Interest in high-frequency trading is at an all-time high, but profit-taking from high-frequency trading strategies focused on low latency is getting tougher.
"The window of opportunity to get into high-frequency trading is almost closed," said Mark Casey, president of CFN Services, a network provider. He defined high-frequency trading as strategies whose underpinning is low-latency order placement and execution.
"If you’re competing primarily on latency, it’s very, very, very, very difficult," added Nigel Faulkner, chief technology officer for the equities technology group at Goldman Sachs.
The cost of the technology and infrastucture needed to support high-frequency trading is "tens of millions of dollars" per year, according to Kevin McPartland, a senior analyst at financial services research firm TABB Group. He moderated a panel sponsored by TABB Group and Switch and Data, a data center operator, last Thursday. This article is based on the panel discussion.
Low latency is necessary, McPartland said, to process market data faster than competitors. And high-frequency trading, which encompasses a range of strategies, depends on that data. "It’s like you’re seeing the Wall Street Journal five microseconds into the future," he said.
High-frequency trading firms must be concerned about latency, but that level of concern should depend on "how much profit they intend to make from every millisecond or microsecond," Goldman’s Faulkner said. He noted that firms must understand the "value of a micro or milli" for the particular strategy they’re running.
"The infrastucture isn’t the barrier" for firms interested in high-frequency trading, CFN’s Casey told the audience. The barrier is competition. In his view, competing with the most latency-focused firms is a tough, elite game because, at that level, microseconds count. A microsecond is one-millionth of a second, while a millisecond is one-thousandth of a second.
According to a recent TABB report on financial services data centers, the financial services industry spends $1.8 billion for co-location and private facilities to support fast direct access to market centers. Broker-dealers account for half of that sum, or $900 million. Exchanges represent 23 percent, proprietary trading firms 13 percent, asset managers 10 percent and hedge funds 4 percent. That report was published in March, but the figures remain accurate, McPartland said, based on TABB’s ongoing research on data centers and trading, including for an upcoming report on sellside technology focused on U.S. equity infrastructure.
McPartland noted that bulge-bracket firms will often have four or five primary data centers to support their own equities trading and the trading of their clients, and 10 or more co-lo sites in the U.S. All brokers, he said, use co-lo at some level, with many operating in at least two or three co-lo sites.
McPartland added that housing servers within an exchange’s data site is costlier than placing the servers near the facility, such as across the street. The chief features behind a firm’s choice of a data center are cost (which is important to 57 percent of firms), exchange proximity (48 percent), space in the data center to expand (33 percent) and power reliability (29 percent), according to TABB. Additional concerns are service, security, control and network neutrality.
CFN’s Casey said that "proximity trading" has exploded over the last couple of years. Proximity trading refers to strategies that depend on low latency by installing computer servers near a market center’s matching engine.
One of the changes in the marketplace in recent years that has fueled high-frequency trading was regulation. In 2007, the Securities and Exchange Commission’s Regulation NMS went into effect. Reg NMS lay down a set of rules to modernize the markets, but it also made the landscape more fertile for high-frequency trading firms. Casey noted that execution strategies that used to be implemented just on Nasdaq, for instance, have given way to more inter-market trading strategies.
But regulation wasn’t the only significant change. The TABB study found that the "game-changing technology" that spurred the growth of high-frequency trading was bandwidth availability and the relative low cost of buying bandwidth. "That’s what is letting equities volume be eight-to-nine billion shares per day," McPartland said.
Several panelists pointed out that while speed is vital, not all high-frequency trading depends on extreme low-latency. Nor is all low-latency trading high-frequency, CFN’s Casey said. Still, Goldman’s Faulkner observed that "if it’s high-enough frequency, it must be low latency." He added that "we increasingly see that the benchmark [for high-frequency trading firms] is low latency."
As more firms now get into high-frequency trading, their infrastructure development has taken different paths. George Hessler, executive vice president at Lime Brokerage, said he thinks the balance for many firms is tipping toward renting components of the technology and infrastructure, rather than building them from scratch. He added that as consolidation takes place in this part of the trading-services industry, the hardware and software services are improving dramatically. Lime services many high-frequency trading clients.
Goldman’s Faulkner, however, said that it would be hard for a truly latency-sensitive firm to be satisfied with vendor products. For big banks, he added, servicing these firms has also become more complex because their needs are different from the traditional needs of high-volume clients. "We’re having to change the mix of our application developers," he said.
Firms that are really latency-sensitive must pull out all the stops to account for every microsecond, since that affects their profitability. They must "account for the last 100 microseconds they can’t find," and be able to figure out if the latency is in the code, switches, applications or elsewhere, Faulkner said.
UBS has a "strong bias" to build rather than rent the various components necessary to support high-frequency-trading firms, according to Josh Schubkegel, executive director for client-facing technology at the big bank. He noted that some clients want to get "close to the metal" and do everything themselves, while others do not.
Schubkegel noted that the focus on serving high-frequency firms has also benefited other clients at some of the big banks. UBS, he said, has leveraged some of the technology platforms developed for high-frequency traders for its direct market access and algorithmic trading business.