EXCLUSIVE: ‘Fast bucks’ – James Beeken and Greg Robinson, Cisco; and Kevin Covington, ChangeAlley in ‘The Fintech Magazine’

As Amazon Web Services floats the concept of a stock exchange in the Cloud, we brought three of the early-enablers of high- frequency trading (HFT) together to reflect on the low-latency journey and whether ‘smart’ now has the edge over ‘fast’

In the film The Hummingbird Project, high-frequency traders in the mid-West go to extraordinary and dramatic lengths to knock a millisecond off the latency in data transmission caused by the 1,172.2 miles sitting between them and the New York Stock Exchange.

Back in 2011, when the film is set, they reckoned a millisecond advantage was worth about $5billion. Things have moved on since 2011. Now, you’re talking picoseconds – that’s one trillionth of a second. But speed still equals money… or does it?

Before the introduction of electronic trading silenced trading pits at exchanges across the world, men (and they were all men) clutching fistfuls of paper, shouted and gesticulated in a frenzy of buying and selling, leaving many floors looking like a Bet Fred bookies on Derby Day when the closing bell sounded.

One of the most riotous and colourful, thanks to the many different blazers worn to help identify the status of who you were trading with, was the Chicago Mercantile Exchange. Ten thousand people traded on the floor of the Chicago Exchange in 1997, the year electronic trading was introduced. Five years later, only 10 per cent remained.

“I was a token rocket scientist for the broking firm that hired me out of my physics department in the mid-80s,” says Greg Robinson, recalling his apprenticeship ‘waving his arms about’ on the trading floor of Australia’s SEAT exchange.

But, after the introduction of electronic trading, it wasn’t long before he and others like him, with engineering backgrounds, were being recruited to start a race against time that’s transformed the industry.

Now director of the engineering department at Cisco, which provides the physical network architecture to facilitate trade, Robinson previously founded a high frequency trading (HFT) firm that moved into building its own hardware ‘in order to start getting the market data off the wire more quickly, and the orders onto the wire more quickly’.

“After that, we decided to sell shovels, rather than stay as miners, and so the trading side diminished and the hardware side grew out of that,” he says. The firm, Exablaze, was acquired by Cisco in January 2020.

Among the staff that moved over with Robinson was James Beeken, who started his career building 1,000-plus-position trading floors, with multiple screens, a complex communication panel and thousands of connections to counterparties all over the world.

“They were all highly-stressed individuals, highly-opinionated individuals, and there were just massive sets of infrastructure,” says Beeken. “The trading floors  consisted of all these chaps in coloured jackets, running around, shouting and waving. The step change happened in ’86, when the London Stock Exchange went private.”

While that was undoubtedly a Big Bang moment that signalled the widespread adoption of electronic trading, he believes it was the MiFID I (Markets in Financial Instruments Directive) regulations in Europe that ‘really put competition on the table’, and fired the HFT starting gun.

“With competition came the ethos of ‘I’ve got to be first’,” says Beeken. “The first to hit a price, to get a trade, that became the all-important thing. And, from there, we tried to make networks faster, trying to get colocation principles for datacentres up and running, trying to get datacentres closer to stock exchanges, looking at everywhere in that trade execution cycle where we could start pulling out time.

“Back in the beginning, we could pull out huge amounts of time, for relatively low infrastructure cost. But, as that journey goes on, we’re looking at increasingly smaller amounts of latency, at ever-increasing cost to achieve it.”

Kevin Covington is familiar with that particular rabbit hole. A pioneer in developing proximity, now called colocation, or colo, services, he helped to come up with the concept of data centres in which equipment, rack space, cooling, security and bandwidth were all made available for rent, together with the connectivity to telecommunications and network service providers. It was an essential part of the ultra-low latency trading story.

“We figured out that you could go around and buy bits of rubbish datacentres that just happened to be located near to the exchanges’ data rooms,” says Covington, CEO of industry consultants Change Alley, who was working for extranet provider Radianz at the time.

“We called it ‘proximity hosting’; it’s now called ‘colo’ and it’s a big industry, but I was buying racks in what were effectively cupboards in buildings, and hosting trading algorithms for people. That fired off this whole madhouse.”

In the US, Covington’s team built the first fibre network to connect this ‘hodgepodge’ of racks to each of the exchanges. As alternative trading networks, like BATS, which provided a gateway to NASDAQ,  started to emerge, there was an even greater demand to colocate near established exchanges, as it was realised that the albeit infinitesimal speed advantage provided the chance to grab a piece of the market.

