantile Exchange. Spread reportedly charges traders as much as $300,000 a month to use its network. Exchanges like the NYSE charge thousands of dollars per month to firms that want to place their servers as close to the exchanges as possible in order to boost transaction speeds. Industry experts estimate that high-speed traders spent well over $2 billion on infrastructure in 2010 alone.
Traders’ need for speed has grown so voracious that two companies are currently building underwater cables (price tag: around $300 million each) across the Atlantic, in an attempt to join Wall Street and the London Stock Exchange by the shortest, fastest route possible. When completed in 2014, one of the cables is expected to shave five to six milliseconds off trans-Atlantic trades.
But why stop there? One trading engineer has proposed positioning a line of drones over the ocean, where they would flash microwave data from one to the next like the chain of mountaintop signal fires in The Lord of the Rings. “At what point do you say, ‘This is fast enough’?” asks Brent Weisenborn, a former NASDAQ vice president.
The acceleration of Wall Street cannot be separated from the automation of Wall Street. Since the dawn of the computer age, humans have worried about sophisticated artificial intelligence—HAL, Skynet, the Matrix—seizing control. But traders, in their quest for that million-dollar millisecond, have willingly handed over the reins. Although humans still run the banks and write the code, algorithms now make millions of moment-to-moment calls in the global markets. Some can even learn from their mistakes. Unfortunately, notes Weisenborn, “one thing you can’t teach a computer is judgment.”
One set of signals the programs have to weigh are countless trade orders other algorithms send out and then quickly rescind. There’s a fierce debate about what these abortive trades might be. Some speculate they are new algorithms being tested or strategic feints, the equivalent of sonar pings probing the market for a response. Some of the fake trades could be aimed purely at gobbling up bandwidth to slow down competitors. “There are doubtless former [high-speed traders] who could tell us,” Black says. “If I worked for the CFTC or the SEC I would be seeking them out to try to learn what was going on.”
On the afternoon of May 6, 2010, CNBC viewers could have mistaken the channel’s programming for an apocalyptic blockbuster. The Dow, already down 400 points on bad news from Europe, had suddenly plummeted another 600. Erin Burnett, wide-eyed, gesticulated at charts to illustrate the “unprecedented” 1,000-point drop. The typically manic Jim Cramer reached a new level of frenzy, shouting at viewers to buy—BUY!—Procter & Gamble, which had fallen 25 percent, and wagging his finger at the screen: “If that stock is there, you just go and buy it. It can’t be there. That’s not a real price!”
Prices of nearly every stock and exchange-traded fund had plunged in minutes. Some 300 securities experienced wild gyrations, with trades executed at prices as low as a penny and as high as $100,000 a share. During the same second, shares of the consulting firm Accenture traded at both $0.01 and $30.
In what was later dubbed the “flash crash,” nearly $1 trillion in shareholder value was wiped out in a matter of minutes before the market rebounded, eventually closing down 3 percent from the previous day. Almost five months later, regulators would conclude that, on a day when traders had already been shaken by the Greek debt numbers, a single massive sell order executed by an algorithm belonging to a firm in Kansas had triggered a series of knock-on events that sent the market into a tailspin. The analysis portrayed “a market so fragmented and fragile that a single large trade could send stocks into a sudden spiral,” the Wall Street Journal reported.
The flash crash spurred regulators to action—but spurs can only make a horse gallop so fast. No one in Washington makes an extra million bucks a year for moving a millisecond faster, and it shows. So far, Congress and the nation’s financial watchdogs have done more hand-wringing than regulating. In classic Washington fashion, when a Senate subcommittee held a hearing in late September on the “rules of the road” for algorithmic trading, the only consensus to emerge was that more hearings were needed.
“Thanks to technology, our securities markets are more efficient and accessible than ever before,” then-SEC chair Mary Schapiro said at an October market technology roundtable. “But we also know that technology has pitfalls. And when it doesn’t work quite right, the consequences can be severe. Just imagine what can happen if an automated traffic light flashes green rather than red, if a wing flap on a plane goes up rather than down, if a railroad track switches and sends the train right rather than left.”
