Regulators Ask Humans How to Deal With Robot Traders

Matt Levine is a Bloomberg View columnist. He was an editor of Dealbreaker, an investment banker at Goldman Sachs, a mergers and acquisitions lawyer at Wachtell, Lipton, Rosen & Katz and a clerk for the U.S. Court of Appeals for the Third Circuit.
Read More.
a | A

My potted history of U.S. financial exchanges goes something like this:

  • Once some gentlemen got together to trade stocks under a buttonwood tree.
  • As time went by, people started trading more things under different types of trees, or even indoors, and developed more and more sophisticated mechanisms to decide what things to buy and sell and how to rip each other off.
  • Then someone discovered that computers could trade even faster than people, could make use of even more sophisticated mechanisms to rip each other offer, and could dispense with the trees entirely.*

Today the Commodity Futures Trading Commission published a "concept release on risk controls and system safeguards for automated trading environments" and, despite that title, it's sort of a fascinating document. The current regulation of exchanges is still mostly a holdover from the second stage of my silly history. And this CFTC document -- which discusses broad approaches to regulation of computerized trading and asks market participants for input and recommendations -- is part of an effort to move that regulation into the third phase, where the markets actually are.

One way to think about the transition is to notice that, in the second phase, diversity was mostly a good thing, and so regulation tried not to second-guess traders' decisions. Different views make a market, and different market participants with different trading strategies, different sources of information, and different plans for how to profit off each other make for a more robust market than one where everyone has the same perspective and is after the same thing. You want to sell a stock at 10 percent below its last sale? Well, you probably have a reason for that, who are we to stop you? You want to sell a stock because of a rumor you read on Twitter? Sure, why not, everybody's getting their information from somewhere.

The rules have, historically, been mostly about preventing people from ripping each other off in ways that were deemed unsporting (insider trading, outright fraud, etc.). Even this wasn't always necessary: Plenty of unsporting practices have, historically, been punished not with CFTC sanctions but with everybody refusing to trade with you because you're a jerk.

But a market dominated by computers looks a bit different. They're so fast, and they're all programmed by like the same 12 guys using the same mathematical models, and they have no emotional intelligence. So much more so than with humans, there's an alarming chance they'll all think and do the same thing.** If that thing is wrong, then it's a big problem.

So a lot of this CFTC concept release is really aimed at preventing algorithms from making wrong decisions. The classic wrong decision that for some reason no algorithm can resist is, "I will sell tons of this $50 stock for a penny," and variants thereon, so much of the CFTC release discusses redundant systems to prevent computers from selling tons of futures contracts*** for a penny for no good reason. "Message and execution throttles," "volatility awareness alerts," "self-trade controls," "price collars," "trading pauses" and trade cancellation policies are all intended at least in part to defend against computers accidentally making dumb purchase and sale decisions. "Market data reasonability checks," also discussed in the paper, are designed to prevent computers from reading Twitter stupidly and falling for fake news. In the olden days, if you fell for fake news, that was your problem. If trading robots fall for fake news, it's everyone's problem.

Of course there's also a lot here designed to prevent people -- sorry, computers! -- from ripping off other people/computers. The CFTC discusses market quality indicators and incentives, for instance, which refers basically to the regulator's desire to actually have market makers quoting prices that they're willing to trade at rather than just sending flickering quotes into the market to try to pick someone off. Once, the incentives for market quality were largely reputational: You were in the market, and if you developed a reputation for scammy trading practices no one would trade with you. (Classically, if you quote a market and then someone calls you and asks to trade at your market and you say "oh, no, I've changed my mind," even if you're not in an exchange that prohibits that, you at least Look Bad.) With electronic trading, nobody knows you're a dog, so there's lots of opportunity to submit a million quotes a second with the goal not of providing liquidity but of pulling a fast one on someone just a bit slower than you.

The challenge of algorithmic trading is that, increasingly, regulators and exchanges will have to decide -- in advance, and in black-and-white objective computer-programmable ways -- what sorts of trading decisions are right and wrong, and then program robust systems to prevent the wrong ones. The old version of financial regulation let you do dumb things because, well, who were the regulators to decide what's dumb -- and because every so often selling a stock at way below its last trading price, or buying a whole lot of stock all at once, or relying on a badly sourced rumor, or quickly pulling down your quote as soon as you put it up, actually makes sense. It is -- was -- a matter for the human judgment of market participants. When the market participants, not being human, have no human judgment, it becomes increasingly a matter for regulators.

* Ooh, absurd fact: this is not quite true! Instead:

The NYSE Euronext data center in Mahwah, New Jersey serves as a bridge between the New York Stock Exchange's history as the nation's oldest trading floor, and a future in which the majority of trading volume will be driven by computers. The halls of the facility are lined with street signs bearing the names of streets in downtown Manhattan, which provide a connection between Wall Street and the data center, and also create a navigational grid for staff within the 400,000 square foot building. ... The grounds of the building feature six buttonwood trees, providing a reminder of the buttonwood tree on Wall Street under which 24 brokers formed the New York Stock Exchange in 1792 ...

Presumably the computers don't care though.

** Consider the various other games -- tic tac toe, Ghost, checkers, eventually chess -- where a diversity of human styles are demonstrably inferior to a single correct computer strategy. Now imagine that there were trillions of dollars invested in winning at Ghost.

*** The CFTC regulates sort of everything but stocks. Including in particular futures contracts, including the S&P futures, which have flash-crashed in the past.

This column does not necessarily reflect the opinion of Bloomberg View's editorial board or Bloomberg LP, its owners and investors.

To contact the author on this story:
Matthew S Levine at