To Gain 0.00007 Seconds, a Trading Firm Paid $14 Million for an Empty Lot | QDaily Data

For high-frequency trading firms, two-thirds the speed of light is still too slow

A blink takes about 0.4 seconds, and we use it as shorthand for fast. One company spent $14 million just to get 0.07 milliseconds faster — 0.00007 seconds, or one 5,700th of a blink.

Jump Trading bought a 120,000-square-meter empty lot across from the data center of the Chicago Mercantile Exchange (CME), the world's largest futures exchange.

It did not put up a building to flip, and this had nothing to do with feng shui. It built a microwave relay tower, to get its trading orders to the CME the instant they were sent.

Jump Trading hasn't said how much time this actually buys, but judging by the gain Nasdaq got from putting a microwave tower on its roof, it is only 0.07 milliseconds.

That gain is hard to come by, because the link was already fast. Ordinary fiber carries a signal at two-thirds the speed of light. Microwaves travel at the speed of light, 50% faster. And fiber is rarely laid in a straight line.

For a trader, seeing the data 0.00007 seconds sooner rather than 0.00014 seconds sooner changes nothing. It takes 0.15 to 0.225 seconds for what the eye sees to register in the brain.

But Jump Trading isn't spending the money for people. It is spending it to make its computers a little faster.

The Trade Is Done Inside 0.2 Seconds

Jump Trading is a high-frequency trading firm.

Whenever a listed company reports earnings, or the government announces a rate change, CPI figures, or anything else that moves financial markets, its servers run pre-written, complicated algorithms to work out where a stock's price will go over the next few seconds, and the computers buy and sell at speed according to a set strategy.

Because high-frequency trading works in such a short window, even a tiny move in a stock's price is enough to book a profit or cut a loss. That is why every high-frequency trader wants to get the news, and finish the trade, before everyone else.

This isn't the first time Jump Trading has bought infrastructure to trade faster than the rest. In 2013 it bought a former NATO microwave tower in the UK, purely to get data to the London Commodity Exchange sooner.

How does one of these trades actually happen?

The Wall Street Journal once laid out an example. At 4:00 p.m. Eastern on December 5, 2014, the cosmetics retailer Ulta released its earnings; the stock was at $122. What followed:

4:00:00.15 p.m. — Business Wire sends the earnings release to high-frequency trading firms and to trading terminals such as Bloomberg's.

4:00:00.20 p.m. — High-frequency trading firms sell $800,000 of Ulta stock at $122.

4:00:00.242 p.m. — Bloomberg is the first to publish the Ulta earnings story;

4:00:00.464 p.m. — Dow Jones publishes it;

4:00:00.7 p.m. — Ulta's stock has already fallen to $118;

4:00:01 p.m. — Thomson Reuters publishes the Ulta earnings story.

We can say with some confidence that by this point no human trader had finished reading the headline on the release. And only 0.85 seconds had passed.

No person can compete with that.

The episode caught the attention of the US Securities and Exchange Commission, and Business Wire came under pressure from regulators. Its CEO, Cathy Baron Tamraz, maintained that the company “hadn't done anything wrong”, but she also thought the coverage “would leave people with serious misconceptions about us, and I worry it will have a lasting effect on the company.” So after consulting Warren Buffett, its controlling shareholder, Business Wire shut down the business of feeding financial data straight to selected paying subscribers.

So high-frequency trading firms now have to wait for Bloomberg to publish before they can start trading — 0.192 seconds later than before.

Sounds fairer? For the machines, it is.

And that is why trading firms pour money into speed without end.

Microwave, Vacuum Fiber, Undersea Cable, Laser — the Most Advanced Technology All Goes Into Trading

Fiber networks are usually the first choice for a long-distance, high-speed link. Trading firms prefer microwave for speed because fiber in the US mostly follows the railways, which makes the cable route longer than the actual distance between two points; inside the cable the light does not run straight but bounces along at angles; and hemmed in by the glass it passes through, the signal moves at only about 200,000 kilometers per second.

Microwaves move through air at close to 300,000 kilometers per second. Firms build microwave towers — usually on hilltops or on the roofs of tall buildings — and relay the signal from one to the next, keeping the distance as short as they can, to get it from one place to another as fast as possible. Because this is close to a straight line, it beats fiber, which bends along its route.

And for Jump Trading, what could be faster than a microwave tower across the street from the CME?

In fact the CME's own data center also relays data by microwave. In 2015 McKay sold the exchange a plot of land for a data center. Last month, DuPage County, near Aurora, approved McKay's plan for a new microwave-tower data center, with the new tower 188 meters closer to the CME's trading center than before.

Same goal as Jump Trading's: to save even 0.00007 seconds.

Microwave has its weaknesses. The signal is easily disturbed by bad weather and cannot reach the speed of light in a vacuum. Rain is the worst: raindrops happen to be exactly the right size to absorb the energy of a millimeter-wave signal, which leaves millimeter wave only about 90% reliable.

It also runs short of bandwidth. Anova, which builds high-speed networks specifically for investment firms, gets only 100 Mbps out of a single microwave base station; fiber carries 1,000 times that. So although a microwave signal is faster than fiber, sending a lot of data at once means a longer wait for it all to arrive. Millimeter wave solves the bandwidth problem, but it does not carry as far — under 1,500 meters — and it is even more exposed to the weather.