Politicians and regulators realize there’s an issue, but by the time Washington gets a handle on the situation, some experts fear, the damage may already be done. “We’re always fighting the last fire,” says Dave Lauer, a market technology expert who has worked for high-speed trading firms.
If it’s a fire the SEC needs to fight, the agency is working with equipment that’s more reminiscent of bucket brigades. The New York Times has called regulators’ tech “rudimentary.” David Leinweber, the director of the Center for Innovative Financial Technology at Lawrence Berkeley National Laboratory, has slammed the SEC and the CFTC for running an “IT museum”—and taking nearly five months to analyze the flash crash, which was essentially over in five minutes.
To enhance its market-monitoring capacity, the SEC has had to turn to industry—specifically, a firm called Tradeworx that specializes in very-high-speed trades—for a new computer program to analyze trading data. That program, called Midas, was scheduled to go online at the end of 2012. But even Midas won’t give the SEC a comprehensive picture of the markets. It offers no data on so-called “dark pools,” private markets where buyers and sellers can trade anonymously, and it won’t tell the SEC who is responsible for a given trade. To fill those gaps, the SEC plans to ask market participants to submit comprehensive information about every trade in the US markets—creating what is called a consolidated audit trail. But the SEC won’t receive this information in real time. Instead, the audit information will be due by 8 a.m. the next day.
“When that data does come in, since we have every single step, we will be able to reconstruct it exactly as it happens,” says Gregg Berman, an ex-physicist and SEC adviser who led the agency’s inquiry into the flash crash. “The only thing we miss is the opportunity to do something the same day. But given that a robust and defensible analysis of even a small portion of the trading day can itself take many days, we don’t give up much by waiting until the next day to receive a complete record of the day’s events.” Studying this market data will help the agency develop rules to address problems in the market—but only after they occur.
Meanwhile, the financial world is getting even more fast-paced, opaque, and downright mysterious. The same week Schapiro spoke at the SEC roundtable, an algorithm consumed 10 percent of the bandwidth of the US stock market. It “ran like a bat out of hell on crystal meth with a red bull chaser, to mix a few metaphors,” Leinweber wrote on his Forbes blog. “It generated 4% of U.S. stock market quote activity,” but the program “didn’t make a SINGLE TRADE, cancelling every order. That is pretty darn weird.” Leinweber suspects that the culprit was a new algorithm being tested, but that’s just a guess—no one knows for sure, least of all the SEC. It used up “10% of the communications capacity of our overly wired market,” Leinweber noted. “Ten of these guys could use the whole market…Scary stuff.”
So far, the problems caused by algorithms appear to be mostly accidental. But what if someone designed a program intended to wreak havoc? John Bates, a computer scientist who, in the early 2000s, designed software behind complicated trading algorithms, worries that the kind of tools he’s created could end up in the wrong hands. “Fears of algorithmic terrorism, where a well-funded criminal or terrorist organization could find a way to cause a major market crisis, are not unfounded,” he wrote in 2011. “This type of scenario could cause chaos for civilization and profit for the bad guys and must constitute a matter of national security.”
Ask the Wall Street lobbyists about things like cascade failures or algorithmic terrorism and they’ll tell you not to worry. They’ll note that transaction costs have never been lower and that the average investor can execute trades faster and cheaper than ever before. In their view, the fact that Knight lost $440 million and didn’t take the rest of the financial sector down with it suggests that the market isn’t nearly as fragile as detractors claim.
Thanks to these arguments, and the nearly $200 million Wall Street spent lobbying Congress around the Dodd-Frank financial reform bill in 2010, that law did almost nothing to regulate high-speed trading. In the absence of actual rules, the most widely discussed safeguards are now the “kill switches” or “circuit breakers” that kick in when a certain threshold is breached. After Black Monday in 1987, when the Dow Jones dropped by nearly a quarter in one day, the New York Stock Exchange instituted circuit breakers that halt trading temporarily when the market falls by 10 percent and shut it down entirely when it falls by 30 percent. Neither of these fail-safes, though, was triggered by the flash crash—the market fell in a blink, but it fell less than 10 percent.
