article

08.07.12

How Wall Street Computers Almost Killed Knight Trading

Knight Trading almost put itself out of business when software ran wild and executed millions of trades. As Alex Klein and Matthew Zeitlin report, the Knight debacle is only the latest example of how machines run our stock markets—and our financial lives.

Early Monday morning, NASA’s Curiosity probe flew unguided through seven minutes of terror, autonomously deploying parachutes, rockets, and robotic cranes to land safely and softly on the Martian surface—a triumph of automation and brilliant engineering.

Back on Earth, financiers, Wall Street executives, and regulators were dealing with the latest failure of automation and shambolic engineering. Last Wednesday, market-maker Knight Capital lost $440 million when its putatively sophisticated market software malfunctioned, and began making millions of unwanted trades. When Knight’s high-frequency electronic trades shuffled shares in 148 major companies, the tech glitch sent markets into turmoil.

Knight is one of those obscure firms that plays a vital role in the functioning of the stock markets. Based in Jersey City, N.J., a few miles from the heart of Wall Street, it is one of the largest market makers and traders in the country, helping to facilitate trades for investors around the world. Last Wednesday, as the company was trying out new software when the markets opened, the stock was worth just over $10 a share.

Soon after the market opened, Knight began to send out orders to buy and sell stocks in massive quantities. According to a New York Times analysis, at 9:58 a.m., trading volume on the New York Stock Exchange was six times more than average; just 17 minutes later, it was back to normal. While Knight is famous for trading the stock of companies like Apple, the 45 minutes Wednesday morning saw wild fluctuations in companies that don’t see normally anywhere near as many trades.

Here’s the scary thing: although the trades started piling up in the minutes after the market opened, there was no simple way for Knight to stop the algorithm from running rampant. If there was an off switch, nobody pushed it. And so some thinly traded stocks saw wild rides up that confused everyone on the floor of the stock exchange, including Knight’s own traders. Thinly traded stocks, like China Cord Blood, went up some 143 percent before the New York Stock Exchange canceled the transactions.

The rogue software nearly destroyed the company. By the end of the day, major customers stopped routing their trades through Knight. The company wound up with hundreds of millions of dollars worth of stock it had inadvertently acquired. When the Securities and Exchange Commission refused to cancel the orders, Goldman Sachs (naturally) stepped in and offered to take the unwanted positions from Knight—at a discount. The upshot: in one day, Knight racked up $440 million in losses, a sum greater than its second-quarter profit.

In an SEC filing, Knight described the error as a “technology issue related to the company’s installation of trading software.” The “issue” almost killed the company.

The fact is, the lion’s share of our assets are controlled, traded, and monitored by machines. Not people. And machines can break down.

On Monday, Knight Capital, however, announced a deal to rescue the firm from a prospective bankruptcy. The investment bank Jeffries Group; Blackstone, the mammoth private-equity firm; and TD Ameritrade, the broker and major Knight customer, were the major investors in a $400 million deal that will allow them to purchase Knight stock at $1.50 a share and would give them control of 73 percent of the company. (Remember, Knight’s stock traded at $10 less than a week ago. Here’s a five-day chart.)

Once is a phenomenon, twice is a coincidence, and three times is a trend. We’ve definitely got a trend of machines running amok on Wall Street.

On May 6, 2010, the Dow lost about 600 points in five minutes—then gained it all back again. The “Flash Crash” culprit? Algorithms gone rogue. It started when Waddell & Reed, a large mutual fund, hit a button to sell about $4.1 billion in stock futures contracts. Its trading algorithm had a fixed sell rate—9 percent of the contracts’ total trading volume—and began flooding the market. Other high-frequency trading firms bought the devalued contracts automatically, but quickly had their own “sell” algorithms triggered by the volatility. A resale bonanza began—essentially, computers trading with computers. The contracts whizzed back and forth, inflating trading volume and crushing prices. As the price of the futures crashed, the spike in buying and selling triggered an automatic shutdown of many electronic traders in the stock market. With liquidity so low, share prices began to wildly fluctuate. Shares of the consulting firm Accenture fell by 99 percent, and shares in Sotheby’s rose 3,000-fold, from $34 to $99,999.99.

The crisis abated only when, at 2:45 p.m., trading in the futures contracts was paused for five seconds by the Chicago Mercantile Exchange’s “Stop Logic Functionality.” Only then did real buyers have time to grab the reins and claim trading back from the army of machines. The computers had taken the Street for one hell of a ride. And it wasn’t difficult—as The Wall Street Journal’s Scott Patterson put it, “Between 60 and 70 percent of all trading on Wall Street is done by automated high frequency trading computers.”

Earlier this year, another technologically driven trading company was nearly undone at the hands of its own software. On March 23, BATS Global Markets—a start-up electronic exchange based in Kansas City that caters to high-speed traders—was about to stage its initial public offering. As soon as BATS put $100 million of its own new shares, originally priced at $16, up for sale, the stock instantly fell to pennies. Why? BATS’s trading software freaked out at the prospect of trading its own shares, and began botching trades on all NASDAQ stocks with A or B ticker symbols. When trading on BATS shares hit 4 cents, the software’s “circuit breaker” stopgap kicked in, shutting down the IPO entirely. “It was going to be a bloodbath,” one trader told The Wall Street Journal. The offering was canceled.

And it’s not just alternative exchanges that are prone to computer foibles. The venerable NASDAQ, host to the century’s most anticipated IPO, ended up with egg on its face after a technical glitch sullied Facebook’s debut. Software errors delayed the IPO more than two hours, and left investors unsure if their purchases had gone through. As underwriters and investors lost millions in the confusion, NASDAQ put its tail between its legs and proposed giving out $40 million to aggrieved brokerages. One of them—none other than Knight Capital—slammed the offer and threatened to sue.

Knight Capital, BATS, and a “flash crash” may make headlines only for a day or two. And it’s hard to resist the human drama of chaotic losses and botched debuts. But behind images of addled traders, these gaffes illustrate a larger issue, one that affects everyone with a stake in the markets—that is, all of us. The fact is, the lion’s share of our assets are controlled, traded, and monitored by machines. Not people.

And machines have a tendency to break down.