In the aftermath of Knight’s HAL moment, Larry Tab wrote a very good and succinct piece on the implications of fragmentation in equity markets:
Most think about fragmentation in terms of trading venues. And yes, the U.S. equity market has a number of trading venues. There are 13 exchanges, about 50 dark pools and many internalizing brokers that can match buyers and sellers. To ensure the best match, sophisticated order routers need to examine many trading venues. And without being in two places at once, the more venues checked, the greater the chance buyers and sellers miss each other.
While checking many venues complicates execution, markets also can fragment in other ways, including price and time. The more price points, the more missed trades. This is especially true as liquidity providers spread orders across more price points and venues. As fractions turn to pennies, and pennies to sub-pennies, through dark pools and the NYSE Retail Liquidity Program (RLP), trading volume becomes fragmented from 16 price points per dollar to 100, or 1,000.
The same problem exists with time. The quicker the quote timeframe, the more messages I need to produce and analyze. If I am quoting across 13 exchanges and 20 dark pools, then I need to manage my quotes across 33 venues. If my quoting frequency changes from seconds to tenths to tens of milliseconds to milliseconds, then my quoting rate increases from 33 to 330, 3,300, 33,000 quotes per second. Multiply that across 5,000 stocks trading at pennies or possibly sub-pennies and think about the massive technology and message infrastructure needed to trade.
Moreover, this dense web of linkages between venues raises the potential for algo errors. More trading venues means more changes in the way the venues operate and their technologies, which means more changes in algos, which raises the odds of a programming glitch. For instance, the catalyst for the Knight implosion was the NYSE’s introduction of its RLP program. Another example that I just read (but can’t find a link to right now) is an HFT firm that had an algo go haywire because it hadn’t been informed of a change in the software at one of the trading venues it connected to: fortunately, this firm had measures in place that turned off the program after it had lost only $5K. There is likely an externality across venues: a change by venue A increases the odds that the change will trigger a perverse algo response that will have adverse effects on venues B, C, D, etc.
So fragmentation has costs. This raises the questions: why does fragmentation exist?, and, do the benefits exceed the costs?
It should first be noted that equity markets have always been fragmented to some degree. Back in the day, substantial volumes of NYSE stocks were executed by “Third Market” dealers (notably, Bernie Madoff!) and in block trades negotiated away from the floor.
This fragmentation was driven by informational considerations. (Relatively) uninformed order flow that could signal ignorance by some means (order size in the case of retail trades executed by Third Market dealers, or reputation and non-anonymous dealing in the case of blocks) reduced trading costs by trading in off-exchange venues. If they traded in the anonymous NYSE marketplace, they could not be distinguished from informed order flow, and paid execution costs that reflected the adverse selection risks faced by liquidity suppliers in such a marketplace: by trading in another venue that had methods of screening out informed traders (imperfectly, but to some extent), the uninformed avoided paying this adverse selection premium.
This type of fragmentation exists today: Dark Pools have evolved to perform the same functions as Third Markets and block markets. Dark pools use a variety of methods to attempt to screen out informed and opportunistic traders. And some HFT/algo strategies (arguably opportunistic) are attempts to circumvent these screening technologies.
The proliferation of non-Dark Pool trading venues-anonymous, non-screening trading platforms-is relatively novel. Until the mid-00s, such venues did not exist. Virtually all informed order flow, and a good chunk of uninformed order flow, in NYSE stocks was executed on the NYSE. The exchange had a market share of 85+ percent up through around 2005.
Since then, NYSE market share has plunged to around 30 percent. Some of that volume has been snagged by Dark Pools, but the bulk has been taken by other exchanges.
This reflects an explicit policy choice by Congress, and the SEC. Since the mid-1970s, Congress has pushed for the creation of a National Market System of competing exchanges. That effort was moribund until 2005, when the SEC implemented RegNMS that obligated electronic markets to direct orders to other markets displaying better prices.
This effectively socialized order flow. In the pre-RegNMS days, market participants had an incentive to direct their orders to the market that they expected to have the best price. This created a self-reinforcing feedback loop. The biggest market (the NYSE) tended to have the best price, so traders sent their orders there, thereby reinforcing the NYSE’s advantage. With the mandated routing of orders to other markets, the NYSE no longer had a lock on order flow. Other venues could compete for it effectively.
This had already been seen in options, which had been subject to order routing mandates much earlier.
From the SEC’s (and Congress’s) perspective, this was a feature, not a bug. The whole idea behind NMS was to create a more competitive marketplace not dominated by a single exchange.
It worked, but it has had unintended consequences. The adverse consequences of fragmentation Larry Tabb discusses are among them.
This points out a fundamental dilemma in securities market design. If order flow is not socialized by order routing mandates, the liquidity network effect tends to lead to the dominance of a single exchange that exercises market power. We see this in futures around the world. Although electronic trading has taken over futures trading, we don’t see fragmentation like in the equity markets, which means that fragmentation is not an inevitable consequence of the computerization of trading. Part of that is due to the fact that information asymmetries are less severe in futures than individual equities (e.g., because there is less private information about an equity index than an individual stock), and so there is no “dark pool” type fragmentation. But most of it is due to the fact that there is no socialization of order flow in futures. Perhaps because it was, historically, a more obscure corner of the financial world than the stock market, Congress never caught the itch to create a National Futures Market System. But regardless of the reason, the order flow feedback effect, where order flow attracts order flow, has led to the dominance of a single exchange in virtually every major futures contract.
This points out a fundamental dilemma. Due to the nature of liquidity, policy makers have to pick their poison. They can eliminate fragmentation, at the cost of market power. They can eliminate market power, but only by living with the adverse consequences of fragmentation.
Those are the choices. Period. The economics of liquidity can’t be legislated or regulated away. All that policy makers can do is affect how those economics will manifest themselves. Don’t like exchange monopolies (or near monopolies)? Socialize order flow a la RegNMS-and live with the consequences of fragmentation like we see today. Don’t like the consequences of fragmentation? Abstain from requiring exchanges to direct order flows to markets displaying better prices-and live with the market power that will result when trading tips to a single venue of price discovery.
Not an easy choice. I don’t know what the more efficient alternative is. But it would be refreshing if debates over market structure would acknowledge that there are no Goldilocks, “just right”, choices. Instead, there is more of a lady and the tiger like dilemma. Bewail the evil consequences of fragmentation if you will, but be prepared to show how those consequences are less malign than the costs of market power. And vice versa.
Leave a Reply