As Advanced Trading first attempted to gauge the extent to which algorithms have been deployed in the fixed-income market in the November 2005 issue, it became apparent that it would be some time before these strategies became as prevalent in the market as they are in equities. While many of the structural issues discussed in that first article — such as the format of the dealer-to-customer trading venues, which use a request-for-quote (RFQ) system — still exist, some of the major interdealer venues now are seeing high levels of automated trading, and there is automated arbitrage between them.
There now are algorithms that can be used on RFQ venues, and perhaps more important, algorithms are being written on the buy side to support decision-making and assist in transaction processing, lowering the effective cost of transactions. And while the entry of a new participant, NYSE Bonds, has yet to cause a major ripple, its unconventional firm-quote platform for the most-active corporate bonds has the potential to change the paradigm for this dealer-driven market.
It is clear that the fixed-income market will never look like the equities market — which has few parameters and security types, and where the main object of algorithms is to capitalize on price disparities between markets offering the same security. In contrast, the fixed-income world is comprised of several markets that are structured quite differently from one another and each of which has thousands of securities and many specialized order types.
However, if one expands the definition of "algorithm" beyond the typical equities function of splitting an order and executing against a benchmark to include any automated routine that processes incoming market data and provokes trading activity, one could conclude that algorithms indeed are prevalent in the fixed-income world.
"Algorithmic trading has two pieces: one, decision support; and two, execution," explains Jon Dean, head of global connectivity at MarketAxess. Algorithms have been used in decision support — for example, tracking price correlation between bonds and futures contracts derived from those bonds, and creating hedging strategies based on the information — for some time. And now that price data is improving and electronic trading is becoming more prevalent, according to Dean, bond algorithms are moving closer to combining decision support and execution. Pricing data has become much easier to obtain since the National Association of Securities Dealers (NASD) began offering the Trade Reporting and Compliance Engine (TRACE) in 2002, with adoption increasing steadily in 2005 and 2006, Dean says, making corporate bonds a much less murky category and paving the way for further automation.
If You Want It, Build It
At present, most of the buy-side firms using fixed-income algorithms extensively build and deploy them in-house. Because of the amount of IT effort this requires, some firms, including Bank of New York (BNY) Asset Management, are making algorithmic development part of a firmwide integration strategy.
BNY recently completed an 18-month IT overhaul that produced a common trading and data information backbone called the Transaction Processing Layer (TPL). The goal of the initiative was to integrate trading and analytical platforms in all asset classes so that traders could see a representation of the market at any given point, says Eric Karpman, a VP at the firm and head of its FIX fixed-income technical committee.
The strategy came about because the bank was concerned with updating its asset allocation strategies across multiple asset classes, part of a growing trend toward sector-based trading, according to Karpman. The availability of TRACE data convinced bank executives that this would be a sound investment. "The proliferation of data gave us a reason to go forward," Karpman says. "Algorithmic trading was just a natural exploitation of the resources that were made available by the TPL."
The FIX-based TPL was built on Informatica's data-integration software. Drawing historical and real-time price and portfolio data off this backbone, BNY Asset Management's algorithms facilitate cross-asset trading across all desks, including the personal, global and institutional fixed-income desks; short-term money markets; institutional equities; and index funds, Karpman says. A team of 11 technologists built the system, but frequent input from other bank divisions, including securities master data, market data and quantitative analysts, also was required in order to obtain timely data and satisfy the business needs for each desk, he relates.
The first tier of assets to be automated consisted of foreign exchange, credit derivatives, government bonds and to-be-announced mortgage-backed-securities (TBA-MBS), Karpman notes. The second tier consisted of corporate bonds, municipal bonds and other structured products, he adds.
About 25 percent to 30 percent of BNY Asset Management's fixed-income trades in liquid areas, such as U.S. Treasury bonds and credit default swaps (CDSs), are conducted algorithmically, Karpman says. "We use the TPL as the glue — there is indicative data, market data, price data, all in one central place," he relates. "We were able to easily create plug-ins for the trading systems to send the trades electronically through our custom algorithms."
Karpman admits that the system is probably more cutting-edge than what reigns at most institutions, which tend to be dependent upon vendors for technology solutions. However, he believes most large institutions working in multiple asset classes are close behind BNY on the way to an algorithmic apotheosis. "All the large buy-side firms have some kind of quantitative framework to analyze liquidity and find the best strategy for execution," Karpman asserts. "That is the first step in building an algorithm."
Buy-Side Demand on Rise
The increasing availability of price data and the shift of the entire financial services industry toward sector-based, cross-asset trading means that the production of elaborate strategies to capitalize on miniscule price adjustments across multiple asset types has probably only just begun, industry observers suggest.
