Ferocious bulls: as on Wall Street hooked up to bid at the speed of light

One of the most interesting things about the collapse of the company Knight Capital Group - a trading company that lost $ 440 million in 2012 - the speed of its collapse. [The original article is dated 2012 - note of the translation.] They wrote in the news that the main part of the wrong trading decisions occurred within an hour, and this computer-controlled plum again forced the financial community to discuss whether the pursuit of profit led to quick, but dull and uncontrolled software solutions. We post this story before its appearance in the September issue of Wired, because it describes how, on Wall Street, they have reached a state in which such failures happen more and more often and what traders who are looking for ever faster trading are ready for.

The two-day conference on algorithmic trading systems “The Battle of the Quants” was held in New York in 2012 just a few days after a group of researchers admitted that they had made a mistake in an experiment trying to turn modern physics. Scientists decided that they found subatomic particles called neutrinos, which moved faster than light. But they were wrong; after six months they recalled their discovery. And while the most predictable headline in the world remains “The speed of light did not give in,” the news that neutrinos obey the known laws of physics, means the end of seductive dreams for “techies” - physicists, engineers and mathematicians who went to finance, who are responsible for 55% all trades in usa. In search of ways to beat the market, the ability to send messages faster than light could provide the ideal head start: the ability to trade in the past, as if we could bet on a horse after the race was over.

“From the moment the first work came out in September to the last week, one comrade in my company managed to write two works explaining how it is possible,” the graying former physicist sadly says, sipping coffee close to Keith Haring’s gigantic picture dominating the room’s decoration auction Christie, in which the conference was held. “Of course, for this you would need a particle accelerator.”
')
If it were only this, then someone would have built it for himself. One of the main topics of the conference was “race to a minimum”, competition without regard to cost in an attempt to reach the theoretical minimum of the time of trading. This variable, called “latency,” is rapidly approaching the physical minimum that exists in the Universe due to the laws of quantum mechanics and the theory of relativity. But even Einstein hardly understood how much electromagnetic waves can be bent in the presence of money. Kevin McPartland of the Tabb Group, which handles financial industry information, predicted that in 2010 companies would spend $ 2.2 billion on trade infrastructures — high-speed servers that process bidding on the optical fiber linking them to the global network. That was before projects were launched that connected New York with London with new transatlantic cables and London with Tokyo across the Atlantic Ocean, all to cut off a few hundredths of a second from the time needed to receive data or send an application.

High-speed traders are a kind of so-called. “Techies”, investors earning in a new-fashioned way: shares of a cent at a time, multiplied by hundreds of stocks, tens of thousands of times a day. They occupy a strange position on Wall Street, feeding themselves with a mixture of modesty and arrogance, which distinguishes them from the simple and uncomplicated arrogance of investment bankers. One of the first high-speed trading companies, Tradeworx, has its rather modest offices two floors above the offices of Urban Outfitters in the sleepy suburbs of New Jersey. Twenty people work there, half of whom are sitting on the trading floor, tracking on triple screens, as they accumulate a second by a second of kopecks. Approximately one and a half percent of the total volume of trade on the US exchanges will pass through this quiet sunlit room with brick walls on any given day.

On the first day of the New York conference, Aaron Brown, the famous “techie” and former professional poker player, stepped onto the stage in crumpled cotton pants and a leather jacket to give a talk on game theory. He began his speech by dictating "3.14159" and making a pause. From the back of the room came the "265358". Thus, they compiled the first 12 digits of the Picogik password number. “You won't find big buds here,” says Charles Jones, a professor of finance and economics at Columbia Business School. “Most of them are such that if you see someone in a tie, it will be their only tie.”

The wheels of finance are spinning faster and faster than increasing the risk of spinning out of control - a disturbance of the system can grow to a global crisis in a few seconds. “For the first time in the history of finance, cars can make deals much faster than people can interfere with them,” said Andrew Haldane, a regulatory official from the Bank of England at another recent conference. “And this gap will widen.”

