When the Securities and Exchange Commission (SEC) signed off on the new National Market System (Reg NMS) in 2005, the goal was to transform the 188-year-old New York Stock Exchange from an old school institution where specialists shouted from the trading floor into a hyper-efficient system of computers quietly matching orders in real time. This modern marketplace was meant to introduce competition, juice efficiency, and ensure investors got the best prices.
But in the decade since Reg NMS was enacted, the proliferation of exchanges has combined with rapid growth in computing power to create a market system that is shockingly complex. There are now more than 50 public and private exchanges, tethered together by a web of regulations that dictate how they must coordinate. It’s as if Comcast, Time Warner, and Charter were competing to send you Game of Thrones, every time you turned on your TV.
This tangle of systems is so complicated that its behavior often appears irrational. Even financial experts have a hard time understanding it. A crucial reason is that the most pernicious problems—latency arbitrage, flash crashes—are not, first and foremost, financial problems. They are computer science problems. On Friday the SEC will issue a ruling with bearing on one of those problems, so-called “latency arbitrage,” as it decides for or against an application by Investor’s Exchange (IEX) to become a public stock exchange.
Latency arbitrage is a byproduct of the fact that Reg NMS neglected to take into account the problems associated with building a distributed computer system. What is “real time” to people is not real time to computers. Computers always take some time to process and transmit information, even if it’s only microseconds. This delay, called latency, allows high-frequency traders to exploit tiny differences in communication between the exchanges to front run stock trades, earning millions by rapidly selling and buying for pennies more or less than they paid.
The SEC has occasionally made angry noises about speed in the market, but has rarely done anything to actively discourage it—instead preferring to wait for the market to develop its own solutions. IEX is one such solution to the latency arbitrage problem. As a private exchange, more commonly known as a “dark pool,” IEX slows all incoming trades down so that no matter how fast you may be, you can’t race ahead.
There’s just one problem: Latency isn’t a market problem, it’s a technical problem that’s been enshrined in regulation. Consequently, it may not be legal for a public exchange to intentionally disrupt it.
To flummox high-frequency traders, IEX uses a “speed bump” consisting of 38 miles of coiled fiber optic cable. The coil slows quotes and trades by about 350 microseconds (millionths of a second). In order to approve the IEX application, the SEC will need to reinterpret sections of the regulation that require quote data to be sent as quickly as possible. Such an interpretation would indicate that regulators are willing to endorse innovations that specifically undercut high-frequency traders. More importantly, it would be an acknowledgement that latency-related problems are ones worth solving.
If the SEC does revisit the regulations, it may open up a range of other speed-related issues to scrutiny. Latency arbitrage is the best known and easiest to understand problem created by Reg NMS, but there are many others, with ersatz names like quote stuffing and layering. Analysts have also connected the dots between high-frequency trading and flash crashes. That link is virtually impossible to prove definitively, but it’s much harder to imagine flash crashes happening in a market where latency wasn’t an issue.
The trick, of course, is to find a way to fix these problems without destroying all the things that work well about the US markets. By many measures, the current system is incredibly efficient. Spreads, or the difference between the lowest price being asked for a stock and the highest bid, are the narrowest they’ve ever been. And even assuming the worst impacts of latency-related issues, that harm is effectively diffused across the entirety of the market. No one person or business is being ruined by a loss of pennies-per-trade.
However, taken as a whole, the social costs could be substantial. The complexity of the market and the poor availability of data make it difficult to pin down the monetary value of latency arbitrage. Eric Budish, who studies market structure at the University of Chicago, has estimated that the arbitrage opportunities for a single pair of related stocks are worth as much as $75 million a year. Another estimate comes from Elaine Wah, a Ph.D. candidate at the University of Michigan, who estimates all latency arbitrage opportunities in the US market were worth $3 billion in 2014. These numbers should be taken with a grain of salt, but they provide a useful sense of the amount of money that can be siphoned off in the current system.
