Just how fast?
Donald MacKenzie on the increasing speed of high-frequency trading
About half of all buying and selling on many of the world’s crucial financial markets is now automated high-frequency trading. HFT is ultrafast. Whenever I speak to someone who might know and be prepared to tell me, I ask them just how fast that currently is: in other words, what’s the minimum time interval between the arrival of a ‘signal’ – a pattern of market data that feeds into an HFT algorithm – and an HFT system responding to the signal by sending an order to buy or sell, or cancelling an existing order? When I first asked, in 2011, the answer was five microseconds: five millionths of a second. At the time, that seemed extraordinarily fast, but now it seems leisurely. Data released last September by Eurex, Europe’s leading futures exchange, indicated that the speed is now 84 nanoseconds (billionths of a second): sixty times faster than it was in 2011.
In a nanosecond, the fastest possible signal – light in free space – travels just thirty centimetres, or roughly a foot. That’s the fundamental physical limit now shaping what we might call the infrastructure of financial capitalism. HFT’s closeness to that limit creates a situation in which the process is exquisitely sensitive to the technology used to transmit signals and above all to where exactly – very exactly – the technical devices involved are located. The most prominent recent addition to this array of devices is a tower – essentially a large pylon – in the grounds of the Chicago Mercantile Exchange’s computer datacentre, which is in the city’s outer suburbs. At just over 100 metres tall, it looms over the low-rise buildings of the datacentre. It is designed to carry microwave dishes, though they weren’t yet in place when I gazed up at it in October. Those dishes are going to provide the fastest form of communication between the tens of thousands of computer systems packed into the CME’s datacentre and the outside world.
The CME is the world’s most important financial exchange, with no real rival in the US or internationally. It trades ‘futures’, which began as standardised contracts between two participants essentially equivalent to one of them agreeing to buy, and the other to sell, a set quantity of a commodity such as grain on a given future date, at a price agreed today. You might think that the price of a futures contract on grain would track the price of the underlying physical commodity, but it’s actually the other way round: the concentration of buying and selling activity in futures markets means that the price of commodities is often in effect set in those markets, not in the direct buying and selling of the commodities themselves.
That’s also largely the case with the financial futures that the CME began to trade from the 1970s onwards. Their prices often move a fraction of a second before the prices of the underlying shares, bonds or currencies. In addition, price movements in the US – especially in the CME’s share-index futures and government-bond futures – frequently lead those in Europe, Asia and Latin America. That makes the speed at which price data from the CME are received hugely important to the world’s automated trading systems.
Fast transmission of price data used to involve fibre-optic cable, but the strands in such cables are made of materials (usually a specialised form of glass) which slow light down to around two-thirds of its speed in free space. In contrast, microwave and other wireless signals travel through the atmosphere at nearly full light speed.[*] Since 2010, no fewer than 17 competing microwave links have been built to connect the CME’s datacentre in suburban Chicago to the datacentres in northern New Jersey in which shares, US Treasury bonds and currencies are traded. Fierce competition has, however, winnowed out the slower links, and now only three firms remain in the race. (A fourth competitor may be about to emerge: it seems a new network is being built by Scientel, a specialist telecommunications firm with origins in the nuclear power sector. Scientel is probably doing this for an undisclosed HFT client, but I haven’t been able to discover whether that’s actually the case or who the client may be.)
Chicago’s outer suburbs are not a tourist destination. There are logistics depots, light industrial plants and scrubby vegetation; the landscape is flat and often dominated by power lines and freeways. If you’re involved in the microwave speed race, however, you will have had to get to know the patch of landscape immediately surrounding the CME datacentre intimately. The CME didn’t allow the three competing firms to install microwave dishes directly on the roof of the datacentre, so instead they were placed on towers several hundred metres away. Until quite recently, this meant that the crucial signals from the CME – and, though they tend not to be quite as important, the market data that flow from other exchanges to the CME – were slowed, perhaps by as much as a microsecond, by having to get from the datacentre to one of these towers (or vice versa) via a fibre-optic cable.
There was a considerable stir, therefore, in May 2017 when the financial news service Bloomberg published a photograph of a diesel generator in a field beside the road that runs along the north side of the CME datacentre. Connected to the generator was a short pole carrying two small antennae. That the two antennae were facing slightly upwards gave the game away: it was an ‘uplink’, which could transmit microwave signals to, and receive signals from, a nearby microwave tower. Bloomberg revealed that a company affiliated to Jump Trading (a leading HFT firm) had paid $14 million for the field. Having an uplink just across the road from the datacentre means that the length of fibre-optic cable through which a signal needs to run is reduced from hundreds to tens of metres, thus saving that crucial microsecond. The other two competitors in the race for speed have now constructed their own uplinks close to the datacentre: they had no alternative, because that microsecond is the difference between success and failure.
I am told that Jump intended to recoup much of the field’s purchase price by reselling all but the small portion of it needed for the uplink. I hope for their sake they were quick about it, because the new tower – which, as I’ve said, is inside the grounds of the datacentre and so at least thirty metres closer to it than any of the uplinks – will have considerably reduced the value of the surrounding land. Each of the firms that compete to have the fastest Chicago-New Jersey link will have no real choice but to install microwave dishes on the tower. I’m told they have been promised that the cables from their dish into the datacentre will be of the same length wherever their dish is placed on the tower. I haven’t been able to discover the prices they’re going to be charged – the tower belongs not to the CME but to the datacentre owner – but it’s clear that they’re going to have to pay them.
