About Time, page 8
The London Stock Exchange was the perfect example, and one of its automatically corrected clocks has survived in a private clock museum in London. It is an oversized and finely constructed mechanical clock made in the early nineteenth century by the noted clockmakers Thwaites and Reed, so it was already decades old by the time it was fitted with a Standard Time Company synchronizing device in the 1880s. From then on, each hour, an electrical signal was received from the time company which flashed across a similar device fitted to every clock on the circuit, causing it automatically to correct the clock to the right time. It was accurate to within about a second of Greenwich Mean Time, the official UK timescale, according to experiments carried out on surviving equipment.
The Standard Time Company’s clock-synchronizing service was not cheap to subscribe to, and the maintenance of this high-tech network was onerous—it was fragile and vulnerable. Yet hundreds of companies, large and small, went to the trouble to get it, and many were in the financial sector. Accurate time was increasingly valuable, and businesses would seek out the latest horological technology that would help them in their ambition to make money.
By the time the Standard Time Company began synchronizing clocks across the City of London, financial markets were no longer the single locations fixed in time and place that the Amsterdam and other early exchanges had been. They were geographically and (therefore) temporally dispersed. Trading on the markets meant engaging with people, institutions and technologies worldwide.
When the concept of daylight saving (or shifting clock-time forward an hour during summer months) was first put forward in the early years of the twentieth century, both the London Stock Exchange and the Liverpool Cotton Market—Britain’s first futures market—were worried about the proposed scheme. It came down to the temporal nature of markets. The UK markets both closed at 4 p.m., partly out of convention but, in the case of London, to catch the evening postal collection which processed the thousands of letters and memoranda that formed the record of its daily trading.
A vast amount of that daily trade was with the New York Stock Exchange, which closed at 3 p.m. Owing to the five-hour time difference between Britain and New York, that meant there was just one hour each day when both territories were open together: 3 p.m. to 4 p.m., UK time, which was 10 a.m. to 11 a.m. in New York. The daily trade during that single hour was, in the words of one broker during the discussion, “immense.”1 If Britain’s clocks shifted forward by one hour in summer, the window would close completely, and Britain’s brokers feared that trade would drift to another global exchange better suited to fit with New York’s hours. Of course, London could stay open later, but it would miss the post each evening, making it less efficient, so trade would drift away, and neither exchange could open an hour later, as they were tied into other temporal windows around the world at that time of day, so working hours would have to increase, so clerks and traders would drift off to other employments, and so on. Alternatively, New York could have changed its hours but it too was tied into an infrastructure that would start to unravel.
When the Liverpool Cotton Exchange moved into brand-new permanent premises in 1907, brokers were surrounded by a network of synchronized electric clocks in the trading hall and surrounding offices, designed, so the Liverpool press explained, “in order to obviate confusion and conflicts of opinion in the transaction of business.”2 A little over a century later, synchronized networks of clocks have moved on, somewhat. When Benjamin Franklin, in 1748, said that “time is money,” he could not have imagined how his maxim to work hard and be thrifty would take on new meaning centuries later.3 Today’s financial markets, which in total trade hundreds of billions of dollars per day, need clocks that are accurate to 100 millionths of a second.
BROADLY SPEAKING, THERE are three ways to buy or sell financial instruments (like stocks or futures) today. The first is human trades, where a person—a real human—issues an order to buy or sell a certain amount of a certain instrument online or over the phone. It is a human finger on the mouse button or a human voice shouting orders on the phone and, apart from the technology, it is the same way things were done in 1611 at the Amsterdam exchange. You issue your instructions and your order gets in line with everyone else’s. It is nice to think this still happens today. Think of it like heritage trading.
The second way to trade on the financial markets is to use computers running complex sets of instructions, or algorithms. These algorithms, or “algos,” automatically issue buy and sell orders according to pre-programmed rules. If the price of this stock comes down below this level, then buy this much, for instance. If this company reports a loss in its quarterly accounts, then sell this percentage of our holdings. That type of thing. Algo trading, which started taking off in the 1990s, is now big business because it happens faster than humans can react, and, remember, this is an arms race of timing.
But conventional algo trading runs at a snail’s pace compared with the third method of buying and selling on the financial markets. A subset of algo trading techniques known as high-frequency trading, or HFT, entered the scene in the 2000s and now accounts for over 50 percent of stock trading in the USA—a little lower in the UK. As its name suggests, HFT is all about speed, making huge numbers of trades every second. The individual trades are usually very small and the profit on each is also tiny—but because of the sheer volume of them, the profits can build up into a sizable amount. At the end of each trading day, HFT traders aim to hold no assets. They sell everything they buy that day. This is not about building up a stock portfolio or holding a futures contract for weeks or months, it is about the friction of trading itself generating money. The assets being bought and sold are of no interest to the HFT traders. They could be company shares, domestic mortgages or exotic financial derivatives. The point is that the computers spot the chance to make a tiny profit—a chance that might last only the tiniest fraction of a second—and they pounce, at light speed.
