New trading rules banning ‘naked access’ that go into effect on July 14 have challenged brokers to revamp their electronic trading platforms so they can perform pre-trade risk checks at blazing speeds.
As a sign of the competitive stakes, Morgan Stanley says it has shaved significant latency from Speedway, its direct market access platform, which the firm says it offers to low latency trading firms.
“We are sub-two microseconds of latency, optimally configured,” says Ari Kaplan, managing director and global head of quantitative client facing business for Morgan Stanley Electronic Trading (MSET). Previously, Speedway’s latency was less than 200 microseconds, “so it’s a very dramatic uptick in performance,” emphasizes Kaplan.
Since January, the system has been doing live trading in U.S. cash equities. Speedway 3.0 is live with five exchanges, including NYSE, ARCA, Nasdaq, BATS and the two Direct Edge Exchange platforms. Next, Speedway is headed for Europe and the Far East, where it will be launched in the next couple of months. While U.S. equities are the focus currently, in other countries equities and swaps may be part of the platform.
SEC Rule 15c3-5 prohibits brokers and dealers from providing customers with “unfiltered” or “naked” access to an exchange or alternative trading system. The way the rule is written brokers are required to perform financial and pre-set credit or capital threshold checks on single and aggregate orders.
Morgan Stanley did not need to create any new checks. All of the core checks have been the same. “It’s the same methodology we’ve always followed but technology has gotten faster,” says Kaplan. “Where anything in the low single digit milliseconds five years ago was cutting edge now you have to be in low single micros to be competitive,” asserts Kaplan.
Not every client needs this speed. The new Speedway is geared to low latency clients, which include quantitative asset managers as well as high frequency trading firms. Quants may have correlated strategies and if someone is trading the same model, they don’t want to be the last to the exchange-matching engine. While low latency clients are core to Morgan Stanley’s client base, the product also serves high frequency traders who want to end the day flat without taking overnight position risk.
But the new rule has set off an arms race among brokers that serve low latency clients and high frequency trading firms. In recent months, other firms have announced similar technology solutions to comply with the rules. Last week, Lime Brokerage said it would provide pre-trade risk controls via the BT Radianz cloud network to other broker dealers. And last December, Nasdaq OMX acquired FTEN, a direct market access technology provider that has provided pre-and post-trade risk controls to clients of the major prime brokers.
However, comparing the latency statistics from one firm to another is difficult and can be confusing. “There are variations in how latency is measured,” cautions Kaplan.
For example, some measures of latency do not reflect the full wire-to-wire latency or the application latency, or the time it takes for getting in and out of the risk checking engine, he suggests. Additionally, latency can vary depending on the exchange the firm is connected to and the protocol it’s using.
Morgan Stanley measures the full wire-to-wire latency, which is the time it takes for the data bits to go from the wire (context – more than one type of on wire material) into the host and then through the risk checks and then through the network layer to get out of the box.
While some of the competitors have worked with FPGA (field programmable gate arrays) devices, in which blocks of logic are programmed into a chip, Morgan Stanley worked with software to get the same performance characteristics. (In March, Deutsche Bank said its Ultra Solution based on FPGAs, was clocking speeds of 1.25 microseconds.)
Kaplan says there are ways to optimize a traditional host to get the same kind of performance with greater flexibility. “Morgan Stanley chose to invest in a technology that would be largely reusable across our entire trading plant.” says Kaplan.
“One of the challenges with FPGA is that the software is embedded in the hardware limiting the capacity for upgrades,” he says. When versions of messaging change at the exchanges, the traditional host will be easier to update, suggests Kaplan.
Meanwhile, a number of technology providers are jumping into the space as well. Kaplan says he recently attended a live demo where there was standing room only for 100 people as the vendor was sending orders and measuring latency. People were asking technical questions about how the speed was achieved.
“A year ago, the same group of people would have said these rates were nearly impossible to achieve. Clearly, there’s an evolution going on here.” says Kaplan.