Pragma Financial Systems wants to bring algorithms to the common man.
Pragma Financial Systems wants to bring algorithms to the common man. A five-man vendor based in New York, Pragma is marketing algorithmic trading software to broker-dealers and money managers. Today control of most algorithms is in the hands of a select few bulge-bracket shops. "The smaller broker dealers don't have the algorithms that the banks have," says Lee Maclin, a Pragma partner, "and their clients are going to the banks which are delivering the service cheaply. So, they feel pressure to implement it."
Pragma is not the only organization marketing algorithms to brokers. However, it is possibly the only vendor with a purely algorithmic trading product. Two big banks Credit Suisse First Boston and Banc of America Securities offer their trading technology to other brokers. One vendor FlexTrade Systems is building algorithmic trading functionality into its portfolio trading system.
Brokers without proprietary algorithmic trading infrastructures are opting for third party offerings. Both CIBC World Markets and Knight Capital Group, for instance, offer third party algorithms to their clients under so-called "white-label" agreements.
CIBC and Knight affix their brands to the services. The source of the technology need not be disclosed. CIBC gets its algorithms from Banc of America. Knight would not disclose its supplier.
Pragmatic POV
Pragma is targeting the buyside as well as the sellside. For the buyside, its plan is to license the technology. For the sellside, it may price on a transaction basis, according to Maclin. To that end, Pragma is in the process of establishing a broker-dealer.
Pragma's technology has been in production for about a year. A Pragma customer is "one of the largest users of algorithms on the Street," Maclin says, although he won't disclose the name. The firm has integrated its system with those of direct access suppliers Neovest and Lime Brokerage.
'The whole concept excited me right away.'
Lee Maclin, Pragma Financial Systems
Pragma was established in 2002 by Maclin, who is also a professor of mathematics at New York University's Courant Institute, and David Mechner, a quant trader at a hedge fund called Pragma. The technology was spun out of the hedge fund, which has since been closed. Pragma, as does its competitors, claims its technology is better than the rest. Underlying the software is a more "rigorously intellectual" framework, according to Mechner. The two founders based their work on the seminal paper "Optimal Execution of Portfolio Transactions." It was published in the Journal of Risk in 2000 by mathematicians Neill Chriss and Robert Almgren. Their work aimed to determine the best way to liquidate a basket of securities. It set the stage for the development of the arrival price algorithm, now gaining favor on Wall Street. "The whole concept excited me right away," said Maclin, who once taught alongside Neil Chriss at NYU. "I saw this as a continuation of portfolio theory." The arrival price algorithm, one of a handful pushed by the Street's largest brokers, is relatively new. Not all brokers have it. Those that do include CSFB, Morgan Stanley, Deutsche Bank and Merrill Lynch. Arrival Price Best? Pragma's technology lets users choose any algorithm they want. However, Pragma execs maintain traders are best served using arrival price rather than a VWAP or TWAP, for instance. "Everyone is crawling towards this solution phase," says Maclin of arrival price, "but very slowly. The systems they must mutate over time are so complex that the operational risk of suddenly having clients trade in a very computing intensive environment is very great. So they are moving very slowly." At the heart of Almgren and Chriss' paper is a formula that encapsulates what most traders know through experience and intuition: If you trade quickly, you will likely incur market impact. If you trade more slowly, you will incur less market impact, but more risk and possibly opportunity cost. Therefore, the optimal trading strategy is one that minimizes the expected cost and risk of a trade. In mathematical terms, that's "min E (x) + V (x)". E (x) is expected cost. V (x) is risk. Lambda, or , represents the trader's tolerance for risk. The calculation determines a trading trajectory, or schedule, that specifies how many shares should be bought or sold over a pre-determined time period. The shape of the trajectory will vary depending on the trader's lambda, or tolerance for risk.
'Everything comes back to E+Lambda V Minimization.'
