In the wake of the recent naked access ban and amid skyrocketing data volumes, Wall Street banks are revamping their electronic trading platforms to ensure that they can perform pre-trade risk checks faster than ever before. The recent ban on naked access prohibits broker-dealers from providing customers with "unfiltered" access to an exchange or alternative trading system. In order to help prevent erroneous orders and enforce preset credit or capital thresholds, the new rule requires broker-dealers to put in place risk management controls and supervisory procedures before providing customers with access to markets. Meanwhile, many firms, particularly high-frequency shops, are concerned that the new requirements will slow their trading strategies.
As a result, a variety of technology solutions are cropping up to help firms obtain low-latency market connectivity that complies with the naked access ban. Earlier this year, for example, Lime Brokerage began providing pre-trade risk controls via the BT Radianz cloud platform. The brokerage says the ultralow-latency direct market access solution is cost-efficient while providing all inline pre-trade checksin less than 10 microseconds. Lime adds that round-trip communication with the platform occurs in less than 10 milliseconds, depending on where the customer's trading infrastructure is located.
Other top Wall Street players also are moving into the low-latency DMA space. This summer, Bank of America Merrill Lynch launched BofAML Express, an ultralow-latency market access platform for U.S. equities that reportedly delivers sub-10-microseconds of wire-to-wire latency and provides embedded risk controls. Initially developed by Thesys Technologies, the platform currently connects to all major U.S. equity exchanges. Morgan Stanley reports that it has deployed software (it has declined to provide specifics) to shave latency from its DMA platform, Speedway 3.0, which is live with the NYSE, NYSE Arca, Nasdaq, BATS and the two Direct Edge exchange platforms. Even Nasdaq OMX has gotten in on the action, acquiring FTEN, a DMA technology provider that delivers pre- and post-trade risk controls to clients of the major prime brokers, at the end of 2010.
Turbocharged Data Processing
As trading firms seek to reduce latency further, a growing number are leveraging hardware acceleration solutions, such as field-programmable gate arrays (FPGA) -- integrated circuits designed to be configured by the customer after manufacturing -- to speed data processing. "There is definitely a trend toward eliminating the latency between the [network interface] and the memory where the message will be processed," says Peter Lankford, director of the Securities Technology Analysis Center (STAC). "FPGA is one of [the solutions]," he adds, noting that a lot of Wall Street firms also are evaluating their network API strategies to reduce latency. (STAC will release research on the various APIs for trading applications in early October.)
But while FPGA technology has been used in the telecommunications field since the late 1990s, many technologists are still taken aback by what they see as its inflexibility and programming challenges. One financial services executive noted that because code is embedded in the hardware, FPGAs have a limited capacity for upgrades.
In addition, an FPGA takes 10 to 20 times longer to program than a standard central processing unit, acknowledges STAC's Lankford, who reports that "Some firms say they have tried FPGA and found that CPU worked faster for them." "CPUs have much faster clock cycles than FPGAs, but FPGAs give you more opportunity to process things in parallel," Lankford adds, explaining the relative speed advantages of each technology. "If your workload doesn't lend itself to parallel processing, it might benefit from CPU" rather than FPGA technology.
But Jose Marques, global head of equity electronic trading at Deutsche Bank, which is among the firms that have embraced FPGA technology, counters that the technology's inflexibility is often misconstrued. "When you think about hardware solutions, you think about creating custom silicon chips fabricated by people wearing space suits," he says. "That would indeed be inflexible, expensive and would require an entire engineering cycle spanning months. The many small but frequent changes required in the real world would clearly be problematic.
"The great thing about FPGA technology is that it is actually very flexible and highly programmable but does require a bit more specialized knowledge," Marques explains. "The real challenge is in finding experienced FPGA talent. But that is no different from finding, say, a specialized GUI programmer."
Deutsche Bank, whose "ultra FPGA" platform conducts pre-trade compliance and risk management checks while minimizing order latency delays, started revamping its electronic trading platform 18 months ago, according to Marques. Latency monitoring service Correlix RaceTeam recently measured the new platform's pre-trade risk management gateway latency at 1.35 microseconds for messages sent to Nasdaq and at 1.75 microseconds for Financial Information eXchange (FIX) messages, Marques reports.
Meanwhile, as data volumes continue to grow, firms also have been boosting their infrastructure for tick databases. "The number of quotes and trades you need to store on a daily basis has been skyrocketing, so that means more data to go through when you need to do queries, and firms are looking for ways to speed that up and support more queries and users from less infrastructure," STAC's Lankford says. "Firms will take in all real-time data and store it so they can go back later and develop algos on what happened in the past."
According to Lankford, firms are looking at an array of scalable, low-latency storage solutions, including flash-based or D-RAM-based storage. "It is primarily to improve response times per query as the amount of data grows," he says.
Keeping up with the massive amounts of data, suggests Deutsche Bank's Marques, is as important as low-latency executions. "You can have the best connectivity and fastest infrastructure, but if you have stale market data, that's not going to help you," he says, noting that Deutsche has been investing in low-latency colocated market data systems. "We are also [colocating] our SuperX ATS matching engine, smart order routing and the algo platform itself," he adds.
"FPGA market access is one small corner of the electronic business. We provide state-of-the-art algos to our buy-side clients that can compete on a level playing field with the best high-frequency traders out there. We have invested aggressively in colocation and have just opened a new datacenter in Equinix's NY4," Marques continues. "Around low latency, it's about giving our clients access to the same state-of-the-art tools as those used by the best high-frequency traders. There is no reason why institutional traders can't have the same high-performance trading goodies."