Don't Bother Joining the Low-Latency Arms Race, Say Low-Latency Traders

"The opportunities in US equities for straight speed low-latency arbitrage are pretty much gone unless you are a major player," said Chris Bartlett, a partner at high-frequency shop Nobilis Capital. "The days of three people who know technology and market data co-locating their computers to take advantage of what's already been done, they're just not there anymore."

The change is reflected in the profits of high-frequency trading (HFT) shops across the country, he said. There are simply too many competing players with the same high-speed technology. Many such shops have turned away from pure speed plays to more quantitative, analytics-based strategies.

Low-latency offerings still have franchise value to large banks, said Deutsche Bank's chief equities architect Jean-Pascal Chauvet, because it adds to the breadth of their portfolio. But he said that, beyond reputational, the profit and loss value is completely diminished from where it was a few years ago. "Back then, it was about shaving milliseconds and microseconds to be first. Now it's really about how not to screw up,” he said.

Screwing up involves allowing small technology miscues to have massive impact because of how many trades a machine can execute before somebody notices and pulls the plug. Improper risk management and controls can lead to existential screw-ups, as in the case of Knight Capital’s near-implosion. Ironically, Chauvet said he wonders if banks, which have skill enough to write field-programmable gate array (FPGA) code, have the right skills to put proper risk controls frameworks in place. The bottom line is that the risks associated with HFT haven't diminished, but the upside has.

Vladimir Danishevsky, Barclays Capital's global head of credit algorithmic trading technology, did point out that these dour descriptions apply only to equities in the US and, to a lesser extent, Europe. Other asset classes and geographic markets are still making the transition to electronic trading and are therefore not as technically developed or sophisticated. They still present opportunities for latency arbitrage.

The opportunities in US equities for straight speed low-latency arbitrage are pretty much gone unless you are a major player. -Chris Bartlett, partner, Nobilis Capital

What Is HFT?
But while the panelists saw eye to eye on the profitability of HFT, they couldn't agree on the definition of what speed actually constitutes low latency. Anything under 100 microseconds qualifies, said Chauvet. But Bartlett said 100 micros got passed three years ago. That number was halved two years ago. Today, even a small shop without cutting-edge technology is operating around 20 to 30 micros. Will Mechem, managing director at HFT shop Pan Alpha Trading, said his machines are 10 micros tick to trade.

So if low-latency execution is flailing as a means of making money, what is the next frontier for institutions with expertise in speed-based machinery? Danishevsky pointed to machine-learning software and neural networking to predict price. Chauvet said broker-dealers are starting to analyze the trading patterns of their clients to tweak and improve their algorithmic machinery. "If we can harness the huge amount of information we have in investment banks, the first ones that can do that will have a massive, massive edge, in terms of risk management and just knowing your clients," he said.

Subtly referencing Michael Lewis's book "Flash Boys," which has brought public scrutiny to the HFT industry, Bartlett ran with Chauvet's theme and imagined a New York Times headline in the year 2018: "XYZ Hedge Fund Made a Lot of Money Using Low Latency Analytics: Are They Fair?"

Perception aside, he said the approach to high-speed trading has been fairly simplistic up to now, and that analytics will combine with speed to add some nuance. "It’s big data, and amalgamating all these data sources, analytics, and relationships, and doing it real time so you can process it fast enough to make near-term market predictions, which are continuously executing throughout the day," he said. "Take something simple like a good old-fashioned volume-weighted average price (VWAP), but multiply that 100-fold in complexity, with multiple data sources, real-time, and combine it with machine learning and human brainpower, so you're getting signals throughout the day. We'll be seeing a lot more of that. It's almost like information arbitrage."

The Bottom Line

• It's not impossible to make money through latency arbitrage of US equities, according to a panel at the North American Financial Information Summit, but it's gotten extremely difficult. There are still opportunities in other markets.
• High-speed technology like FPGA can still be used, however. It just requires a more nuanced trading strategy that a pure speed trading play. High-speed analytics are a possibility.