U.S. Securities & Exchange Commission
SEC Seal
Home | Previous Page
U.S. Securities and Exchange Commission

Speech by SEC Staff:
Market Participants and the May 6 Flash Crash


Gregg E. Berman

Senior Advisor to the Director, Division of Trading and Markets
U.S. Securities and Exchange Commission

11th Annual SIFMA Market Structure Conference
New York, New York
October 13, 2010

Good morning. It is a pleasure for me to be here today — though I've been in field of finance for almost 20 years, I've only recently joined the regulatory community and very much appreciate SIFMA inviting me to speak with you about a topic that has received such wide-spread attention, namely the events of May 6.

Since the Report of the Joint Staffs of the SEC and CFTC was released a few weeks ago, an enormous amount has been written on the subject, on the report itself, and on potential next steps. My goal today is not to rehash what has already been written. Rather, I thought you might be interested to hear how the analysis was performed, how we actually know what we say we know, and what the process itself may suggest for future policy.

If our report was made into a movie, then my remarks this morning should be taken as the directors-cut DVD, with bonus tracks. As such, these remarks are indeed my own, and do not necessarily represent the views or opinions of the Commission, any individual Commissioner, or my colleagues on the Commission staff.

My story begins with the release of our Preliminary Report on May 18th, a mere 12 days after the flash crash. As a preliminary report it contained facts and lines of further inquiry, but no real explanations.

At that time, what we knew for sure was that at about 2:40 in the afternoon of May 6, prices for both the E-Mini S&P 500 futures contract, and the SPY S&P 500 exchange traded fund, suddenly plunged 5% in just 5 minutes. More so, during the next 10 minutes they recovered from these losses. And it was during this recovery period that the prices of hundreds of individual equities and exchange traded funds plummeted to ridiculous levels of a penny or less before they too rebounded. By the end of the day everything was back to "normal," and thus the event was dubbed the May 6 Flash Crash.

As part of our preliminary analysis, the CFTC had found that a single firm, a large fundamental trader, had sold 75,000 E-Mini contracts between 2:32 and 2:51. But based on an examination of just the price and volume patterns for the E-Mini, it would have been premature to conclude that this trade played a leading role in the main event. After all, half of the contracts had been executed after 2:45 while the E-Mini was on the mend.

We also knew that trading on the New York Stock Exchange in a significant number of NYSE-listed stocks was intermittently "halted" around the time of the flash crash by NYSE's use of "Liquidity Replenishment Points" that effectively band stocks to trade only within certain price limits. And finally, we knew that exchange traded funds suffered disproportionately during the flash crash. That was notable since most ETFs are listed the NYSE's Arca exchange, and on the afternoon in question, other exchanges reported that they declared self-help on Arca and stopped routing orders to that exchange.

It didn't require a huge leap of logic to jump to the conclusion that these issues were all somehow related, suggesting this was yet another case of a "confluence of events." And so it was with these notions that we began our work.

The investigation itself comprised two prongs. The first prong was all about collecting and analyzing market data. We knew that to really understand how and why prices fell so fast we needed to examine underlying order books and trading patterns. The CFTC took the lead on the E-Mini and we at the SEC took the lead on the equity markets.

Because the E-Mini is traded on a single market, the CFTC was able to assemble all the relevant data in just a few short weeks. And here is what they found: Buy-side and sell-side depth in the E-Mini market within about 10 basis points of the current price was stable throughout the entire morning and most of the afternoon. This "near-inside" liquidity data would have been observable by market participants trading the E-Mini, and perhaps its stability is what led to spreads in the E-Mini remaining very tight even when prices rapidly declined. However, the CFTC was also able to analyze the full E-Mini order book which allowed us to track the evolution of total market depth, not just near-inside market depth. And this data revealed a completely different dynamic.

At the beginning of the day total buy-side market depth in the E-Mini started at about 100,000 contracts. This depth trended down throughout the day, and by mid-afternoon buy-side interest was less than 50,000 contracts. At about 2:30 buy-side market depth started to erode even faster and by 2:40 there were only 15,000 contracts left. Then, over the next four or so minutes, buy-side market depth in the E-Mini virtually vanished, falling to about 1,000 contracts, or less than 1% of its value of that morning. It was at that time that the E-Mini and SPY both plunged over 5%. It was apparent from the order book data that the E-Mini had suffered some sort of a liquidity crisis, but exactly how it happened was yet unknown.

