For Your Eyes Only: In Undercover World of Intelligence Agency Desks Utilize New Technology

On Wall Street, firms guard intelligence secrets with a passion rivaled perhaps only in international espionage.

But the programmers and scientists guarding the secrets of so-called intelligent trading systems more closely resemble James Bond's pal Q the brainy inventor of 007's electronic gadgets than they do the British secret agent.

Since their introduction into Wall Street in the late 1980s, intelligence systems have made significant inroads into proprietary trading. Employing mathematicians and physicists, certain firms have used the systems based on mathematical models, or algorithms, that help process information and recognize patterns to locate mispriced securities and project future stock movements.

But recently, a number of trading firms have used intelligence software to trade stocks on an agency basis.

"Agency trading will have to go this way," said Dr. Mark Gimple, managing director of quantitative methodologies at New York-based Reynders, Gray & Co., an institutional agency-trading firm. "Investors are looking worldwide for liquidity, and they will need that technology to find it."

Gimple a Stanford University Ph.D. and former aerospace and biomedical scientist developed his firm's distinct intelligence system. With initial start-up costs of more than $400,000 for equipment and software, it took Gimple six months to get the trading system running.

Aimed at lessening market impact, the quantitative trading system is programmed to follow the daily volume in New York Stock Exchange-listed stocks. The system equipped to trade up to 6,000 client orders each day will trade for orders when a particular stock's volume is peaking during the day.

The system does not handle Nasdaq orders. "Because of the nature of the dealer market," Gimple said, "Nasdaq orders require a little more human attention."

Affectionately nicknamed "the walking algorithm" by his co-workers, Gimple said the system keeps track of multiple liquidity sources and accurately projects when volume in certain stocks will rise.

When the desk receives an order from a client at a pension fund, trust fund, institution or from a broker, Gimple or a Reynders Gray trader will give the client a first assessment of the market, hoping to get a sense of how the client wants the order to trade. Some orders may need to trade quickly, others can be stretched out over days. Based on the client's instructions, the desk will enter the order for execution within a certain time frame.

The desk then enters the order into the system, providing the system with a designated price range and time frame for execution. The system will execute an order within the designated parameters when it predicts volume will peak.

"We call it trading in the shadows,'" Gimple added. "We try to be present in the market, but we don't want our orders impacting the marketplace."

When the system determines an order should be filled, it will be sent to the NYSE's DOT system if it is a market order, or to SuperDOT if it is a limit order. The Big Board's electronic systems, in turn, route the orders directly to the specialist for execution.

Reynders Gray has eight equity traders in New York and Boston. Of the 1.5 million shares the desks average each day, almost half are executed by the trading system.

Across town from Reynders Gray's midtown Manhattan offices, New York-based Investment Technology Group (ITG) also utilizes a distinct quantitative algorithm to trade securities on an agency basis.

QuantEx ITG's quantitative trading system monitors market conditions and is programmed to make trading decisions based on weighted factors. Like the system at Reynders Gray, QuantEx will route orders directly to liquidity sources when market conditions are most favorable.

"It is almost like the system has a series of questions it asks thousands of times a day," said Suzanne Christensen, ITG's director of product marketing for QuantEx. "If all of the market conditions click, the system will trade the order."

In 1990, ITG acquired a small technology company, Integrated Analytics Company, that had developed a successful quantitative system. Hoping to transform it into a transaction system, ITG spent almost three years reprogramming the quantitative software.

Orders are entered into QuantEx during the day, and are routed to DOT, over-the-counter market makers, electronic communications networks (ECNs), crossing networks and brokers when market conditions fulfill the system parameters.

Aside from handling orders on its own agency desk, ITG has licensed 115 QuantEx links to 60 clients. The ITG clients with QuantEx links route orders directly to liquidity sources through ITG's routing server.

David Cushing, ITG's director of research, provides all coverage and strategy writing for QuantEx clients. Aside from its research responsibilites, Cushing's staff works with clients to help integrate QuantEx into the client's trading operations.

"Working so closely with clients, we can help identify what clients want on their end," Cushing said. "That lets us know what enhancements will keep our system up to date."

QuantEx clients route more than ten million shares a day, and ITG routes up to seven million shares daily on the intelligence system.

"What QuantEx does would be impossible to do on a human level," said Scott Mason, ITG's president. "It would take a large team of traders to trade the way it does. I can't see how agency desks will be able to continue to trade without some kind of intelligence system."

Algorithms, Quants and Expert Systems

An algorithm is a mathematical program that surveys inputted data and charts predictions based on that data.

Applied to trading, algorithm programmers believe that market behavior, although appearing very chaotic and random, follows a pattern. By analyzing various data sources and weighting those sources based on importance programmers hope a system can find a pattern and make predictions based on that pattern.

A genetic algorithm programmed to mimic the human thought process runs whole sets of algorithms together, constantly updating and modifying predictions based on changing inputs. A genetic algorithm attempts to solve problems through constant trial-and-error processes.

"Genetic algorithms are built with more parameters than simple algorithms," said Richard Bauer Jr., a finance professor at San Antonio's St. Mary's University, and author of "Genetic Algorithms and Investment Strategies," a book on the application of algorithm models on Wall Street.

Bauer said that genetic algorithms distinct from the most basic algorithms are more dependent on human programmers to produce forecasts.

"An algorithm can be generic and just react to information," Bauer added. "But because it has a memory, a genetic algorithm reacts the way humans would react if they could analyze data that quickly and efficiently."

A neural network as described by Dr. J. Doyne Farmer, a senior scientist at Santa Fe algorithm investment firm Prediction Company is a series of algorithms working together for more detailed predictions.

"You set up a neural network with algorithms connected in layers," Farmer said. "They work like nerves do, with the firing of one triggering processes in subsequent nerves."

