Ex-Goldman trader teams up with millennials to make AI dark pool

(From left) Richard Suth, Kelly Littlepage and Stephen Johnson have rented a 240 sqft office in a building on a quiet side street in Lower Manhattan. Bloomberg
Mathematical formulas, scrawled in red and black, cover a glass wall. The office is small, but the three men who work there, dressed in jeans and sneakers, hovering around an array of computers, have big plans. They want to run a stock market, and they say they and their algorithms can do what even their leanest competitors would need 80 human beings to achieve.

Kelly Littlepage, Stephen Johnson, and Richard Suth have rented a 240-square-foot office in a building on a quiet side street in Lower Manhattan. Next door, a shop sells black leather motorcycle jackets (“Ride or Die”) and retro lamps made from old cassette tapes. The three entrepreneurs say their company, OneChronos, will fundamentally alter the business of buying and selling stocks established more than two centuries ago by the New York Stock Exchange, located just over a mile away.

OneChronos is seeking to be the first venue trading mainstream securities to use artificial intelligence to match trades. Whereas stock exchanges use relatively simple math to match buyers and sellers trading shares of individual companies, OneChronos proposes matching trades among multiple stocks, trading in different currencies all at the same time. This would be impossible at the kind of stock exchange with which we’re all familiar. And that’s why OneChronos needs to use artificial intelligence, the great transformational force in finance today.

The company is currently awaiting approval from the US Securities and Exchange Commission and the Financial Industry Regulatory Authority to operate a trading venue known as a dark pool for US equities. Assuming regulators give it the go-ahead, it will match its first buyers and sellers later this year. Littlepage, Johnson, and Suth, who each have an equal say in running the firm, plan to add European equities and currencies as soon as possible after that. “Our market is for someone who wants to sell out of Apple, buy a European equity, and do the FX trade all at the same time,” says Littlepage, 30, who founded OneChronos in 2016 with his schoolmate Johnson, 31. “If we didn’t use AI, it would take longer to match those trades than the amount of time left in the universe.”

In markets at least, AI has mostly been used to generate returns. OneChronos is different: It wants to generate matching orders. It’s promising to complete trades at a lower total cost than any competitor. Its fees will be low, but, more important, the company wants to show that trades on its platform move the market less than trades on other platforms, alleviating the concerns of big fund managers who live in fear that the value of their shares will drop when they have a stake to sell and climb when they have a stake to buy. When that happens, Littlepage says, “it’s death by a thousand paper cuts.”

OneChronos is the brainchild of Littlepage and Johnson. They were in the same running club at Arapahoe High School in a Denver suburb in the early 2000s before Littlepage went on to study applied computational mathematics at the California Institute of Technology. There, he was taught by Preston McAfee, who now works at Microsoft  as the technology giant’s chief economist.

It was McAfee who introduced Littlepage to the math of combinatorial auctions. US aviation regulators had pioneered the use of these auctions to allocate landing slots at key airports to rival airlines. The auctions had also been used to sell bands of wireless spectrum to competing mobile phone companies. McAfee saw they might be adapted to financial markets with their vast numbers of orders and near-instantaneous trades, but, as Littlepage says, the computing power to make that happen has only recently become available.

Littlepage had stayed in touch with Johnson, who went on to become a cybersecurity specialist at Accenture Plc. As well as sharing a sport, they liked to talk about how emerging technologies could solve mathematical problems. In the summer of 2016, Johnson and Littlepage, who’s worked at Crabel Capital Management LLC, a futures and currencies trader, decided to take their OneChronos idea to Y Combinator, a Silicon Valley startup funder. The seed accelerator gave them $120,000 to set up OneChronos. Another five venture capital firms, including DST Global and Data Collective, also invested.

