Leaving It to the 'Expert System'
Is the market maker of the future a computer? Automated Trading Desk, a proprietary shop in which computers do most of the trading, is betting it can make a buck as a wholesaler. The Charleston, S.C.-based broker dealer bought a flagging Chicago Stock Exchange specialist last year and is now running many of its trades through its system.
That system is a so-called "expert system," which uses artificial intelligence to mimic the thought process of a market maker or sales trader. By posting limit orders on ECNs and the New York Stock Exchange's DOT system the machine trades about 80 million shares per day in 2,000 Nasdaq and listed securities.
ATD was co-founded by Rutgers University Professor of Finance Emeritus David Whitcomb in the late 1980s. It was designed to apply his pioneering work in market microstructure to the trading of stocks. Market microstructure looks at how a market's rules affect the price formation process. Whitcomb is chairman of ATD. Steve Swanson is president and chief executive. The firm has about 75 employees.
Last July, ATD bought Chicago Securities Group, the third largest specialist on the CHX by number of securities traded, as part of a strategy to diversify its sources of order flow. In addition to the largely retail flow it expects to get from CSG's broker dealer customers, it has also begun tapping into the institutional market with a "guaranteed VWAP" execution program.
ATD has integrated its expert system into the order management systems of the Chicago Securities Group. Thinking and acting like a human trader, the system is charged with laying off CSG positions and representing customer orders.
"We're taking a conventional market making business and making it unconventional by using technology," said Whitcomb. "We think we can give better order execution than firms now give and make money as well."
"Making money" is not a phrase uttered by many wholesalers these days. ATD bought CSG at a time when many long-time players are getting out of the business altogether. NDB Capital Markets and Fleet Trading, for instance, shut down last year. Knight Trading, the biggest wholesaler, and Schwab Capital Markets are scrambling to replace lost retail order flow with more institutional flow.
Chicago Stock Exchange specialists have been hit hard as well. Top market maker Dempsey & Co. sold out to E*Trade last year. SG Cowen exited the business last month. Chicago Securities Group itself was losing steam when it was acquired by ATD. In 2000, CSG traded about 500 Nasdaq and listed names with a staff of about 50. Last year, it was down to just 200 or 300 stocks and 25 employees.
Diminished profitability is behind the retreat. Decimalization and the bear market are behind the diminished profitability.
Decimalization reduced the minimum trading increment from 6.25 cents to a penny. Dealer competition then compressed quoted spreads on the most active stocks to one cent. That made it impossible for dealers to profit on the difference between the price they gave their customers and the price they got when laying off the position.
Nasdaq Commissions
Institutional Nasdaq market makers have been able to stay afloat by charging commission-like fees. Retail dealers have responded by cutting back on payment for order flow and instituting small transaction fees. They still face a tough environment.
ATD is undaunted. It believes it can prosper in a penny-spread environment by making trading more efficient. A system programmed to think like a trader should be more adept at profitably laying off the often hundreds of small positions taken on during the day. That's the thinking anyway.
Most wholesalers offer their customers automatic executions on orders up to a certain size. To minimize their risks they try to lay off their new positions as quickly as possible in the market. But human traders are often ill-equipped to deal with the sheer number of transactions and digest all the necessary market data, according to Whitcomb.
"It's about speed and good handling," Whitcomb said. "If a good trader has the time to focus [on the layoff] he will make a profit. But the problem for the conventional market maker is that the volume and the number of issues are so huge that the trader cannot use all of his intelligence and focus all of the time." At ATD, computers process 95 percent of the trades, according to Whitcomb.
Whitcomb stresses ATD has been functioning like a market maker since its inception. When buying, it posts bids. When selling, it posts offers. It does not hit bids or lift offers. It is usually a supplier of liquidity, not a taker. "While our limit order is posted, we're making a market," Whitcomb said. "We're making money at an activity which is basically electronic market making."
The legal definition of a market maker, however, is a firm that posts continuous two-sided quotes. ATD does not. It is typically on either the bid side or the ask side. Most dealers though, Whitcomb notes, are typically only aggressive on one side. They may post both a bid and an offer in a given security, but only one side comes close to the National Best Bid and Offer (NBBO).