This movement has picked up speed for more than ten years. People investing on the basis of their economic assessments and analysis of individual companies are leaving. Their place is taken by decision-making computers based on market trends and news feeds generated by other computers, communicating only with each other. Conventional economists see this as an ideal option: it is considered an axiom of the economy that all transactions benefit the economy because they increase “liquidity”, the ability of investors to sell or buy goods at the best price. Indeed, in 2007, the US Stock Exchange Commission created new rules for the game: the state market system, which opened the doors to dozens of new trading platforms. But now the time for one transaction is calculated in microseconds, and prices are calculated to the sixth digit, so these possibilities have already gone beyond the simple problem of reducing profits.

So, excluding possible breakthroughs in physics, we have already approached the last frontiers of the trend, begun by the Rothschilds, who according to legend used post pigeons to beat the result of the Battle of Waterloo with profit. For about a hundred years, until 1970, the highest technical achievement remained the telegraph (for receiving data) and the telephone (for transmitting orders). Now this is a high-speed server connected to the exchange via a fiber-optic channel of the smallest possible length, since each mile adds 8 microseconds of delay. Cash profits are more than worth the costs of research and the cost of creating the necessary infrastructure that can save these microseconds.

Everything rests on the iron. DC NYSE Euronext, an international conglomerate that includes the New York Stock Exchange, is located in a building in Mavi, a suburb of New Jersey, 27 miles from Wall Street. In addition to computers processing all transactions, there are high-speed trading servers that receive data and spit out commands based on software algorithms. Traders pay for hosting servers in the same building, and to create equal conditions, engineers specifically make some cables longer to equalize them among all the servers there. Yes, yes, give or take a few meters. Almost at the speed of light.

Now multiply these attempts by the width of the continent or ocean. Add the quantum and relativistic effects of machine-to-machine communication, not requiring human intervention, occurring faster than any human reaction. It all started with pigeons, and ends with subatomic particles, transferring data, becoming obsolete almost before they arrive at their destination.

Not all trade goes in New York. Historically, derivatives (futures and options) are traded mainly on the Chicago Mercantile Exchange, 720 miles from there. And a few years ago, Spread Networks began quietly buying back the rights to laying cable on parcels of land, which would reduce the length of the fiber optic cable between the Chicago stock exchange and the main DC Nasdaq in Carteret, New Jersey, by about 140 miles.

Existing networks usually follow the railway and are intended for servicing settlements, and not for communication of traders. Instead of turning south towards Philadelphia, the Spread Networks path goes northwest through Pennsylvania and then turns west to Cleveland. Delays are usually measured by the time required to go both ways (for example, sending an order and the arrival of confirmation). The shortest path before the appearance of Spread Networks required 14.5 ms for the signal to pass, but its throughput was low and customers had to be content with a delay of 15.9 ms. Spread Networks lowered this figure to 13.1 ms in the case of the premium “dark fiber” tariff - a connection that you do not share with other customers. Prices for services are kept secret, but according to traders, they are "quite high."

Spread was quickly gaining customers, but by the spring of 2012 a faster competitor appeared. According to the laws of physics, the speed of light propagation in a medium is inversely proportional to the refractive index of the medium. Therefore, light travels through fiber at a speed of 200,000 km / s, not 300,000, as in the atmosphere. The fastest method of communication between New York and Chicago will be a line of sight through the air, for which you want to build a chain of microwave transmitters. Such a network is being built by Tradeworx and McKay Brothers, a Californian company that expects their system to be the fastest, with a delay of no more than 9 ms. Their path, according to co-founder Bob Miad [Bob Meade], uses a minimum number of towers - 20, and deviates from the geodesic line (the shortest path along the surface of the planet) by only 4 miles. It includes about two miles of cables in the towers themselves, connecting the dish antennas raised a hundred meters above the ground, with ground amplifiers.

The disadvantage is that microwave radiation at a frequency of 11 GHz can deflect storms and special atmospheric phenomena. Spread Networks Director David Barksdale [David Barksdale] declares the reliability of his system in the "five nines" - 99.999% - and claims that he is not afraid of microwave competitors. “People have been talking about building low-latency wireless networks for years. We do not believe that these microwaves are well suited for complex trading applications due to certain limitations ”(which he refused to list). But Miad, a former Harvard physicist and techie, is convinced that speed is more important than reliability. If your connection is reliable at 99%, you do not earn money 1% of the time. If it is slower than the competition, you do not earn money 100% of the time.