One of the VC firms, Green Visor Capital Management, approached Richard Suth, 47, a former New York-based partner in the equities division of Goldman Sachs Group, to join the startup. Suth, whose role is to sign up customers, has met with hundreds of fund managers, banks, proprietary traders, and hedge funds in the past 12 months. Big financial firms won’t commit to use a new venue until it’s received regulatory approval and is ready to launch, but Suth is bullish. “We will have a good portion of the major market makers participating,” he says. “We expect a couple of dozen market makers and banks.”

Technology has driven change in financial markets since the NYSE was founded in the aftermath of the American War of Independence. Trading quickened as telegraph poles and then submarine cables replaced messengers on horseback. But the pace of change has been fastest in the last quarter of a century. In the 1990s, men (and they were almost always men) in colorful jackets shouted orders at one another in grand buildings with important-looking facades.

By the turn of the century, most markets had gone electronic. Sometimes men picked up telephones to shout orders at other men in distant offices. And then toward the end of the last decade, electronic trading fragmented. While some of it stayed on the exchanges, a lot moved to new venues—including dark pools—operated by banks or start-ups. A new breed of trader—the high-speed, or proprietary, trader—came to dominate trading of stocks and currencies. The men (they were still mostly men) hung up their telephones and sought alternative employment.

Whether speed traders will flock to OneChronos is a different matter. A senior executive at a large algorithmic trader says the company’s processes are just too complex, while an exec at another major trading house says the new venue will only work if firms supply it with lots of orders in every asset class. But Larry Tabb, the founder of New York research firm Tabb Group LLC, is more sanguine about OneChronos’s opportunities.

“The whole way of thinking about market making is changing,” he says. “A couple of years ago, the buy side would not listen to this idea. Now the bigger guys are all doing it.”

Several stock markets already offer auction services. Cboe Global Markets Inc. operates the largest so-called periodic-auction service in Europe, while London Stock Exchange Group Plc’s Turquoise division and Goldman Sachs also run auctions that last roughly 100 milliseconds—about the length of time it takes a­ hummingbird to flap its wings.

OneChronos is different from these other auctions in that it uses AI to choose which of the mathematically possible matches will result in the maximum number of shares changing hands. That may deter some potential traders because the machine-learning system that OneChronos relies on is autonomous; it cannot explain why it chose the match it did. “The thing about machine learning is that it’s a black box; you do not know why the machine is doing something,” says Tim Cartledge, the global head of FX at NEX Group Plc’s markets division. “Having a black box in the model, that would put the fear of God into most people.”

Littlepage and Johnson, who train the machine, think they can dispel such fears. They have set upper and lower limits on the number of shares that can be matched by the AI during a single auction. The lower limit is the number of shares that would have changed hands if OneChronos was run like a stock exchange, only matching orders of individual stocks during continuous trading. The upper limit is the theoretical maximum number of shares that could be traded if the AI had much longer to find an answer.

OneChronos may be alone in running an AI-powered stock market, but it faces lots of competition. The U.S. already has 12 stock exchanges and 32 off-exchange markets or dark pools. IEX Group Inc.—the last eye-catching new exchange, which began life as a dark pool in 2013—accounts for little more than 2 percent of U.S. equity trading. How much volume does OneChronos need to succeed? “For equities, even less than 2 percent would be plenty,” Johnson says. “We want to get some size done there and then move into different asset classes.”

For Suth, the real opportunity lies on the other side of the Atlantic. Europe’s recent overhaul of markets rules, MiFID II, has prompted fund managers to make greater use of auction services. And having six different major currencies across Europe ensures that any basket of stock trades is likely to involve a currency transaction, which OneChronos can handle in its auctions. “The U.S. is not the golden goose,” Suth says. “It’s a huge market, but the European market is potentially better.”

Without a regulatory green light, the potential and ambitions clung to by Suth, Littlepage, and Johnson amount to little. But if the regulators give the go-ahead, OneChronos may well need to look for new—and larger—digs.

Hadfield covers market structure for Bloomberg News in London. Massa writes about finance in New York.

Business Standard is now on Telegram.
For insightful reports and views on business, markets, politics and other issues, subscribe to our official Telegram channel