Europe’s equivalent of the Chicago-New Jersey route is the succession of microwave links between Frankfurt – where Eurex is based, along with most of the trading of German shares – and Greater London, where most of the rest of Europe’s share trading and nearly all of its foreign-exchange trading takes place. (The electronic trading of Eurozone sovereign bonds that currently takes place in London is going to move to a datacentre in Milan, at least if Brexit goes ahead.) There is a similar competition in speed on the London-Frankfurt route as on the one from Chicago to New Jersey, and the same three firms are in competition in Europe. (The race has been documented in remarkable detail by the researcher Alexandre Laumonier in a blog that has been hugely popular in the world of high-frequency trading.[†]) Europe’s geography, though, has exposed the chief limitation of microwaves for use in trading: it is a line-of-sight technology. There has to be an uninterrupted straight line between each successive dish in a microwave link and the dishes before and after it. The curvature of the Earth therefore limits the distance between microwave towers.
That’s a real problem when it comes to the sea crossing from southern England to the Continent. The shortest path (the ‘geodesic’) between London and Frankfurt crosses the east coast of Kent near Richborough, a point at which the crossing is too long to be achievable with standard microwave towers. (Lake Michigan is a challenge, but less so, for the Chicago-New Jersey links.) In 2016, two HFT-owned microwave companies applied for planning permission to build 300-metre masts – three times as tall as the CME tower, as high as the Shard or the Eiffel Tower – on the Kent coast near Richborough to establish a line of sight that would make a geodesic-hugging link to the Continent feasible. Unsurprisingly, in 2017 Dover District Council turned down their applications, so as things stand the three competing microwave links have to deviate from the geodesic and cross the Channel further south, closer to the narrow Dover Straits, at a cost in speed of around ten microseconds.
The English Channel is an obstacle to microwave signals, but the world’s oceans are a currently insuperable barrier. There have been proposals to suspend microwave dishes from balloons, or have them carried by ships, helicopters or drones, but so far none has been adopted. The crucial signals from US datacentres to Europe still travel by fibre-optic cables on the Atlantic seabed. So far as Brexit is concerned, that is perhaps fortunate. Hibernia Atlantic’s cable, which is the fastest, makes landfall near Brean on the Bristol Channel, which means that signals from the CME reach London a thousandth of a second or so before they get to Frankfurt and the other financial centres on the Continent. That just might give London a continuing slender advantage as a centre of trading.
A competitor to submarine cables is, however, starting to emerge. Over the last couple of years, one of the contributors to Laumonier’s blog has been an engineer, Bob Van Valzah, who is a keen cyclist and lives in the Chicago suburbs. On his outings, he began to notice the appearance of new shortwave radio antennae. Like microwave, shortwave is an old technology: it’s what Radio Moscow used to use, and the Voice of America and BBC World Service still rely on it in some parts of the world. The point of shortwave is that at least a small portion of a radio signal in that frequency band will often bounce back to the Earth’s surface from the ionosphere, the layer high in the atmosphere in which electrons can be stripped from molecules in the atmosphere by the sun’s rays or cosmic radiation. Shortwave signals are therefore not restricted to a line of sight: they can travel over the horizon, sometimes for thousands of kilometres – potentially across the Atlantic, for example.
The new antennae spotted by Van Valzah are almost certainly being used to find out if shortwave might work in transoceanic HFT. Shortwave’s bandwidth is very limited, but the critical signals for trading are often extremely simple: you don’t need many binary digits to convey the fact that the price of a futures contract on the CME has ticked up or ticked down. The problem is that shortwave is a fickle technology, affected by conditions in the upper atmosphere, by solar flares, and even by whether it’s day or night. If in your youth you used to listen to shortwave radio stations in distant countries, you will have experienced the effects: a station you could hear clearly one night could be inaudible a day later.
With skilled planning and sufficiently powerful transmitters, shortwave signals will probably manage to reach London and Frankfurt from Chicago, much of the time at least. But what about Mumbai, Singapore, Hong Kong, Shanghai or Tokyo? Shortwave is unlikely ever to make direct transmission over those distances feasible. This potentially brings satellites into the picture. The cost of launching them is coming down sharply, and satellites could become a faster alternative to transoceanic cables. When I first heard discussion of this, I was sceptical. A whole constellation of satellites would be required: you can’t use a single geostationary satellite, since it would have to be more than 30,000 km above the surface of the Earth, and that would make a signal transmitted by satellite slower than one travelling by undersea cable.
What changed my mind was talking to two people in the industry whom I know not to be dreamers but serious people who actually build things. One of them led the construction of the earliest Chicago-New Jersey microwave link; the other has built several impressively fast links. Both have been doing research on the geometry of the problem (the low-Earth-orbit satellites required constantly change position with respect to the Earth) and its economics. Both think that a global network of a hundred or more satellites, constantly passing financial signals among themselves, is technologically feasible and potentially within the budget of the world’s HFT firms. It may not happen even so, but initially there were sceptics about microwave transmission and these two people proved them wrong. I wouldn’t be entirely surprised if, before too long, a world-girdling satellite network became part of the infrastructure of finance.
[†] Laumonier’s findings are included in his book 4 (Zones Sensibles, 106 pp., €15, January, 978 2 930601 36 6).