“That caused an increased feeding frenzy on these proximity sites,” says Covington. “The trouble with that was that we were running out of power and footprint, and people were arbitraging on footprint by buying all the racks, so that somebody else could get in there. That was really the first public realisation that latency awas important. From that point, in 2004, onwards, it’s become a mantra and everybody’s focussed on improving it.”

As Robinson explains: “When we first started, hundreds of milliseconds were fast, because you were comparing yourself with human reaction times and we thought that getting down to anything in microseconds was going to be super-fast.  But we all have customers, these days, that prune their fibre cables to a minimum length, to save that extra nanosecond.

“How much a speed advantage is worth is determined by the venue you’re trading in. If you’re trading into a venue with a very high jitter [congestion] in its  network, then you may be fastest, but you may get shuffled to number five in the queue as the orders find their way into the matching engine. So, beyond a certain point, you’re just buying a ticket in the raffle rather than winning the prize.

“I know, from personal experience, when we were trading that just finding a few extra microseconds multiplied our revenues by a factor of five overnight.  But, of course, someone else comes along, and you go from peacock to feather duster very quickly.

“Everybody’s on their toes all the time, and that’s part of the reason why this industry has moved so fast.”

As a result, the people in it have markedly changed: chancers with a nose for a deal have been replaced by those with doctorates who can literally engineer one better.

“Some of the smartest people I’ve met  find their way into the HFT world,” says Robinson. “They live and breathe how they can find an edge and they find their edges in the most remarkable places.”

HFT really took off after the 2008 crash as exchanges, which desperately needed liquidity in the market, were prepared to pay a supplement or fee for HFT, which is characterised by high volumes, albeit those volumes are only in the market briefly. HFT  became a feature of the industry, but its impact led to concern that it’s made the markets less fair and less accessible.

A report published in 2020 by the UK’s Financial Conduct Authority concluded that the practice causes the overall volume of trading on global stock markets to decrease by as much as US$5billion a year, its conclusion based on the fact that latency arbitrage, allows for better prices to be snatched up by HF traders before regular investors.

The arbitrage practice also has the effect of reducing the incentive for those on the other side of a trade to offer these better prices, which also costs retail participants.

Despite the regulator’s misgivings about the negative impact of HFT, it is not going away anytime soon. Indeed, tech giants like Cisco are continuing to develop products such as smart network interface cards – ultra-low latency and high-resolution timestamping adapters – which can deliver sub-600 nanoseconds wire-to-application-to-wire performance out of the box.

“That, in itself, is a fairly eye-watering performance,” says Beeken. “And then we give our clients a set of tools we call ExaSOC so they can fine tune these cards to their networks and improve the performance characteristics even further.”

Field-upgradeable SmartNICs promise to provide up to 10x latency performance.

“Our whole ethos is to drive out those nanoseconds in the platforms we build,” says Beeken.

Work is also going on in the sphere of FPGA (Field Programmable Gate Array) hardware, which allows customers to write their own complex codes to develop trading solutions – again emphasising the role of tradetech engineers.

But that’s a situation only enjoyed by three or four companies which are tuned to the lowest possible latency, according to Robinson. And that has resulted in an ironic twist in the tale.

Having spent the past 30 years getting data to transmit faster and faster, it’s not about the speed of trade on the day anymore, but rather how fast you can forecast it.

“People are now starting to move away from the nanosecond-based trades, into maybe microsecond forecasting and having models which are dynamic, where you can be a bit smarter, even if you’re not faster,” says Robinson. “Nobody wants to be slow on purpose, but if you can’t just get those super-fast trades there’s still plenty of room for sophisticated, high-frequency trading. There’s a sweet spot for everybody.”

With larger firms of high-frequency traders now experimenting with machine learning and updating their models of the world accordingly, intelligence as
much as speed equals money – creating opportunities for ‘another wave of startups that are a little bit smarter’, says Robinson.

The Hummingbird Project ends with… well, we won’t spoil it for you. But where does this tradetech story go next? Will exchanges evolve into being Cloud-based institutions, for example, to reduce the huge costs of maintaining their network infrastructures?

Amazon Web Services already works with capital markets, including NASDAQ and the Stock Exchange of Thailand. It’s recently undertaken a ‘proof of concept’ with The Singapore Exchange and London-based Aquis Exchange to show that trading shares in the Cloud can be fast and reliable.

Covington foresees that there could be a conflict in ensuring market regulation and compliance are maintained with that approach. Beeken, though, is less sceptical: “We’ve got artificial intelligence, we’ve got machine learning; we’ve already got data from the exchanges in the Cloud, so we’re already on that journey towards an entire exchange operating in the Cloud.”

We’ll perhaps leave that plot twist for a sequel interview.

This article was published in The Fintech Magazine #19, Page 72-73