A high-frequency trading firm does not buy and sell one company's stock at a time. It usually trades thousands of stocks at once, plus the stock indices, futures and currencies that its various strategies call for.

So fiber and undersea cable, “slow” as they are, suit the trades that carry a lot of data — when a listed company puts out an earnings file a few megabytes in size, for instance. Microwave is saved for the trades where speed matters most.

Spread Networks, another high-speed network provider, started around 2010 to dig a fiber tunnel straight through the Appalachian Mountains, at a cost of $300 million, so that the cable would not have to go around — cutting about 3 milliseconds off transmission time.

By 2013, an icebreaking crew and a purpose-built polar cable ship were due to start laying the first trans-Arctic undersea cable. Two of them, “Arctic Fibre” and “Arctic Link”, would run through the Northwest Passage in Canada's Arctic Archipelago. A third on a similar path, Russia's trans-Arctic undersea cable (ROTACS), would be laid around the Scandinavian Peninsula in northern Europe and along Russia.

Those three cables would connect London and Tokyo, two of the world's largest financial trading centers, almost directly — nearly 8,000 kilometers shorter than the existing route, and transmission time between the two cities down from about 0.23 seconds to 0.17 seconds. The project would cost around $1.5 billion.

There is also an approach that offers both microwave's speed and fiber's bandwidth: laser.

Anova has installed white equipment on rooftops between Manhattan in New York and Carteret, New Jersey, home to the NYSE's and Nasdaq's data centers. These boxes, which look like oversized telescopes, are the base stations of Anova's laser network, and they can shoot an infrared beam as thin as a hair 10 kilometers.

The two exchanges' data centers are about 55 kilometers apart in a straight line. Anova will build six or seven laser stations along the way, creating the most direct channel possible for trading data, to replace fiber and microwave wireless networks.

Anova CEO Michael Persico told The Wall Street Journal that in practice laser runs about twice as fast as fiber. The company has also put equipment at 1275 K Street in Washington, DC, near one of the US government's data centers, just to receive economic data the moment it lands.

On bandwidth, Anova's laser network delivers 2 Gbps, a few hundred times what the firm's earlier microwave network offered. And the infrared laser it uses is barely affected by weather; it carries data normally through rain and heavy fog.

The catch is that a laser only travels in a straight line, so Anova has to mount its transmitters on the roofs of tall buildings to reach further. Tall buildings sway a little back and forth in the wind, and the hard part is landing the beam on a receiver plate the size of a pot lid — and keeping it there for the whole transmission.

High-Frequency Trading Firms Say It Makes Markets More Efficient. What Value It Creates Is Harder to Say

“High-frequency trading is drawing more and more attention, and broad criticism from many market participants. What are they doing wrong?” The Wall Street Journal asked Larry Tabb.

“Markets today are more efficient than they have ever been; high-frequency trading is what makes all of these strategies work,” Tabb answered. He founded the consultancy Tabb Group, has studied high-frequency and quantitative trading for years, and is a believer in the strategy. His point is that institutions can now finish a trade within milliseconds, and technology is what changed that. From its first appearance to its rise, barely thirty years have passed.

But what high-frequency trading firms save, at bottom, is their own time: buy and sell faster, and in the end make money more efficiently.

“The ultimate hack” is what Mark Cuban — owner of the NBA's Dallas Mavericks, a tech investor, and one of high-frequency trading's critics — calls these firms.

Shaving a few milliseconds off a trade is a numbers game inside the trading process. It has nothing to do with what a company is worth.

Warren Buffett has also mocked, hard, the kind of investing that leans on complicated formulas instead of seeing what a company actually is. In 2005 he staked $1 million on a 10-year bet that hedge funds could not beat an index fund. Ted Seides, a partner at Protégé Partners, took him up on it in 2007.

When the ten years were up, Buffett wrote in his annual letter to shareholders this February that the index fund he chose had grown 7.1% a year on average, while the five hedge funds Protégé Partners chose had grown 2.2% a year. Buffett's fund returned more than three times as much.

Hedge fund managers' pay has slipped as well. On Institutional Investor's ranking of hedge fund manager income, only the managers of Renaissance Technologies and Bridgewater Associates made more than $1 billion in 2016. Their returns are in fact lower: for several years running, they have not kept up with the market as a whole.

Early on — when Amazon founder Jeff Bezos was doing computerized trading at D.E. Shaw in the 1980s, for instance — a computer could put a trading firm far ahead of its Wall Street rivals.

But today more and more firms trade at high frequency and let computers run their funds outright, and every one of them is fast. In a sense the market has become fairer, and the returns for everyone in it are nothing like what they were.

Even so, computers taking over from human traders is still the trend. This March, BlackRock — the world's largest investment manager, with $5.1 trillion under management — started using artificial intelligence to pick stocks, and cut more than 30 analysts and fund managers, about 7% of that department's headcount.

In a race where the time left to save approaches zero and machines take over from people, no one can stop.

Graphics: Feng Xiuxia

Cover image/QuantStart