Where do clocks come into this? The first way is in the technical synchronization of the computer networks over which the trading takes place, which, as we will see in a later chapter, relies heavily on accurate clocks to work properly. But the second way in which clocks have become important in the era of HFT and algo trading is in many senses the same as in 1611 in Amsterdam. There, the clock was used not just to bring buyers and sellers together but also to time-stamp financial transactions to ensure everything was done officially and above board. Physical markets with clocks allowed oversight by regulators—in that case, the city council. Anyone who broke the rules would be fined. It is the same in today’s financial markets. Clocks enable regulation to happen.
The risk today is that firms with computers trading at almost light speed might be able to “see” the financial markets a tiny split second earlier than their competitors, which would enable them to profit on the information they saw in that tiny window of time. In some circumstances this would be illegal, so what regulators need is to be able to read the sequence of every trade and every data feed, to make sure that nobody is breaking any rules and unfairly jumping the queue. Assuming everyone in the marketplace agrees on what time it is, then the trades and data feeds can be time-stamped, and it is clear what happened when—and what came before or after what. It means that regulators can keep a watchful eye on the conduct of the markets.
It sounds easy but, in practice, there are three problems. The first is agreeing on a timescale. In today’s world many timescales are in use. We have TAI, which comes from atomic clocks, and UT1, which comes from the Earth’s rotation, as well as UTC, which is a hybrid of the two, but there are seventy-five very slightly different versions of that depending on where you are in the world. Or we could choose GPS time, or time from other satellite navigation systems such as GLONASS, BeiDou and Galileo. We have, too, time from radio stations, internet time services, cell-phone providers and broadcasters. Deciding which timescale is “the” time is not a trivial problem to solve.
The second problem with time-stamping in the financial markets is a hardware problem. Once you have accepted which timescale is the one to use, how does every market participant get access to that timescale so that all the time stamps agree with each other? It is easy enough to get the time to a clock on the wall of a stock exchange for those human traders, but what about the thousands of computer servers exchanging high-frequency trades around the world? Each server, each microprocessor, each network switch needs to have access to the same clock, somehow. Again, this is not trivial, to say the least.
The third problem is in some ways the most challenging. Imagine that two financial orders were placed half a second apart. If the time-stamp clock ticked once per second, as many clocks do, it would be quite possible that both trades would hold exactly the same time stamp, and nobody would know (or be able afterward to find out) which came in first, opening the system up to abuse. The third timing problem therefore is the precision of the time-stamping clock—how small its timing intervals are—as well as its accuracy. The more precise and accurate the time stamp, the more likely it is to get the trades in their proper order. But of course more precision and accuracy is harder to achieve and even harder to distribute over the whole network.
It is important to be clear about what is needed here. The European directive on markets in financial instruments which came into force in January 2018, known as MiFID II, selected UTC as its timescale, and demands time stamps precise to one second for human trades, or one-thousandth of a second for normal computer algo trades. But for HFT trades the time stamp must not deviate from true UTC by more than 100 microseconds, or millionths of a second, and it needs to have a precision—the gap between two successive stamps—of no more than one-millionth of a second. Put another way, instead of stamping once per second, MiFID II–compliant clocks stamp a million times each second. And those time stamps need to be found on every chip in every computer server in every trading exchange across the entire European financial market.
Just think about that for a moment. All the computers that are involved with financial trading across the whole of Europe need to show the same time as each other to 100 millionths of a second. There are thousands upon thousands of such computers in a single data center alone. Records need to be kept documenting the time stamps at every moment, in every location throughout the network, for future audit. The requirement is for at least five years; the best time suppliers keep them for seven. There are 221 million seconds in seven years. A million time stamps per second. This is a requirement for the most astonishingly accurate, precise and reliable time, twenty-four hours a day and documented for several years in case the regulators want to check something, and any firm that does not comply risks being slapped with a fine of 10 percent of its global revenue. Not profit—revenue.
THE AMSTERDAM STOCK Exchange clock made in 1611 would have been accurate to better than half an hour a day. It kept time using a horizontal bar or wheel that rotated first one way and then the other, powered by falling weights and regularly corrected by its keeper using a sundial. In the 1650s, a new horological technology entered the scene, when pendulums were first used to control clock timekeeping. They were a radical improvement in accuracy because they oscillated at a frequency that was predictable, unlike the older clocks, and the clock-making industry jumped on the chance to develop and refine them. By the 1920s, the very best mechanical pendulum clocks had reached an accuracy whereby they would gain or lose no more than a second in two or three months compared with true time as measured by the astronomers at national time-finding observatories.
Then, electronics engineers started building quartz clocks. These used pieces of quartz crystal that vibrated like a bell, and the frequency of the vibrations could be measured with electronic equipment, so they could be used as a clock. These were much more accurate than pendulum clocks—good to about a second in thirty years to start with and soon much improved.
