Latency Becomes Costly Race of Attrition

As the economy continues to bite into trading firms' bottom lines and data budgets, market data and technology professionals are reevaluating their spend on low-latency data infrastructures, and weighing the increasing costs against diminishing returns as it becomes harder and more expensive to gain advantage through speed alone.

"Low latency is a moving target and a relative term -- you have to be faster than your competitors, and not much faster than that," so as not to be spending prohibitively more than rivals, nullifying a firm's advantage, said Gil Tene, chief technology officer at Azul Systems.

However, "You need to be willing to pay for it," said Bill Ruvo, head of Elektron Hosting and managed services at Thomson Reuters. "We see a top tier, then people who are happy with 400 milliseconds of latency because they know their competition also has 400 milliseconds of latency."

And with the costs involved rising, "People are stepping back now and saying that going from one second to milliseconds wasn't hard... but going to a microsecond or 100 microseconds is harder," said Peter Nabicht, chief technology officer at Allston Trading. "Low latency is about having the right tool for the right job -- it depends on the trade you're running... so don't over-invest."

But latency will remain an important factor for traders, even if their firms aren't competing over nanoseconds: "For us to put orders into the market... we need to be able to execute hedges. And with higher latency, we have higher risk that we will not be able to execute various legs of a trade to manage our risk," said Robert Creamer, chief executive of Geneva Trading.

"People are stepping back now and saying that going from one second to milliseconds wasn't hard... but going to a microsecond or 100 microseconds is harder," - Peter Nabicht, chief technology officer, Allston Trading

Ugur Arslan, founder of Aien Capital, agreed, saying that "Most of the shops out there are trying to go after the same opportunities, so they have to be faster to respond... to get orders in first, and for defense, because when you need to get out of the markets, you need to be able to cancel orders."

Creamer suggested that as the latency race slows in the most liquid US markets, high-frequency trading will expand into new marketplaces and geographies, and that global trading will become more integrated. In fact, deploying a strategy that has run its course on its original market into new geographies and asset classes could reinvigorate that strategy, said Ruvo.

However, Arslan warned that the biggest expense associated with low-latency infrastructures comes not from chasing zero latency, but rather from attempting to expand low-latency strategies into new markets. "I don't think building a low-latency infrastructure is expensive -- you can build one with a few million dollars over nine months. But taking that infrastructure global is where the cost is," he said, citing the cost of datacenters in far-flung regions and the cost of opening offices. "A distributed infrastructure is harder to achieve than a low-latency infrastructure, and the hardest part is finding the right person who knows what has to be done and how it can be done."

Other factors that can impact cost include whether firms invest in specialized technologies or leverage commoditized components to build their low-latency infrastructures, Tene said. "We work in Java, which isn't traditionally thought of as being low latency, but is easy to work with in terms of development spend, bringing something to market... and the cost of supporting that over five years," he said. "So we are seeing people using commodity technology -- because going beyond that has an extremely high cost curve."