Neural networks, according to Farmer, mimic the human brain's repetitive processes of measuring data, making adjustments and remeasuring data until predictions run smoothly. Once a certain algorithm makes a projection, it spurs the firing of a subsequent algorithm.

At Prediction Company, Farmer helped develop the detailed neural network that manages a portfolio of investments for international banking giant Swiss Bank Corp. Farmer started the company in 1991 with Dr. Norman H. Packard, his former University of California at Santa Cruz physics classmate.

But according to Lawrence Tabb, a group director at Newton, Mass.-based research company The Tower Group, many Wall Street firms have turned away from neural-network technology because of the complexity of the forecasting processes.

"Because the processes are so layered, I think people on Wall Street have backed away because they can't understand how the systems reach conclusions," Tabb said. "Data goes in, but a lot of people don't trust the answers that come out."

Tabb added that more Wall Street firms have turned to expert systems and quantitative models in recent years.

An expert system is a series of algorithms based on expert analytics. With rules derived from professionals in the designated industry, an expert system will process data based on those rules.

In trading, an expert system could be programmed with algorithms based on rules followed by certain traders. The system would rapidly process data and make predictions based on the preprogrammed trading rules.

A quantitative system or quant in trading uses complex algorithms programmed into the system as rules. Based on market conditions, like current prices and order flow, for example, the rules would ensure a desk's orders are handled within certain bounds.

"Often, a firm setting up for quantitative analysis knows what kind of results it wants, but it doesn't know how to program the system to achieve those results," Gimple said. "Quantitative systems may not actually trade, some just make sure a desk stays within its limits."

Intelligence Systems in Trading

Intelligence systems are used all over Wall Street, from exchange order books to money-management software and trading systems.

D.E. Shaw & Co. is the most well-publicized Wall Street firm using intelligence systems in proprietary trading.

Launched in 1988, D.E. Shaw utilizes algorithm models to exploit securities-pricing inefficiencies. The investment bank, founded by former Columbia University computer-science professor David Shaw, is reported to have spent over $100 million to research, construct and maintain its systems.

D.E. Shaw's mathematical models and risk-management systems try to identify pricing anomalies in over 100 worldwide markets. In recent years, the investment bank has expanded, creating a broker-dealer subsidiary for third-market trading and an international derivatives business.

Like most Wall Street firms using intelligence software, D.E. Shaw works to maintain a level of secrecy surrounding its trading systems. Few firms are open about their specific technology, and most keep volume figures hidden.

But according to Reynders Gray's Gimple, the growth in volume on the DOT and SuperDOT systems may in some ways mirror the development of intelligence trading over the last ten years.

In 1988, according to NYSE-published figures, the Big Board averaged more than 82 million shares each day over the DOT systems. That same year, more than 238 million shares were traded each day on the NYSE floor off of the DOT systems.

In 1993, DOT volumes jumped to 192 million shares a day, while almost 330 million shares traded off of the systems. Last year, volume on the DOT systems boomed, handling more than 475 million shares each day, close to the daily NYSE floor volume of almost 550 million shares for 1997.

Santa Fe's Prediction Company was one of the firm's that contributed to the growth in DOT and SuperDOT trading in recent years.

The firm's neural network makes all of the firm's investment decisions, and routes limit orders directly to SuperDOT. Prediction Company is bound by their contract with Swiss Bank Corp. from discussing its trading volume, but Farmer said since 1991, the firm has grown from just seven employees to more than 20 today.

Farmer added that Prediction Company has invested more than $10 million since 1991 in updating its neural network.

David Whitcomb, a Rutgers University finance professor, started an algorithm trading firm in 1988 that used to route limit orders to SuperDOT.

"I think we were the first firm to use an expert system to trade limit orders on the NYSE," Whitcomb said.

His Charleston-based Automated Trading Desk began trading limit orders on an agency basis for an investment bank. Although business remained profitable, the investment bank broke its relationship with Automated Trading Desk in 1995.

Today, Whitcomb's firm trades three million shares a day through its system for the 18 high net-worth customers of its brokerage subsidiary. Having spent millions to update its technology, the firm's expert system analyzes market information based on its clients' portfolios, makes trading decisions and routes orders to liquidity sources. The system now trades only Nasdaq securities, sending orders to SelecNet, SOES and licensed ECNs.

"Firms will have to do some of their trading this way," Whitcomb said. "Small trades should be done by expert systems because they can handle hundreds of orders at once. And human traders generally don't handle small retail orders with the same kind of attention they give to big blocks."

While still trading for its brokerage subsidiary, Whitcomb would like Automated Trading Desk to get back into trading on an agency basis as well. "It's our roots, and it kept us in business our first seven years," he added.

Agency Intelligence Systems

Although a number of agency desks are trading for clients with intelligence systems, it is by no means a standard practice among firms.

"That type of technology is used for proprietary trading, not agency trading," said Jim Hanrahan, head trader at New York-based Lynch, Jones & Ryan, a leading agency-trading firm. "Quants. Algorithms. That's business for a D.E. Shaw. What we do is follow customer orders."

Junius Peake, a University of Northern Colorado finance professor and Nasdaq market-structure expert, was unaware of any agency firm using intelligence software, but felt it was a logical and necessary development.

"I can't say I'm shocked agency desks are using that technology. It can do calculations that traders just can't," Peake added.

At ITG, Mason believes that although software and programming may differ greatly, intelligence systems will become the standard on Wall Street trading desks.

"There will always be agency trades that have to be handled the old-fashioned way, but intelligence trading will just continue to grow" Mason added.

Gimple agrees. "Investors are looking worldwide for liquidity, and they increasingly need technology to find it," he said. "A trader can't keep track of all the market information out there. You need the smartest technology to stay ahead."