David Mechner, Pragma Financial Systems
The trajectory for a trader with a high tolerance for risk will appear close to a straight line. He is more patient and willing to trade more slowly. He reduces his impact costs, but may suffer as the market moves away from him. The trajectory for a trader with a low tolerance for risk will be more sharply curved. The trader is less patient and is therefore more likely to conduct much of his trading up front. He increases his market impact costs, but lowers his opportunity costs. "Lambda is that parameter that allows you to tune this minimization to your exact trading style," explains Maclin. While the Almgren and Chriss paper was a breakthrough, the Pragma execs say it wasn't ready for prime time. To apply the work to real world trading, the execs had to "fix" five problems. First, the paper did not incorporate the concept of "alpha decay," or the fact that a trader's expected profits grow smaller over time. The trader typically makes more money in his first five minutes of trading than in his second. And he makes more in the second five minutes than in the third. And so on. Pragma built the concept of alpha decay into its system. Competitors have not, says Mechner. "This is one of our biggest value-added points," he notes. Second, the authors' assumption of how traders "feel" risk was at odds with how traders actually feel risk, according to Maclin. Third, the authors assumed that market impact increased in linear fashion with trade size. Actually, contends Pragma, market impact is less severe initially because New York Stock Exchange specialists act to mitigate it by aggregating small retail orders against large blocks. Fourth, the Almgren and Chriss paper did not take into account the vagaries of intra-day trading. There may be better times during the day to trade than others, such as near the opening and the close. Finally, the paper did not consider an "optimal compromise trajectory." That's when trading a basket, the combined trajectories of all the individual stocks may create a best-case trajectory. Almgren and Chriss ignored this compromise trajectory, looking only at individual trajectories.
'I saw this as a continuation of portfolio theory.'
Lee Maclin, Pragma Financial Systems
"There may be characteristics of their baskets that traders want to preserve as they trade," explains Maclin. "If someone is long value and short technology, they don't want some stock trading in ten minutes and another one trading over the next five hours. They want to preserve the original ratios." Strategies With the five holes plugged, Pragma execs claim their system can implement any algorithmic trading strategy, whether its VWAP, market-on-close or arrival price, etc. "Everything in our system comes back to E plus lambda V minimization'," says Mechner. "That's our real value added. We've organized our system around that. Whereas many other systems were built in ad hoc fashion." The exec contends that while many players built their arrival price algorithm taking into account E plus lambda V minimization', their other algorithms did not get constructed using as rigorous a framework. While some use E plus lambda V minimization' to support their market-on-close algorithms, none use it, for instance, for VWAP or market making. "We first built a framework for the E plus lambda V minimization'," Maclin says. "Then we built the algorithms on top of that. So every algorithm we have in the system makes the correct trade-off between cost and risk." At the core of Pragma's system are two main components-an optimization engine and a micro-trading engine. The optimization software, called RACE, for Risk Adjusted Cost Evaluation, determines the trading schedule. It evaluates millions of different possible trajectories to finally settle on one that produces the best cost-versus-risk profile. To derive a trajectory, RACE uses six months of historical data. Information is included from the NYSE and the INET databases. It will also use real-time data although it does not have access to the relationship of volatility to the bid-ask spread at the opening. Once a schedule is outlined it is up to the trading engine to implement the strategy. The micro-trading engine determines how, where and when orders are actually placed. Market orders? Limit orders? On the bid? On the ask? In between? Orders are typically no bigger than a few hundred shares. Trading can be slowed or speeded up depending on the level of progress. Traders view graphically the progression of their trades. A red line represents the trajectory. A creeping blue line represents the actual fills. How closely the blue line follows the red line determines the success of the strategy. So do brokers need something like this? CIBC and Knight decided they did. But not all are convinced that white labeling is the way to go. "We're not going to buy somebody else's algorithms," said Doug Rivelli, in charge of program trading at Weeden & Co., a mid-sized brokerage of 150 people. "Where is the value-added in that?" Rivelli maintains that brokers who white label algorithms offer their clients nothing that they can't get elsewhere. They do so largely to maintain a relationship. They also run the risk, Rivelli notes, of not having the most up-to-date algorithm. If the supplier of the algorithm tweaks the system he may not immediately make the change available to the white labeler. Weeden the Builder Weeden has built its own algorithms including a VWAP, a short-term momentum model and time optimization models. It does not offer a market-on-close algorithm. That's because it believes it can trade MOC orders better manually. Do Weeden's customers want to own their own algorithms? Many don't, says Rivelli, even the largest. Rivelli argues they are better served choosing from an array of offerings from brokers rather than bringing the technology in house. That relieves them of the continual updating and maintenance that a changing stock market often demands. The buyside does want customization though. It may prefer a more passive VWAP, for example, and ask Weeden to supply it. "We've built specific algorithms for clients," he says. "It provides them with customization, but it doesn't tie their hands."