I just read Haim Bodek’s “The Problem of HFT.” It reinforced my belief in what I’d long concluded, namely, that the problematic aspects of HFT result from the decision of the SEC to create a Simulacrum Limit Order Book-a SLOB-rather than a true Central Limit Order Book-a CLOB-when it promulgated RegNMS. As I noted at the time, the SEC chose an “information and linkages” approach to create a National Market System that consisted of interconnected competing exchanges, rather than to create a single CLOB. In achieving competition, it was hugely successful. Whereas NYSE dominated trading in its listings pre-RegNMS (with a market share in the low-to-mid 80s), now executions are divided rather equally among a handful of exchanges. There are few rents in securities execution in the US anymore, in contrast to the old floor days. NYSE’s equity business contributed relatively little to the price paid by ICE for NYSEuronextLiffe.
Part of the decision to was political economy: politically, the SEC couldn’t have realistically favored one venue, or displaced existing exchanges by creating a new national CLOB. Part of this decision was predicated on a defensible and understandable desire to encourage competition in securities trading, in fees, rules, liquidity, and technology. However, the SEC didn’t anticipate how the market structure would evolve, and in particular, how HFT would transform securities trading. I can’t say that anybody did. I certainly didn’t. I thought that automated trading would link markets, but I didn’t see it playing out like it did.
I call what the SEC wrought a “simulacrum” order book because although it incorporates some aspects of a central limit order book, specifically price priority across exchanges for the top of book, it lacks several crucial features of a true CLOB. Price priority is for top of book only. Time priority (or other secondary priorities) are not enforced across trading venues. Markets can cross or lock-which can’t happen in a CLOB, where locked or crossed orders execute automatically. Moreover, latencies, particularly in the public feed of the NBBO create behaviors and incentives that diverge from a CLOB.
Because it links multiple execution venues, the SLOB has many seams. And one thing we know is that traders traders look for and exploit seams like Tom Brady looks for and exploits the seams in a zone defense. Bodek’s little collection of essays goes into considerable detail on how traders have worked the seams, and how their methods have changes in the last 6 years.
The most critical seam in the SLOB, in Bodek’s accounting, is the rule on locked and crossed markets. The SEC rules state that an order on exchange A that locks or crosses an order on exchange B cannot be displayed. This complicated the task of quoting aggressively: an HFT that submitted an order that locked or crossed a market would not be able to establish position at the head of the queue. To address this, traders first used spam and cancel: repeatedly sending orders that were canceled if they locked or crossed, but which would go to the front of the queue as soon as the quote in the away market changed. Exchanges had implemented algorithms that re-priced orders that locked or crossed away markets, and put them at the end of the price queue. HFTs circumvented this using hidden order types that were not displayed if they locked or crossed, but which were displayed-and crucially went to the head of the queue at the new price-the instant their price would no longer lock or cross an away market. HFTs also used sweep orders to establish priority.
The common features behind all of the things Bodek criticizes are that (a) they are all intended to establish priority, and facilitate aggressive quoting, and (b) they can happen in a SLOB, but not a CLOB. They give an advantage to HFT firms that invest in the technology and in understanding the order types, to the detriment of those who don’t (including many institutional investors): this is Bodek’s main beef. They are an artifact of the prohibition on locks and crosses, and this can only occur in a fragmented market structure that precludes locks and crosses. If locking and crossing orders were matched against one another and executed, as would happen in a true CLOB, much of the gaming of order types Bodek laments would go away: I suspect, however, that latencies across a system of linked exchanges would make this impractical.
This relates to, and complicates, points I made in my writing on exchange competition in the late-90s and early-00s. Competition between privately operating CLOBs with no obligation to route orders to markets displaying better prices is likely to result in the survival of a single dominant platform that exercises market power. This is arguably the case in futures markets. A CLOB utility that is regulated to mitigate this market power problem raises myriad difficulties. Who will operate it? How will rates be regulated? How will access be regulated? (One way to exercise market power is to limit access to the CLOB.) How will the CLOB utility be governed? How is capture (by major intermediaries, for instance) to be avoided? How can the CLOB be incentivized to be technologically innovative, rather than a lazy monopolist (especially since some technological changes may be very disruptive to the business models of important market participants)?
Socializing order flows by requiring exchanges to route orders to other markets displaying better prices encourages competition, and seemingly avoids the problems posed by private CLOBs and utility CLOBs, but raises its own difficulties. Indeed, these are the problems that we are experiencing today, and which Bodek identifies. This raises the question: can we clean up SLOBs so they are more truly CLOB-like, while retaining interexchange competition that mitigates market power problems? The answer to this question has both technological and political components. A system of linked exchanges necessarily has seams and latencies and inter-market coordination issues absent in a true CLOB, and these are vulnerable to gaming and breakdown, and even during normal operation represent a more complicated and error-prone way of allowing all orders to interact with one another than a true CLOB. The political components arise from the fact that interested parties, most notably exchanges, major sell side intermediaries, and big buy side institutions, will strive to shape the system’s design and rules to advance their interests.
The trade off is therefore pretty stark: it’s a pick-your-poison kind of thing. Don’t socialize order flow and live with market power, or socialize order flow to facilitate competition, and grapple with the technological challenges of creating a virtual CLOB, and fighting the never ending battle against attempts to exploit the seams. I don’t claim to know the right choice, but I think that framing the issues in this way will permit far more constructive debate than the one currently ongoing. This debate has become Manichean: good vs. evil: black hats vs. white hats. It is oddly ahistorical, in its failure to recognize the problems with the previous system.
Perfect is not an option here. Which is why debates over securities and derivatives market structure have been raging for as long as securities and futures have been traded. The debate would be more productive-less sterile-if the choices were framed more clearly. I’ll keep trying.