NEW YORK (Reuters) - Algorithmic trading, in which computers execute orders at lightning speeds, has financial exchanges relocating computer systems to shave milliseconds off execution times, executives said this week at the Reuters Exchanges and Trading Summit.
Transaction decisions are measured in fractions of a second in algorithmic trading in which computers are modeled to execute orders based on criteria like price or timing.
One-fifth of global equity trading last year was algorithmically driven, up from 14 percent in 2003, according to financial consultant Celent.
When speed is a priority, as it is for hedge funds or other momentum investors that use algorithmic trading strategies, the amount of time it takes bits of data to travel from one city to another can make all the difference, executives said.
"Algorithmic engines are now near our data center, and the broker-dealer firm would control the logic and running of those models," said Nasdaq Stock Market Inc. (NDAQ.O) Chief Executive Robert Greifeld. "In a real sense, you are reconstituting the floor, except this is a computer floor."
Proximity to data centers has become so crucial that Greifeld said renting space near his servers is a growth business for Nasdaq, which is the top U.S. electronic stock exchange.
"For some customers it matters, and those are actually the customers that add a lot of liquidity to the system," said Dave Cummings, CEO of electronic trading platform BATS Trading, of millisecond latency. "If you satisfy them, you have liquidity, and that's why the longer-term participants come in the market."
"The standard now is sub-one millisecond," said Philadelphia Stock Exchange CEO Sandy Frucher. "If you get faster than sub-one millisecond you are trading ahead."
About eighteen months ago, the standard for executing a trade was around five milliseconds, he said. A millisecond is one-thousandth of a second.
The focus on milliseconds is so intense that trading platforms like the Philadelphia exchange and Kansas City-based BATS have moved their servers closer to broker-dealer hubs in the New York City region.
Cummings said BATS' matching engines are in Weehawken, New Jersey, across the Hudson river from Manhattan, adding that the same building hosted servers from NYSE Arca, an electronic exchange. "We chose that building strategically. A lot of our customers already had their servers in that building."
When Frucher began revamping the Philadelphia exchange's equity trading platform about 18 months ago, he realized that a three-millisecond delay between Philadelphia and the New York City region -- about 100 miles away -- was noncompetitive
"I had one of the great comeuppances in my life," he said.
Frucher decided to move some of the exchange's equipment closer to New York City, but did not specify where.
Even the New York Stock Exchange NYX.N, the No. 1 U.S. stock exchange whose once-packed trading floor has long been synonymous with Wall Street, is feeling the squeeze.
"The New York Stock Exchange itself is too slow, and we have to be faster," said NYSE Euronext NYX.N CEO John Thain. "We are in the process of doing that."
Thain added that its Arca unit "is as fast as anyone."
Asked about the effect of the algorithmic trading boom, Thain said it did not pose a trading risk in normal environments.
"Most of the time the algorithmic trading actually provides liquidity to the system," he said. But he added that it was unclear how algorithmic trading engines would behave in a dramatic market decline.
"Will they exacerbate the decline?" he asked. "I don't think anyone really knows the answer to that because we haven't really seen it happen."