Other critical paths are from New York to London and from London to Tokyo. (Trading hours in the United States do not intersect with the Asian, so the demand for ultra-fast connection from New York to Tokyo is small). At least three companies have announced plans to lay fiber optic cables under the Atlantic Ocean between Europe and Japan. One path borders the coast of Russia and goes ashore on the northern tip of Murmansk; the other is on the northeast route through the Canadian Arctic spaces. When they start work, they will reduce the delay from 230 ms for existing paths through Asia to 155-168 ms.

At this time, manufacturers began manufacturing a cable for a new connection New York - London, designed to cut 311 miles from the existing path and reduce the delay from 65 ms to 60. To do this, the cable will pass through the shallow water of the Great Newfoundland Bank . Most transatlantic cables immediately go to a depth to avoid sharks. For reasons of almost karmic justice, sharks threaten the financial industry by biting cables. They are attracted to the electromagnetic fields generated by the wires feeding the amplifiers located at certain intervals along their entire length. Near the coast, cables must be enclosed in expensive protective armor against sharks and preferably immersed in order to protect them from fishing trawlers and anchors. The new cable will be protected for 60% of the entire length due to the shortest path along which it will be laid. In the summer of 2013, two ships will start laying a cable to meet in the middle of the Atlantic in three months - this project the company called Project Express [ project completed successfully - note of translation]. The cost of the project will be about $ 300 million. Estimated time to life before it loses relevance: hard to say. But what else to do? Unlike the New York - Chicago route, the Atlantic Ocean is very poorly suited for installing microwave towers. On the other hand, when I asked Alexander Dziejma, the main architect of the high-frequency trading company, Dymaxion Capital Management, at the Battle of the Quants conference, he grinned.

“Now with the help of drones you can do amazing things,” said Dzayima.
"Dronov"?
Of course, he answered. A fleet of solar-powered drones carrying microwave repeaters can hover at even intervals across the Atlantic. I began to argue for the insanity of this idea, and then I realized that I had already heard a dozen more insane ideas that day. “Someone, in the end, will accomplish this,” he said.


168 batteries provide backup power for a period of more than 24 hours of DC Spread Networks companies in Carteret, New Jersey.

High-speed traders make money from a vacuum, snatching pennies that appear and disappear like virtual particles of quantum field theory. Their goal is to end the trading day by exiting the market so that the profits are safe in the bank. Depending on the model of work, they can win only 55% of transactions. They check prices all the time, looking for relationships and trends, or chances to buy something for $ 1 and sell somewhere else for $ 1.01, or $ 1,001. Sometimes they don't even want to make money on a deal. According to the seller-buyer model, some exchanges offer incentive rewards, or interest deduction, for placing quotations, as a result of which a transaction is made. The exchange charges a little more than a commission from the party responsible for the transaction and assigns the difference. So the algorithm can buy shares, get a reward, then sell them and get a reward for them.

All this is controlled by algorithms that can live no more than a few weeks. Sometimes the algorithm simply looks for stocks that go up in price for several transactions in a row. An “instant” algorithm can buy stocks in anticipation of their continued growth. The algorithm of "average return" can sell shares, waiting for them to fall to an average value. Both of them can work from the same company. Within a minute, both of them may be right.

One of the common algorithms is to search for pairs of stocks, the prices of which are historically related. The canonical examples are oil company stocks, rising with rising oil prices and airline stocks going in the opposite direction. But they can not do it at the same time, so one of the strategies is to buy or sell lagging stocks and wait for them to catch up. Similarly, derivatives, options or futures can be out of balance with major stocks. Some algorithms are trying to control the market - they are trying to buy at a low price and quickly sell at a slightly larger price, pocketing the difference, or spread. The people involved were called specialists, and when the spreads reached 12 cents, with this money it was possible to exist quite well. But after the New York Stock Exchange switched to the metric system in 2001, the spreads dropped to 1-2 cents, and to receive a similar income, it was necessary to operate with a large number of shares and do it much faster. And this is not a task for people.

