Cracking The Street's New Math
Ted Oberhaus used to have one big headache. As director of equity trading at the $93 billion Jersey City-based Lord Abbett & Co. mutual-fund group, he typically needs to buy and sell about 7 million shares a day. Trouble is, the average-size deal on the New York Stock Exchange has fallen to just 400 shares, down from 1,477 in 1998, when prices were last quoted in fractions, according to market researcher The TABB Group. As a result, if other traders get wind of what Oberhaus is up to, market prices could move swiftly against him. "We're trying to put elephant-size orders through a keyhole," says Oberhaus.
These days, thanks to so-called algorithmic programs executed on high-powered computers, Oberhaus and other professional money managers can ram orders for an average trade of 155,000 shares through such small openings. Simply put, an algorithm -- first identified by a 9th century Arab mathematician -- is a series of calculated steps strung together to produce a desired numeric goal. Now, these formulas are used to buy and sell large blocks of stock that have been traditionally shopped around traders on the floor of the NYSE or through Wall Street brokers.
Trading algorithms automatically break up large orders into bite sizes and feed them directly into the market. But they can be tuned to execute almost any strategy. Some aim simply to capture the average price over a day, others to gain an edge by, say, trading more heavily at the opening and close, when volume is high, and less when it slows down around lunchtime. And they can be customized, for example, to sell stock stealthily over several weeks if a manager holds a 3% position in a particular stock and wants to cut it to 1%. Not only is algorithmic trading in principle more secure than using a human broker but it also costs less than a penny a share to trade electronically, vs. 6 cents for full-service trades, which include research.
Wall Street sees algorithms as a way to recapture the large volume of commission revenues lost to rival electronic exchanges. In addition, banks and brokers are selling new services that allow money managers to test the performance of the trades before and after they are executed to help managers improve their trading techniques. In the past 18 months or so, every major broker (and a handful of independent software vendors) have spent millions ramping up the business, including Merrill Lynch (MER ), Citigroup (C ), JPMorgan (JPM ), and HSBC (HBC ). Others have bought into the business. Last year, Bank of New York Co. (BK ) and Piper Jaffray & Co. (PJC ), respectively, bought algorithmic specialists Sonic Financial Technologies LLC and Vie Securities LLC.
Algorithms are a step up from the more familiar program trading, which institutions for years have used to buy or sell bundles of 15 or more stocks worth a combined $1 million. Algorithms handle trades in individual stocks, and the exchanges don't ban their use when trading becomes extremely volatile, as they have done with program trades since the 1987 market meltdown. As the use of algorithms moves from hedge funds and Wall Street's trading desks to mutual- and pension-fund managers, it will account for more than 40% of total U.S. equities trading on all markets by 2008, up from about 25% today, according to Boston-based researcher Aite Group.
But as with all Wall Street feeding frenzies, there are dangers. Some critics say that when less experienced hedge- or mutual-fund traders use the software they've bought from Wall Street, they inadvertently expose their trades. How? Canny traders, mainly those who trade on behalf of big banks and brokerages with the firms' capital, may be able to identify patterns of algorithms as they get executed. "Algorithms can be very predictable," says Steve Brain, head of algorithmic trading at Instinet (INGP ), the New York City-based institutional broker.
Coupling that with a few phone calls to chat about order flows and some traders may be able to piece together enough information to "front-run," or profit by trading ahead of the customer. "There are people out there who seek to reverse-engineer algorithms," says J. Mark Enriquez, chairman of Pulse Trading Inc., which operates an aggregation service that can tap into all available ECNs and alternative exchanges. "The larger firms will swear on a stack of Bibles that there is no info leakage...[but] given the history on Wall Street, there is a potential for someone to figure out how to access" details about algorithmic trades. The chief executive of a leading alternative exchange, who spoke on the condition of anonymity, says: "These tools are in effect a Trojan horse."
Many Wall Street firms strongly deny that details of client orders ever circulate outside their algorithmic groups -- or that algorithmic patterns can be traced. "We live or die by how well we trade," says Dan Mathisson, Credit Suisse First Boston's (CSR ) global head of advanced execution services. "A significant deterioration in performance would kill us; there are no information leakages." At JPMorgan, the bank's own traders are separated physically from the algorithmic unit, and the computer systems aren't shared either, according to Emily Portney, chief operating officer of JPMorgan's equities Americas business.
Some critics simply question Wall Street's wisdom in investing large sums in algorithms. Says Randy L. Grossman, capital markets research manager at Financial Insights in Framingham, Mass. "Many firms have spent millions of dollars to develop algorithmic capability. Many, we believe, have not done a realistic return-on-investment analysis of this tool."
All the same, some of the bigger firms are looking into ways to use algorithms more widely. They're mulling systems to trade futures, currencies, international stocks, or baskets of stocks. Already, algorithms are having a huge impact on trader productivity. Banc of America Securities LLC (BAC ), for example, has slashed the number of its traders by almost half in the past two years while increasing its overall equity trading volume by 160%. Today about half its trades are made with algorithms, up from none two years ago.
Regulators are aware of the potential for algorithmic trading problems. The NASD, with the blessing of the Securities & Exchange Commission, is collecting documents and interviewing traders at several major brokerages to learn more about the programs and their potential for abuse. Unlike ongoing probes of alleged front-running at the NYSE and specialist firms, the inquiries on algorithms aren't full-blown enforcement investigations -- and won't be unless they turn up evidence that brokers are abusing the computerized orders. For now, at least, money managers seem quite happy to embrace algorithms to ease their trading headaches -- and cut their costs.
By Mara Der Hovanesian in New York