Algorithms are widely recognized as one of the fastest moving bandwagons in the capital markets. And everyone is jumping on board. Westborough, Mass.-based TABB Group projects the total portion of buy-side orders flowing through algorithms will reach 8 percent globally by 2007, which translates to a hefty compound annual growth rate of 63 percent.
With this kind of anticipated growth, it is no surprise that sell-side institutions are jockeying for market position with a veritable glut of product offerings. For instance, when Instinet launched its dark liquidity aggregation algorithmic strategy, Nighthawk, in June, it entered a marketplace cluttered with similar strategies, including Fusion from Piper Jaffray, Dark Server from Investment Technology Group and the long-established Guerilla from Credit Suisse.
The constraints on the market for algorithmic strategies clearly are not from the supply side, observes Jonathan Cohn, senior manager in the strategic IT practice at Mercer Oliver Wyman (New York). "The constraint is really from the demand side," he says. "There's the issue of, strategically, how the asset managers are going to use these algorithms in their business. ... The flood of options is growing for them every month."
In a market that saturates so quickly, being the first to innovate can give a broker a significant advantage over the competition both in capturing the order flow of early adopters and building a reputation as a thought leader. And maintaining an efficient development process is critical to effectively managing time to market.
Making the First Move
"It is possible to create an algorithm and enjoy a significant time window ahead of the competition if that algorithm addresses a really unique execution strategy," notes Jeff Wecker, CEO of Townsend Analytics, the Chicago-based data and trading technology provider owned by Lehman Brothers. He is quick to add, however, that "as more broker-dealers enter the market with more and more algorithms, and as vendors do a better job of delivering algorithms, that first-mover advantage is getting smaller and smaller."
Further, explains Wecker, there are parties other than the broker-dealer that have a meaningful impact on time to market. Institutional investors must have some method of interacting with a trading strategy, which generally requires that the algorithm be integrated into an order management system (OMS) or execution management system (EMS). Facilitating this process requires strong relationships with the vendors of these systems and a homogenization of the technical parameters of algorithmic offerings, Wecker says.
"In addition to working through and providing the API [application program interface] so that an OMS or EMS vendor can access the algorithm, you have to do the work and have the relationships with those vendors to get in their development queue," says Wecker. "One way to dramatically simplify and shorten the time frame for the delivery of an algorithm is to make the [parameters] for your algorithms look very similar," he adds. "Another way the broker can reduce the time to market is to work with one or two vendors who they use religiously to showcase their stuff."
The implementation challenge is not limited to white-labeled algorithms, such as Townsend's offerings, that are integrated into third-party vendor systems. Even for large firms integrating homegrown algorithms with proprietary systems, such as Charlotte, N.C. - based Banc of America Securities (BAS), the implementation of algorithms into the front-end system is a resource-intensive process. For BAS' August release of the Ambush algorithm, designed to execute large orders with minimal market impact, integration into the firm's proprietary trading platform required considerably more manpower than the product's design, reports Bill Harts, head of strategy for equities.
"In the case of Ambush, there were clearly a lot more people working on the actual implementation than were dedicated to the design of it," Harts says. Although he declines to specify the number of people involved in the process, Harts notes that BAS' electronic trading services team, including front-office systems and implementation, is more than 100 people strong.
For small broker-dealers without those kinds of resources, playing second fiddle seems to be inevitable. "This does take quite a bit of academic and development horsepower -- and cash," notes Mercer Oliver Wyman's Cohn. "The interesting stuff that you're seeing pumped out from people ... tends to be in the medium to large [firms]."
But one firm bucking this trend is New York-based agency broker-dealer Miletus Trading, which specializes in algorithmic trading strategies. Miletus keeps up with its well-funded competitors with a strategy of high-end customization of its algorithms.
"One thing we do to try to differentiate ourselves is to focus on a high level of service. Instead of offering a one-size-fits-all algorithm, we offer customization service as well," explains Richard Johnson, senior managing director of Miletus. "We're a small, independent firm in the algorithm space competing with a lot of bulge-bracket firms. We really have to stay ahead of the curve and be more innovative than them to be able to stand out."
Such was the case with Miletus' Close algorithm, which was released in June. "We knew the objective of the algorithm -- to try to achieve the day's closing price," says Johnson. "As algorithmic experts, it is up to us to determine the best way to achieve this." While not all of Miletus' algorithms are custom designed, for those that are, there tends to be a high level of collaboration with a client to meet its specific ideas of how the strategy ought to behave, Johnson adds.
Typically, Miletus will present the client with results of back tests and analysis using historical tick-level data, Johnson relates. The algorithm then is released to one or two beta clients, which begin using it on small volumes of live trades. From that point, Miletus will engage in a period of iterative feedback, lasting as long as a month, during which the firm and its client conduct post-trade analysis to ensure that the desired result is being achieved.
Mercer Oliver Wyman's Cohn advocates an even greater degree of dealer-client collaboration. All algorithm dealers would be well served by leveraging the "agile" software development practices exercised in other parts of their businesses, whereby the final product is moved up and down the development chain and deliverables are received piece by piece with constant feedback from the end user, he asserts.
Using an agile development strategy, "You could get a much more customized product for a client or suite of clients much more rapidly," Cohn contends. "The intellectual trick here is: Could we develop something innovative that could be tweaked with parameters by other clients?" He adds that by tweaking, pushing and pulling a strategy based on individual client needs, "Clients Two, Three, Four and Five are benefiting from the input that client Number One gave you."
This Better Be Good
Still, being a first mover or providing custom algorithms is meaningless if the strategies themselves don't perform.
"The client, from my perspective, is just interested in results," says Miletus Trading SVP David Fellah, who reiterates that for Miletus' customizable strategies, time to market is not the most critical element for success. "[Clients] are only interested in performance. They demand good performance and they demand speed of execution, so the manner in which we test [an algorithm], or the manner in which we implement it, is rarely of concern to them."
Even for a large firm such as BAS, all the quants, developers and marketing dollars in the world won't sell an algorithm that loses money. "The first-mover advantage can be important," says the firm's Harts. But, "If you're offering unique products and new and innovative concepts to the market, then it's not so much a question of if you're there first; it's a question of how good the algorithm is."