This analysis led to an obvious next question — did the same thing happen in the equity markets? Did a similar liquidity crisis perhaps precede or even trigger the one in the E-Mini? Unfortunately, assembling comparable order book data for equities and ETFs proved to be a much more difficult process. Every exchange has its own formats, conventions, and databases, and we found we needed to deploy a dedicated team focusing directly on this problem for three months. Though collection started in June, it would not be until the middle of August that all the required data was aggregated into a single database for analysis.

Ideally, we would have liked to see every order, on every venue (whether public, OTC, or dark), who submitted it, how it was routed, how it was executed or canceled, and what the market conditions were at the time of execution. At present, this type of market structure data does not exist in a comprehensive form. Even the seemingly simple task of lining up a quote with a trade price proved to be very difficult.

When we requested data from the exchanges and other participants we never did so blindly — we always explained to those providing the data exactly what we were trying to do. And those in the know cautioned us that trying to reconstruct the world of trading down to the millisecond is just not possible. There are too many inherent lags and uncertainties. Fortunately, we did not require such precision for our analysis. In general we aggregated information on a minute-by-minute basis. And though we therefore may have missed fleeting intra-minute events, what we saw on a minute-by-minute basis was more than sufficient to provide a robust analysis of liquidity.

So, for the purposes of our report we had what we needed. And we now know that there was indeed a liquidity crisis in the equity markets, but it followed the one in the E-Mini, and did not precede the initial rapid decline in prices. This then led to two further questions: First, what were the conditions that caused the liquidity crisis in the E-Mini prior to 2:45; and second, why did the equity markets suffer a similar crisis after 2:45?

But before answering these questions I'd like to go on a little sidebar and add a third set of questions to the mix that may have some implications for policy. Why did it take us three months to complete the data-driven portion of our analysis of the equities markets, and why should this matter?

Many market participants currently trade at the millisecond level, but it seems that reconstructing events at that level is fraught with uncertainty. While it is true that market makers and high frequency traders track the equity markets in real time, this comes at significant cost, requires a high-degree of skill to deal with overwhelming complexity, and most of these firms only track the markets from their own touch-points as they interact with exchanges and other market venues. The very process of collecting data for our analysis leads me to wonder-has the structure of our equity markets become so complex, and has trading become so fast, that only the most sophisticated players can effectively monitor and manage their interaction with the markets in a robust and holistic fashion? And what does this mean for the future of a market place in which the quest for speed continues unabated? I'd like to keep these questions in mind as we continue with the main story.

The second prong of our investigation involved gathering whatever information we could on the specific actions market participants undertook on May 6, and the reasoning behind those actions.

Throughout June, July, and the first half of August, we had teams of 20 or more people from across the SEC's various divisions conduct extensive multi-hour interviews with registered equity market makers, options market makers, high-frequency traders, over-the-counter internalizers, brokers, asset managers, proprietary funds, and of the course the exchanges themselves. At the peak of the process we were holding three or four of these interviews per day.

We began each interview by asking participants to describe their business model and how they normally trade. Then we asked them to detail their actions throughout the afternoon of May 6. When we started the process in June, our expectations were that we would hear a lot about disparate trading conventions, LRPs, and self-help. But this was not to be the case.

Instead, market participants had told us that they pulled back from the markets, and stopped providing liquidity, because the markets were moving too fast, they were hitting position limits, risk limits, P&L limits, and system limits. They weren't sure if the prices their automated systems tracked were indeed moving that fast, or if these observations were just an artifact of some data problem. If it was a data problem they couldn't continue trading, and if it wasn't a data problem they didn't necessarily want to continue trading. It's hard to convey in words some of the emotion and concern that was expressed during those interviews. Participants told us they didn't know what was going on, and decided it would be more prudent to reduce or altogether stop their trading activity than to risk trading during such extreme conditions.