But things really started to change after 1955, when the UK’s National Physical Laboratory, or NPL, built the world’s first successful atomic clock, which used fundamental and unchangeable properties of atoms for its time base. No longer did the world’s best clocks take their time from the shape and size of a physical oscillator such as a pendulum or a slice of crystal, which put practical limits on the quality of timekeeping. With atomic clocks, the accuracy and precision that could be measured was almost limitless. And since that pioneering NPL clock first began to keep time in 1955, physicists have provided the world with a procession of exponentially more accurate clocks.
The 1955 atomic clock was accurate to one second in 300 years. By the 1980s, NPL’s atomic clocks were keeping time to an accuracy of one second in 300,000 years. Homo sapiens, the human race, appeared about 300,000 years ago. It is a long time to keep time within a second, but today’s atomic clocks are much more accurate than that. Clocks now keeping time at NPL, known as cesium fountains, are good to one second in 158 million years. But even these timekeepers are sluggish compared with the next generation of atomic clocks the scientists and technicians are now developing. These have an accuracy of plus or minus one second in 30 billion years. That is more than twice the age of the universe. Or, to put it another way, if one of these clocks had somehow been set running at the Big Bang—the explosion that brought the entire universe, even time itself, into existence—it would currently be wrong by less than half a second.
But how do we set our clocks right? NPL does not just keep time. It gives it away: NPL clocks feed into radio time signals and internet time services that allow all of us to know the time they keep. But neither of these is accurate, precise or stable enough for MiFID II, with its demand for time stamps every millionth of a second. The clocks at Teddington are much more accurate than that, but the hardware that transmits the time to the outside world could not cope. Time signals from GPS satellites would be one alternative, and some financial companies use GPS time for their stamps, but the signals are technically vulnerable. As MiFID II was published and the financial sector scrambled to build systems to comply with it, NPL spotted a gap in the market. Leon Lobo, an NPL business development manager, recently said, “At the microsecond level, I would say no-one in the City of London has the same time.”4 His job is to sell it to them.
THE FIRST THING that hits you is the noise. The rooms full of six-foot-tall, black-painted steel-mesh cabinets filled with computer servers do not hum, they roar. They are surrounded by fans and air-conditioning equipment to keep the equipment cool and the noise is so great you must shout your conversation, not that most people visiting these facilities are there for a chat. Telehouse North, which opened in 1990 near the old East India Dock in London’s Tower Hamlets, was Europe’s first purpose-built data center allowing businesses like those at nearby Canary Wharf to site their computer servers in the same building as those of their customers, partners, exchanges and super-fast connections to other facilities around the world. Now, the area has become a bleak city district filled with gigantic windowless structures, bristling with razor-wire fences and CCTV cameras.
Below the streets that divide up these structures is a dense network of fiber-optic cables over which the light-speed commands of modern finance pulse, day and night, serving the financial powerhouses nearby. But there is a single pair of fiber cables running into Telehouse North that is worth looking at more closely, as it has come a lot further than Canary Wharf. In fact, it goes all the way to NPL at Teddington, fifteen miles to the southwest.
I was at Telehouse North with an NPL scientist, Ali Ashkhasi, one of a small team running its fiber-optic time service; I wanted to see what the modern equivalent of Amsterdam’s 1611 Stock Exchange clock looked like. It is easy to assume that modern-day time signals, like cloud computing or electronic financial trading, are somehow virtual; that this all exists in what we call cyberspace, as if it has no material form. The reality is very different. This is a world of big, heavy, loud machines, using vast amounts of electricity and requiring industrial-scale cooling, which occupy huge buildings covering acres of valuable city real estate. It is a world of steel and concrete, bricks and mortar, security guards and staff canteens. As I discovered only too clearly during my visit to Telehouse with Ashkhasi, a lot of his working day is spent filling in forms, signing out passes and keys, and finding parking spaces. The lifts smell of lasagna. Cyberspace and the services that use it are decidedly real, not virtual. And so are the clocks that regulate today’s financial markets.
Telehouse data-center complex, London, photographed in 2020
The single pair of fiber-optic cables runs from NPL in Teddington on a circuitous forty-six-mile route through central London out to Telehouse North. First set up in 2014, the cables carry nothing else except time. At the Teddington end, two hydrogen maser clocks, each the size of a small domestic refrigerator, provide the time to a further pair of clocks known as the Grand Masters (one primary, one backup). The masers in turn are part of NPL’s ensemble of super-accurate atomic clocks that together form the UK’s version of UTC (which is checked against every other country’s UTC once a month in a laboratory near Paris in France). The Grand Masters feed their time into the fiber link. At the Docklands end, the time signal feeds into an identical pair of clocks in NPL’s steel cabinet buried in the Telehouse data center, which become, in their turn, the Grand Masters for all the customers subscribing to the time service.
