Attack of the Machines

Is your stockbroker a robot?

Every day, hundreds of reporters from CNBC, Bloomberg, Dow Jones, and other outlets concoct a story about the stock market. From the chaos of the New York Stock Exchange, they discern rational human behavior to explain why the S&P 500 rose precisely 11.46 points today.

Tech stocks up? Why, it’s because Intel’s CEO made positive comments. Oil stocks down? A respected analyst issued a bearish forecast. When stocks fall across the board, it is frequently attributed to investors “taking profits.” (Strangely, in the zero-sum game of investing, stocks never seem to rise due to investors “taking losses.”)

But last year, for the first time in the NYSE’s long history, a majority of the market moves may have been dictated by machines, not by human agents. “Program trading” has been around since the 1970s. It’s defined simply as “the simultaneous trading of a portfolio of stocks, as opposed to buying or selling just one stock at a time,” and defined technically by the NYSE as when somebody buys at least 15 stocks at the same time worth a total of more than $1 million. In casual parlance, program trading is when computer algorithms automatically trigger trades in response to market signals, such as price changes in a given stock.

According to the New York Stock Exchange, program trading for all of 2004 was a record 50.6 percent of volume, up sharply from 37.5 percent in 2003. (The most recent detailed weekly statistics can be found here, while weekly statistics for the last several years can be seen here.) In most weeks in 2000, for example, program trading accounted for only about 20 percent of volume.

Does this mean that half the activity on the markets today was generated by soulless machines? Not necessarily. A good chunk of program trading is initiated by humans. Big securities firms conduct program trades on behalf of clients—think of the efficiencies you gain when you buy cases of beer instead of single bottles. If a big investor wants greater exposure to the pharmaceutical sector, or a pension fund manager wants to put a big chunk of cash to work in the S&P 500, it will execute program trades through Goldman Sachs, scooping up lots of different stocks in huge quantities, all at once.

But a pretty big portion—it’s hard to say how much—of program trading seems to be on autopilot. In today’s markets, the purchase of one security can automatically trigger the sale of another group. When Wall Street firms sell the increasingly popular Exchange Traded Funds—single securities that represent batches of underlying stocks—to investors, they often automatically buy the shares underlying the index through program trades.

And frequently the trades are initiated at computers’ suggestions to exploit meager price differentials between securities. In the first trading week of January about 10 percent of the NYSE’s program volume was related to index arbitrage—when buyers simultaneously buy, say, a futures contract on a stock index while selling the stocks that comprise the index to profit from minuscule price differences.

Then there are the trading geeks, guys with black boxes in Lower Manhattan and Greenwich, Conn., who have written ultra-secret algorithms that dictate the purchase or sale of stocks whenever prices hit certain tripwires. In the past few years, quantitatively driven hedge funds have proliferated. And every day, the code on which they rely can trigger a buy and a sell on the same groups of stocks—sometimes several times a day. Thanks to program trading, a relatively small quantitative firm with only several hundred million dollars in capital can nonetheless account for a big chunk of the NYSE’s daily volume on a given day.

It all sounds like a recipe for disaster. In theory, sell programs could trigger other sell programs, and cataclysm could ensue. (Program trading was partly blamed for the 1987 stock crash.) But as program trading has risen, the market counterintuitively seems to have grown less volatile. The exchanges have built in mechanisms to stop program trading from moving the market once certain limits are breached. And there’s also a degree to which program trades are self-regulating. Every day, there are hundreds of different computer-driven strategies at work, and they frequently work at cross purposes. If IBM breaks through 100, it might simultaneously trigger a sell program at one hedge fund and a buy program at another.

The rise of the machines may undermine the journalists’ narratives of the market. But at another level it’s somewhat comforting. Program trading creates a highly unpredictable, occasionally unnerving universe in which sudden reversals can crop up out of nowhere, only to be followed by sudden outbursts of optimism. In other words, a market dominated by computerized program trades is a lot like real human life.