And while ATD is unlikely to commit capital for million-share blocks, it does guarantee execution prices for institutions. Its guaranteed VWAP service ensures that institutions or broker dealers will receive no worse than the volume-weighted average price for the day for a particular security or basket of stocks. ATD assumes the risk associated with the trade.
ATD's system does two things: It predicts prices and makes trading decisions. A very short-term price prediction model forecasts stock prices 30 seconds forward. It makes a new prediction for 2,000 securities every second. Using that information, the expert system determines when to trade; how much to trade; and at what price to trade. Said Whitcomb, "It tries to post limit orders that are aggressive so that the trade will get done, but intelligent so we won't be buying when the market is falling."
Assume, for example, ATD is a buyer of Microsoft and the forecast is for Microsoft to drop two cents within the next 30 seconds. The expert system will either decline to trade or post a bid three or four cents away from the inside. The minute the pricing engine flags the stock as neutral, the expert system will post an aggressive bid.
Unlike many so-called expert systems built for stock trading, ATD's system does not take as its inputs traditional technical or fundamental variables. It looks neither at 30-day moving averages nor cash flow ratios.
ATD bases its trading decisions on those variables encountered minute to minute by market makers and sales traders on trading desks. It looks at the quantity of bids and offers, for example, and tries to assess whether they are "real."
Fragmented Market
In addition, it synthesizes Whitcomb's pioneering academic research into market microstructure. For example, it takes into account the complex and fragmented Nasdaq market with its various ECNs and their various trading protocols. It asks the question: Should an order of this size for this particular stock be sent to Instinet or Archipelago?
"What a human does heuristically with ten issues, the machine does with 100 issues," explains ATD sales exec Derek Morris. ATD's trades account for 10 percent of all ECN volume, according to Whitcomb. The firm trades 80 million shares per day in 300,000 trades, giving it an average trade size of 267 shares. It trades about four percent of Nasdaq's daily volume and two percent of the listed market's. Largely because of CSG, it is now trading 20 million listed shares per day.
ATD actually began life trading listed securities on DOT. But by the mid-90s, that strategy was puking red ink. Fortunately, for ATD, Nasdaq promulgated the Order Handling Rules in 1996. That allowed limit order traders to post quotes in the Nasdaq montage via Nasdaq desks or ECNs. ATD prospered along with the Nasdaq technology bull of the 1990s.
That strategy too eventually hit a brick wall. Despite the fancy technology, ATD has not been immune to penny ticks and the bear market. In 2000, the firm netted about $66 million on revenues of $100 million. The following year, profits dropped to $49 million. For 2002, they are expected to slide even further to $25 million.
ATD is betting the additional order flow brought in by both CSG and its institutional efforts will enable it to improve its profitability. While it can do little about the direction of the market, it can make up for the narrower spreads by trading more volume.
But, why go to all the trouble to become a market maker? Why not just throw more capital at the prop trading business? "By simply changing a single parameter in our software we could move our volume tomorrow from 80 million to 160 million shares," Whitcomb said. "But we would lose money."
The exec explains that pouring more money into its core trading model would drive out all the inefficiencies that allow it to make a profit.
Becoming a market maker though exposes ATD to new risks. Its core prop trading operation doesn't have to trade if it doesn't want to. If its price prediction model indicates a falling Microsoft, ATD has the luxury of buying later.
The customer business affords no such luxury. It will build positions whether it wants to or not. "That's one of the costs of doing the customer business," said Whitcomb. "But the benefit is you get a lot of good order flow."
Whitcomb explains that most of the time the market is in neutral. It is neither rising nor falling. So, the market maker expects to make a spread. And because a machine is faster than a human trader, it can layoff those troublesome 100-share positions in a hurry. A human may not bother until the position hits 1,000 shares.
Whatever the case, Whitcomb is not over confident. Replacing human traders with a machine is a largely untested proposition in the dealing world. Ever the academic, he says: "The whole thing is a demo project. It will take six months or so before we have much of a sense of whether we can do what we say we can do. But it's a very important strategy for our future growth."