Budish supports IEX, but he also believes the underlying regulation should be revisited to fix the problems created by latency. “In [the] transition from humans to computers we made a mistake. There was a glitch that was introduced to the electronic market design,” he says. Last year he, along with two colleagues, published a paper proposing that glitch be solved by replacing instantaneous trades with a technique they call frequent batch auctions.
In a batched system, all trades that happened over, say, a millisecond, would be processed as though they occurred simultaneously. All buyers and sellers that beat the best price in that window of time would trade at that same price. This would almost completely eliminate the opportunity for latency arbitrage and make innovations like IEX’s speed bump unnecessary. Such a system would also have the advantage of being technically straightforward—many other computer systems work in similar scheduled time increments.
Batch auctions are a financially nuanced way of dealing with the latency issues that would not require reintegrating the markets. There are a variety of other possible solutions—computer scientists have been thinking about these sorts of problems for decades—though most would require rolling back at least some of the competitive mechanisms introduced with Reg NMS. For example, trades across the markets could be fully synchronized, such that a trade could not complete on one market without first verifying that it had the latest price data from all the others. This would slow down the markets, but would also guarantee price consistency. In another scenario, quotes and trades could be sent through a central routing point before being routed to and from the exchanges. This would be the simplest possible solution, though it would largely defeat the purpose of separating the exchanges in the first place.
The speed bump offered by IEX is in some ways a more complicated solution than any of these. It requires no modifications to the market structure, but that’s because it’s really just a Band-Aid. It only protects IEX customers and only from a fairly small subset of speed-related problems. If adopted without careful regulatory adaptation by the SEC, it may actually end up making the market as a whole more complicated, by fragmenting it into “speed-relevant” and “speed-irrelevant” zones.
Thanks to widespread media coverage, the IEX proposal has received hundreds of public comments letters. A surprisingly large number come from ordinary investors angry at the perceived rip-off posed by high-frequency trading. Other supporters include T. Rowe Price, Franklin Templeton, and several large pension funds. (The first two are also investors in IEX.)
It’s no surprise that institutional investors, such as pensions, would endorse IEX. More competition among trading venues is always good for them. They are also most vulnerable to the so-called “high-frequency tax.” When a large pension fund rebalances its portfolio, that has reverberations in the market that high-frequency firms detect and trade against. From a pension manager’s perspective, IEX offers an opportunity to prevent fast traders from skimming the hard-won earnings of their beneficiaries.
High-frequency traders, such as Citadel and Hudson River Trading, are among IEX’s harshest critics, but their critique pales when compared to that of the established public exchanges. NYSE and Nasdaq have taken every opportunity to heap opprobrium on IEX. Over the last decade those exchanges have cultivated a lucrative business selling speed-oriented services to traders. IEX represents not only new competition, but also an existential threat to their business model.
In March the SEC put forth a regulatory proposal that would make speed bumps legal for any public exchange—one possible path to approving IEX. To accomplish this, they would reinterpret Reg NMS to allow any intentional delay that is shorter than a thousandth of a second. The nature of this proposal is emblematic of how the market got so complicated in the first place. Not only does it avoid the larger structural questions, the choice of an arbitrary threshold will inevitably add complexity to the market by encouraging exchanges to adopt a wide variety of inconsistent implementations.
Nasdaq has threatened to sue the SEC if the proposed rule is adopted. “The Commission’s proposal to sanction intentional delays would rewrite Regulation NMS, not interpret it,” they wrote in a comment letter. And they aren’t the only ones opposed to such an arbitrary change. Even IEX has written in to oppose allowing exchanges to adopt speed bumps for any purpose. Not only would that undermine what is unique about IEX, it could also open the door for exchanges to sell access to slower trading—effectively monetizing both sides of the debate.
Despite the lack of clarity about how to proceed, SEC staff has recommended the committee approve IEX’s application. Whatever the committee decides on Friday, it will undoubtedly be interpreted as a signal of their willingness to regulate high-frequency traders. Unfortunately, as long as the structural issues of Reg NMS are relegated to the sidelines of the conversation, isolated decisions such as this one won’t accomplish much more than preserving a broken and incoherent system.
The image at the top of this post was shared under a Creative Commons license on Flickr. It has been cropped.