Revisiting the Flash Crash
The Joint CFTC-SEC Advisory Committee on Emerging Regulatory Issues has issued its findings regarding the Flash Crash and potential regulatory remedies to such events. Overall it is a solid effort that makes a serious attempt to grapple with the thorny issues arising out of the May 6 events, and changes in financial market structure more generally.
My primary criticism is that the report assigns far too much importance to the speed of modern, electronic markets dominated by high frequency traders that are the primary liquidity suppliers. This over-emphasis is pronounced in section I, titled “Volatility.”
The report confirms what has been pretty evident from the start, namely, that liquidity was eroding before May 6, and then virtually disappeared for a short period of time on that date. But it does not provide any reason to believe that this was something that was driven particularly by high speed electronic trading. The economics of market making have been the same for a long time under a variety of trading technologies. These economics are dominated by considerations related to information–and especially information asymmetry–and risk. These considerations are present in every trading and technology environment. There are, moreover, informational and risk conditions in which liquidity suppliers leave the market in droves. This happened in old-fashioned face-to-face markets: witness the fact that on 19 October, 1987 most futures market locals left the floor–or were pulled off by their clearing firms. The Flash Crash just demonstrates that these factors are relevant in electronic markets too.
By focusing on high speed computer trading, and suggesting that it is fundamentally different from liquidity supply in the old days, the report distracts attention from the more fundamental causes of crashes. Yes, HFT is new, but HFT market makers are responding to the same economic factors in similar ways to their old school floor trader counterparts. The main novelty of HFT is that high frequency traders trade across markets, rather than in one or two. The report would have done better to focus how to best mitigate the age old plagues of market making in an electronic environment.
The report discusses the advisability of imposing obligations on market makers, but wisely demurs from recommending this. It recognizes the difficulties, though it does not acknowledge the most important problem, namely, that obligations are costly and will affect adversely liquidity supply in non-crisis periods.
The report also correctly lauds the CME’s stop functionality, and recommends it be adopted in other markets as well. It makes constructive and reasonable recommendations about how this functionality can be tweaked to make it more robust. In brief, it recommends what could be described a “reload” feature that would kick in if the initial application of the stop functionality failed to stem a crash.
One particularly interesting part of the report relates to “Preferencing, Internalization, and Routing Protocols.” It is particularly interesting because it highlights an issue that I analyzed in my 2005 article “The Thirty Years War.” In that article, I noted that the SEC had deliberately chosen an “information and linkages” approach rather than a central limit order book (CLOB) approach to securities market structure. I further noted that the rules adopted by the SEC did not preserve time priority across trading venues, and did not protect all limit orders in all books against execution on another venue at an inferior price to those not at best of book. These were problematic features of the SEC RegNMS, and I was critical of both (especially the fact that only limit orders at the top of the book are protected).
The report argues that the lack of time priority inherent in the “no trade through” rule has encouraged internalization of orders, and that the lack of integration across books at prices away from the inside market can exacerbate price movements. It therefore recommends that the SEC study the costs and benefits of changing these aspects of RegNMS. Specifically, it recommends study of the imposition of a “trade at” rule which would effectively restore time priority.
I agree with the recommendation, but think that these issues are somewhat of a red herring with respect to the Flash Crash. This is especially true of the “trade through” rule. The report authors argue that the current rules favor internalization; internalizers supply liquidity; but internalizers stopped providing liquidity during the Flash Crash, leading to an additional flow of orders to public markets that stressed an already liquidity constrained situation. Imposing a “trade at” rule that would make it impossible for internalizers to match quotes standing in the public market would encourage additional liquidity supply in the public markets.
This is no doubt true, but the report doesn’t analyze the counterfactual, i.e., how things would have worked under the trade at rule. Yes, there likely would have been less internalization. Thus, in normal circumstances, more liquidity would be supplied via public quotes and less via internalization. But we saw during the Flash Crash that public market makers fled and as a result quotes disappeared. Internalizers provided less liquidity at the same time. But isn’t it highly likely that the liquidity suppliers that entered the public markets during “normal” times to replace internalizers would have also fled the market in conditions like those that provoked the Crash? I think so. The report doesn’t consider this counterfactual, so it cannot support a different conclusion.
That is, I think that it is accurate to say that a “trade at” rule would shift liquidity from internalizers to public quotes during normal times, but that this does not imply that liquidity would be higher overall (across all sources of liquidity supply) during stressed periods. The quotes that appear in public markets because quote matching becomes harder would be highly likely to disappear in Flash Crash-like conditions because it is particularly costly to provide firm quotes under those conditions–that’s why the quotes disappear. Thus, it is worthwhile to have a debate/study about the wisdom of the trade at rule, but that debate study should recognize that just because internalizers fled when the market went crazy doesn’t mean that the new public quoters attracted to the markets when internalizers are hobbled will stick around.
A look at the futures market bolsters this skepticism that internalization was a major contributor to the Crash itself. The index futures markets are not fragmented, and are not subject to internalization. The quote matching problem doesn’t exist there. At all. But it suffered serious liquidity withdrawals on 6 May. Thus, in my view, internalization is a red herring with respect to Flash Crashes. It is an important issue, but its relevance is primarily during “normal” times and that should be the focus of any study.
I was also intrigued by the report’s discussion of peak load pricing and the problems of system capacity constraints driven in part by high cancellation rates for HFTs. This subject was, in fact, the subject of the first substantive post on SWP.* There I puzzled as to why exchanges didn’t set traffic-sensitive price schedules for access to exchange matching computers. I noted that the lack of pricing of capacity could lead to overconsumption of computer resources, lags in getting orders executed, and even shutdowns of the trading system. Interesting to see that 5 years after I wrote that piece the issue is getting some attention. I’m still puzzling, and look forward to seeing how the exchanges and regulators deal with this issue. (The report mentions the difficulty of pricing cancellations in a fragmented market structure. This is another manifestation of the US equity market-centric mindset of the report’s authors. That isn’t an issue in futures markets, or foreign markets that are less fragmented. The puzzle is not explained by market fragmentation.)
In sum, the report moves the ball forward some. I think it gets sidetracked on some issues, but its focus on the need to encourage liquidity supply during stressed times is salutary. A lot of work remains to be done to see just how that can be accomplished in an efficient way, however.
And speaking selfishly, that’s a good thing. For the report confirms something I wrote in the conclusion of “The Thirty Years War”:
Therefore, the proposed rules [Reg NMS] are not the final battle in a ThirtyYears War. I fully expect that in 2075, some professor will write an article about the latest clash in an ongoing Hundred Years War over securities market structure regulation.
These market structure issues are thorny indeed, and will not be resolved any time soon. I am sure they will not be resolved by early-2013, when my next book, Financial Market Macrostructure: The Organization of Securities and Derivatives Markets should appear. Which should be good for sales 🙂
*The post was titled “VOLT”, a neoacronym (which might be a neologism) for “Value of Lost Trade.” The idea was that in figuring out how much generating capacity is needed in an electricity market, it is conventional to assign a “value of lost load” (VOLL) to the consumption of power lost when system capacity is reached. Similarly, when deciding how much trading capacity to build, it would be advisable to determine the value of a trade lost or delayed due to a shortage of system capacity.