"Corporates have been very interesting," notes Brad Bailey, senior analyst at Aite Group. "There has been more transparency in that market, and the credit default swap has risen as a means of giving greater transparency in pricing." Credit derivatives, typically used as a hedge against corporate bonds, have grown at about a 200 percent annual rate for the past few years, according to Bailey.
In the interdealer world, Icap and eSpeed have indicated that more than 20 percent of their order flow is from automated strategies, Bailey adds. Increasingly, this traffic is from hedge funds, such as those operated by Citadel Investment Group, as much as traditional dealers, he says. Algorithms are being used in the market both to arbitrage one platform against the other and to support the decision-making process, Bailey contends. Icap, eSpeed and Citadel officials did not return calls seeking comment.
In the dealer-to-customer area, algorithms are being deployed on the dealer side in order to generate prices, and to modulate those prices based on the class of customer requesting them and the up-to-the-second price data being fed in, says John Bates, founder and VP of Progress Apama Software, which creates risk management, event processing and trading algorithms for financial firms. "The more-recent engines are skewing the price of the bond in real time, based on data changing in real time, and may be changing the spread as corresponds to the tier of customer," Bates comments. "When a request comes in, you spawn a millisecond-length calculation that uses your analytic libraries built up over the years." In other words, dealer firms now can offer not only up-to-the-millisecond pricing based on real-time data feeds and historical information, they also can offer more-loyal customers a better price, or selectively offer improved pricing to less-frequent customers as an incentive to trade more frequently.
Algorithms also come into play when firms take a position in both futures and bonds at the same time, Bates adds. They can be programmed, he explains, to rehedge when set thresholds are crossed.
On the dealer-to-customer sites, such as MarketAxess, which primarily deals in corporate bonds, a few customers have created algorithms that attempt to exploit the latency in the RFQ model, MarketAxess' Dean says. "You can perform intraday arbitrage between RFQ and an order-driven system," he relates. "It can be solved programmatically, but it is not as optimum as in an equities scenario." Traders still must manually press the button to ensure the order has been executed because the information the algorithm was acting on was only an indicative price, Dean explains.
Dean says he believes that nearly all of the 100-plus buy-side firms that write to MarketAxess' application program interface (API) are using some type of automated strategy to inform their trading. But, he predicts, it will be six to 12 months before execution algorithms and cross-platform strategy trading really takes off among the mainstream institutional customers. And it may be even longer before traditional vendors of buy-side order management systems (OMSs) offer algorithmic fixed-income capability, Dean notes.
This frustrates potential customers, such as Travis Bagley, head of fixed-income transitions at Russell Investment Group in Tacoma, Wash. Bagley says his main goal in trading on behalf of his fund customers is cost-minimization rather than rapid profits.
"We are ready and poised to include some kind of algorithmic trading into our process as soon as they become available from vendors," Bagley says. "The algorithms that are out there and working today are created by proprietary users, such as hedge funds and prop trading desks doing arbitrage and alpha-generation strategies. What we'd like to see is one of the trading software vendors create an algorithm for cost minimization as we trade across multiple venues."
It seems that buy side-focused vendors, most of which grew up in the equities marketplace, may still be overwhelmed by the flurry of algorithms that brokers continue to develop for equities.
Sell Side Priorities
For the sell side, it makes sense to allocate technology and resources for the business lines that are most likely to pay off in the shortest amount of time. That means that fixed-income algorithm development ranks behind foreign exchange, options and futures at Credit Suisse's Advanced Execution Services (AES), according to Guy Cirillo, AES global sales channel manager. "We would develop fixed income further down the road as that pent-up demand matures," says Cirillo. "Once these markets are ready for algorithms, we will develop them."
There also must be a global market for the technology in order to fully commit to it, Cirillo notes. "With everything we do, we want to see it applied to not only the North American market, but also Europe and Asia," he says. "If there is something that is only in demand in one region of the world, we are more hesitant to develop that."
Another factor preventing widespread deployment of fixed-income algorithms is the variety of FIX flavors in the marketplace, according to Gary Maier, CIO at Five Mile Capital. Version 4.4, which has the greatest support for fixed income, has yet to be adopted by many brokers, and FIX 5.0 already is on the horizon, notes Maier. "They are mainly doing 4.2," he says.
In the structured product arena, FpML [Financial products Markup Language] is probably better than FIX. "Algorithmic trading will become more pervasive, but it is hard to anticipate when that happens," Maier adds.
FOR MORE ON ALGORITHMIC TRADING in the fixed-income space, view Wall Street & Technology's Editorial Perspectives TechWebCast at advancedtrading.com/events/ondemand.