Two microseconds per round-trip. That’s about how fast Deutsche Bank said its newest market access product takes for an incoming order to undergo a pre-trade risk check, hit an exchange and return the acknowledgement.
The product, from Deutsche Bank’s low-latency Autobahn Ultra suite, was designed for high-frequency traders. About 25 clients have shown interest in it, said Ralf Roth, the bank’s global head of product development for equities trading. And while some of those are existing Deutsche Bank clients, Roth estimated, many are new.
"There’s no one out there who’s published anything faster," he said.
The time represents the 1.25-microsecond trip an order takes from the Deutsche Bank network interface card, through a risk check, and to the exchange matching engine. The return trip, minus the risk check, takes a blistering 0.8 microseconds, Roth said.
The new market access product–which Roth referred to as an "appliance"–is awaiting a patent, as Deutsche Bank has filed for one, he said. Deutsche Bank just installed it at the Nasdaq market center. The bank’s first client, a U.S. firm, started using it earlier this week. Deutsche Bank will roll it out to all major U.S. markets over the next two months, Roth added. It will be available in Europe later this year.
The technology is Deutsche Bank’s answer to the Securities and Exchange Commission’s move to ban brokerages from providing unfiltered–or naked–sponsored access to the exchanges. The SEC requires brokers to perform pre-trade risk checks for their customers who use their market participant identifiers to access the markets.
In fact, many of the biggest brokers have long since been ready. Deutsche Bank joined other large banks that have entered the low-latency business in the wake of the SEC’s call to ban naked sponsored access. Over the past year, or so, banks such as UBS, Credit Suisse, Barclays Capital and Nomura Securities International have introduced low-latency products.
These firms wanted to be prepared just in case the SEC’s proposal to end naked sponsored access became a rule. That is exactly what happened, so these firms already had systems in place. That included investing in the technology to cash in on the revenues to be made in the burgeoning HFT space.
One high-frequency trader said he thinks the product can attract interest. The speed increase, he said, would be significant to a firm that had been working with slower technology. There is a noticeable difference, say, from making cancel and replaces at speeds of 500 microseconds, he said, compared with making them at less than 10 microseconds.
What’s more, Deutsche’s new product might be fast enough for some prop-trading firms and hedge funds that they won’t need to become broker-dealers. It’s possible, the HFT added, that firms on the fence about their status might remain non-broker-dealers if they could get to market just as fast–or faster–and stay within the market access rules.
Deutsche Bank has arrived at its speeds by taking a different approach–one it sees as the technology of the future for low-latency trading. Unlike most providers that build software solutions, Deutsche Bank delivers its speeds through a piece of hardware, Roth said.
The bank built proprietary risk-checking logic onto a field-programmable gate array (FPGA) card. Deutsche Bank would place the card at a co-located site, in a cabinet at the exchange. The client connects to Deutsche’s card through a cross-connect fiber cable.
"This is so revolutionary that we believe that over the next 18 months, most clients that care about low-latency will move to these types of hardware appliances," Roth said. "We believe that with the proliferation of hardware in the low-latency trading space, it will be hard for anybody to compete without using some kind of hardware appliance."
Hardware can be built to interact with a very specific program or algorithm, Roth said. Software, by comparison, must interact with a computer’s operating system, which is ultimately less efficient. Also, FPGA cards are excellent at parallel processing, he added, and therefore well-suited to working with large amounts of data.
