DARK POOLS: The Rise of AI Trading Machines and the Looming Threat to Wall Street
by Scott Patterson
Random House Business, 368pp
You can quit worrying about naked short-sellers or traditional market manipulators, says Scott Patterson — the ‘bot algos’ (algorithms) have taken over. Dark pools are private markets hidden from the average investor — the opposite of displayed or ‘lit’ markets such as the NYSE or Nasdaq.
The vast computerised systems of the exchanges were designed by ‘plumbers’, but then a new breed of trader emerged who, says Patterson, ‘focused on gaming the plumbing itself, exploiting complex loopholes and quirks inside the blueprints like card-counters ferreting out weaknesses in a blackjack dealer’s hand’.
Because robot trading machines were front-running long-term investors such as mutual funds on the lit markets, anticipating their buy or sell orders and driving the prices up or down accordingly, those long-term investors began switching to dark pools. By 2012, 40 per cent of all trading volume was taking place in dark pools. But now the dark pools are swarming with predator bots as well.
This may sound like science fiction — and Patterson, like Dickens’s fat boy, ‘wants to make your flesh creep’ — but it is actually happening. The Flash Crash of 6 May 2010, during which the Dow Jones index fell by 800 points, was caused by technical glitches in the system and horrified the regulators. There is even a name for an event that hasn’t happened yet but might, if this were to occur on a global scale — Splash Crash — and that could cause what Bank of England economist Andrew Haldane has called a ‘double liquidity void’ (a lack of short- and long-term buying).
Patterson, who previously wrote a bestselling book called The Quants, about the mathematical whizz-kids behind the derivatives that caused the global financial crisis, has traced the history of the bots (and the human personalities behind them) from the introduction of computerised trading in the early Nineties. Ever since 1996, when computer programmer Josh Levine created Island, a trading system that bypassed market makers, the market has been dominated by trading machines. What followed Levine’s Frankenstein moment became known as the Algo wars, in which highly leveraged, high-frequency traders vied to be the fastest, with talk last year of one day being able to measure trades in picoseconds (trillionths of a second).
As in an arms race, both sides are constantly seeking to outwit and outdo each other in technological terms. Algos were ‘like hunted prey attempting to cover up their tracks through feints and dodges’, but hunter-seeker radars ‘adapted to the new stealth techniques and watched for them, anticipating every move’. With names like Shark, Guerilla, Stealth, Thor and Sniper, by 2011 the ‘mindless algos had evolved into dangerous beasts of prey’.
High-frequency trading became so competitive that no trader could make money at high volumes. The big scandal has been that some of the high-frequency traders have been collaborating with the exchanges to clip ordinary investors in a rigged market. Only trading companies that were in on the arrangement are able to survive by using exotic order types — routing or trading instructions.
At the same time, Cerebellum Capital, a fund with no highly paid traders but an ‘Invention Machine’, was able to generate a steady return of 7 per cent through mini-robot-traders. ‘The machine would kill off the traders that had done the worst and shift money to the traders that performed the best. This would lead to mutations in the strategies — entirely new algorithms.’
But the holy grail of artificial intelligence on Wall Street is to create the AI-equivalent of Warren Buffett. In 2007 a hedge fund named Rebellion launched an AI programme called Star. Its designer, Spencer Greenberg, told a gathering of quants in February 2011: ‘The goal is to have our software learn, on its own, to become a long-term-oriented stock investor. We do not assume that we already know how to invest, and are not using machine learning just to optimise a few parameters in our model. Rather, we are leaving it up to our learning algorithm to learn to invest.’
So will the SEC be able to keep up? It is planning to create its own machine called CAT (Consolidated Audit Trail) to detect manipulative patterns, but it will cost billions. And even Greenberg, the cheerleader for AI on Wall Street, has said: ‘Machine learning can be disastrous in the hands of people who don’t know what they’re doing.’
Patterson is adept at explaining the abstruse strategies of the trading machines and at sketching the characters of the men behind the machines. If his prose is a little lurid and frantic, it befits his subject. We should be afraid. We should be very afraid.