After the flash crash, the SEC implemented new circuit breakers that kick in when an individual stock experiences rapid, unusual price swings. But those didn’t prevent the Knight debacle—it was mostly trading volume, not unusual prices, that cost the company hundreds of millions. New SEC rules slated to take effect in February will halt trading for five minutes if prices of individual stocks move outside of a set range for more than 15 seconds. But those are “a Band-Aid,” complains Lauer, the technology expert. “You’re treating the symptom, not the cause.”
Most participants at the SEC’s October market tech roundtable endorsed the idea of installing more kill switches at various levels—for individual firms, individual stocks, and perhaps for the market as a whole. But there’s a problem: If a kill switch or circuit breaker is automatic, it does nothing to reintroduce human judgment. Conversely, if a person has to pull a kill switch, he or she has to take responsibility for doing so—which creates its own problems. “No one wants to be the guy who cried wolf and got you onto the front page of the Wall Street Journal,” Black says. “The word that’s going to be used is that [you] panicked.”
So, if kill switches and circuit breakers don’t prevent future problems (and they haven’t before), how do you avoid the algorithmic apocalypse? Reformers are advocating what amount to speed limits. One of their proposals involves implementing what could be viewed as a temporary “no backsies” rule, requiring trading firms to honor the prices they quote for a minimum amount of time unless they execute the trade or make a better offer. Even a minimum quote life of just 50 milliseconds “would have eliminated the flash crash,” says Eric Hunsader, the CEO of Nanex, a company that makes software for high-speed traders.
In a more far-reaching proposal, Rep. Peter DeFazio (D-Ore.) and Sen. Tom Harkin (D-Iowa) have proposed levying a financial-transactions tax—they suggest 0.03 percent—on each trade, as a way of discouraging churn and raising revenue. (The United States had such a tax until 1966.) Economists, activists, and even some finance big shots—Warren Buffett among them—have endorsed the idea. “Even at the modest level we’ve proposed, [the tax] would raise $35 billion a year, which would either be used to defray the deficit or be used for job-creating investments by the government,” DeFazio told me. Eleven European Union countries (though not the United Kingdom) are pressing ahead with the idea—and they’ve talked about a tax as high as 0.1 percent. Wall Street lobbyists have pushed back against both speed limits and bringing back the transaction tax. But in the wake of the Knight episode, some industry experts are expressing doubts about the status quo.
“I believe this latest event was handled better than the flash crash, but the larger question is whether our markets are adequate to deal with the technology that is out there,” Arthur Levitt Jr., a former chairman of the SEC and a dean of the financial establishment, told the New York Times in August. “I don’t think they are.” That view is becoming more widely accepted, even among corporate CEOs, traders, and the algorithm builders themselves.
The chief executives of publicly traded companies—who are hired and fired based on stock prices—increasingly worry that their shares could be sent into a free fall by an algorithmic feeding frenzy. The current markets have created a “somewhat disjointed world between what a company does and what its stock does,” the CEO of one billion-dollar, NYSE-traded company told Mother Jones.
According to Ben Willis, a longtime NYSE trader, “When you have the heads of the Fortune 500 companies say, ‘Hey, wait a minute, guys: Our stocks look like hell and…no one can tell me with any certainty who is doing what to my stock and why,'” the critics might gain political momentum. Then again, the financial sector has a pretty solid track record of stymieing reform. And, given the extent to which the international financial markets are intertwined, would slowing down Wall Street make a difference if similar measures weren’t taken in London and Hong Kong?
As market-shaking episodes pile up, even some of the tech geniuses who helped usher in Wall Street 2.0 now worry about their innovations running amok. Wall Street Journal reporter Scott Patterson’s book on high-speed trading, Dark Pools, recounts the story of Spencer Greenberg, a young math genius who built a hugely successful trading algorithm named Star but later came to have reservations about what he had unleashed on the world. “In the hands of people who don’t know what they’re doing,” Greenberg warned a gathering of algorithmic traders in 2011, “machine learning can be disastrous.”