A more specific example: a simplified algorithm presented at the conference by Mani Majury [Mani Mahjouri], Tradeworx General Director for Investment. His imaginative algorithm buys and sells SPY shares - securities tied to the level of the S & P 500 index, based on the value of the 500 leading companies listed on the stock exchange. SPY is traded on various exchanges, including Nasdaq in New York, but in Chicago there is a trade in S & P 500 futures - contracts, the price of which reflects the rates on the index value in the future, in a few weeks or months.

The cost of SPY and futures usually move together, but not synchronously; as a rule, futures are ahead of SPY by a few milliseconds. Why this happens, it does not matter, the main thing is that this rule is observed often enough for the algorithm to earn money by tracking changes in S & P futures and predicting changes in SPY that should occur in a few milliseconds. Earnings depend on how quickly the algorithm can transfer data from Chicago to New York. As an experiment, Maury applied this scheme to the history of trading for the day and found that, ideally - when data is transmitted at the speed of light - such an algorithm would conduct 64,000 transactions with an average profit of $ 0.0001 per share. The model set the trading volume for one transaction; Usually, high-speed traders operate with small volumes, several hundred stocks at a time, because operations with large volumes affect the market.Just for example, let's imagine that on average about 150 million SPY shares pass from hand to hand per day. Tradeworx claims that it processes 4% of all SPY trades, this results in 6 million SPY shares, multiplied by $ 0.0001 ... We get $ 600. Slow way to get rich. But we multiply this number by the number of simultaneously running algorithms, several hundred, and then something interesting begins to emerge. High-speed trading works in particular because for human intervention, everything happens too quickly.

What you can try by yourself: go to Yahoo Finance during trading hours, choose a company to track, say INTC - Intel. On the quotes page go to the Order Book. In the “Bid” column you will see 4-5 prices for reduction and a similar sheet with prices in the reverse order, in the “Ask” column. They submit bids to buy or sell a certain number of shares. Prices will be about equal and the top numbers should match the current value, reflecting the latest deals.

, , , . Intel, , , , . . . , , . , .

Yahoo . , , – , . . – . « , , , , 50000 »,- [Sal Arnuk], Themis Trading « » [Broken Markets]. , 90% . ; , , – , « » [quote stuffing]. , . « » [denial-of-service attack] , . “” .

, 12 – , - , . « , , , ,- [Eric Scott Hunsader], Nanex, , . – , . 2001 56. , ».



BATS . - : BATS BATS , , . , «A», 10% . , BATS, Nasdaq, - . 900 , , $15,25 $0,28, , .

BATS , IPO . , . , Nanex , . « , , , »,- . , , ; Nanex [Hunsader] , BATS, 10% .

Representatives of BATS do not comment on this theory, walked on the forums of stockbrokers in the spring of 2011. Other market watchers were skeptical. In general, according to the head of Tradeworx Manoj Narang [Manoj Narang], “computers do not engage in market manipulation, people do it. Any HF trader who incorporates the logic of manipulation into an algorithm that anyone can sue is too stupid to earn. ”

- . BATS , , , Flash Crash 6 2010 – Dow Jones Industrial Average 5 600 . - , - , 1929 . « , , »,- [Irene Aldridge], Battle of the Quants. 100- , «» «», , 2010 «- – „ “, ». , , , . $300 . .« - , ,- Themis Trading,- ».

- , : ? , -, , – . 1792 2006 - , , . NYSE Euronext, -. – , , , - . «, , »,- - [Thomas Philippon] Russell Sage Foundation. « , ?»

- , , , « , ». , , . . «» , - . , – Tradeworx.

. « , - ,- Subotnick Financial Services Center, . -. – , , 9:30 16:00 ». , [Joseph Saluzzi] Broken Markets, , - . .


-

. ; “” . , , , , . (Dow Jones & Company — ). , , , , . 2007 , – , . , , «» , , . « », , RavenPack, 100 000 , . « , „ “ , „“,- , - Dow Jones Institutional Markets. , , , , , , , .

. , , . GPS . , , , . „ “, [Johan Bollen], 10 2008 , , , „“, „“, „“, „“ „“. , „“ Dow Jones Industrial Average 2-6 . , , , , “” , : „ , “.

, , . , , , - [Alexander Wissner-Gross], , - - , . , , , . , , , , . [Dan Stancil] , 0,1 , , „/“. „ , ,- , “ ». , . IBM « ». , , , . “” , .

Buying neutrinos.