Within the first few weeks we had spoken with those participants who, in one form or another, were party to many of the trades that had been executed at absurd prices. This was not hearsay reporting. These were the participants who actually withdrew liquidity. We asked whether self-help issues caused them problems. In general, the answer was "no" since these same participants use their own routing engines to send orders directly to each public market. How about LRPs? Yes, they were concerned that the number of LRPs being triggered was indicative of very volatile market conditions, but no, LRPs did not cause any routing issues nor inhibit their ability to trade. How about data delays on the consolidated public feeds, an issue that was brought into the limelight repeatedly throughout this summer? Definitely noticed by some, but not really an issue for trading individual securities since most of the participants who provide liquidity to the markets use proprietary feeds coming from the exchanges, and not the consolidated feeds.

In general, many liquidity providers pulled back because they weren't confident about what they were seeing, and they weren't confident about trading under such extreme conditions. I find this quite telling.

Consider that at its low point, Proctor & Gamble traded at an intraday loss of 36%. And during the minute in which this low point was hit almost 500,000 shares were executed. However, in the minute before, Proctor & Gamble was only down 23%. I can understand how P&G might jump to a loss of 36% in one fell swoop, perhaps due to sudden news or a technical glitch. But I don't understand how it can fall 36% over the course of a few minutes. Collectively, across the entire market, participants could not find a mere $25MM to buy P&G at a 23% discount? That stock rose more than 25% in the next 5 minutes. For a market that spends so much time and effort on trading within spreads of a few pennies, it seems like this was the deal of the century. Yet it took another 13% loss before P&G was able to recover. So I guess I agree with what most market participants told us about May 6. They really didn't know what was happening, or at least they didn't feel confident enough to buy P&G at a 23% discount.

I'd like to now repeat the question I posed before. Has the structure of our equity markets become so complex, and has trading become so fast, that only the most sophisticated players can effectively monitor and manage their interaction with markets in a robust and holistic fashion? Apparently, for many participants, the answer may be yes.

The good news is that these issues can be addressed, and indeed some of the fixes are already underway.

One of the key lessons we learned from the flash crash is that many market participants have their own versions of trading pauses that kick in during extreme events. Unfortunately these pauses don't all occur at the same time or in the same manner. If everyone in a stadium stands up together, then everyone can see the field. And if everyone sits down together, then everyone can still see the field. But if half the people are standing and half the people are sitting you have…a mess. This is what effectively happened on the afternoon of May 6.

To treat this symptom, a few months ago the SEC approved a pilot program by the exchanges and FINRA to pause trading in an individual security when its price moves "too much, too fast," which is currently defined to be 10% or more in 5 minutes or less. Initially this "circuit-breaker" mechanism covered only stocks in the S&P 500 but recently the program has been expanded to include all stocks in the Russell 1000 as well as a number of ETFs that track those stocks.

As a pilot, this program proved successful. Over the last few months circuit-breakers were triggered in about a dozen stocks, providing us a great opportunity to study the mechanism and its effects. We observed when and why individual circuit-breakers were triggered and have a very good sense of what needs to be improved. Across the industry there seems to be a general agreement that some form of a limit-up/limit-down structure would best balance the desire to pause trading in an individual security when prices move too fast, but allow prices to recover on their own if possible. This same structure has some very appealing side-benefits — by banding prices within pre-defined limits the probability of a clearly-erroneous trade is significantly reduced. Price-banding also eliminates the need for, and use of, stub quotes.

Though promising, limit-up/limit-down treats only the symptom, and not necessarily the cause of erratic price moves. We know from the flash crash that some of these causes are related to orders that overwhelm existing liquidity, and some are related to orders that are placed when liquidity is being withdrawn. But before I get to those points I'd like to come back to idea of market participants having more confidence in the infrastructure of our markets so they feel more comfortable participating even under extreme conditions.

How can this issue be addressed? I think my take on this might surprise you.

At the end of May, the SEC released a proposal for the creation of a Consolidated Audit Trail. Allow me to quote from the release "A consolidated audit trail would significantly aid in SRO efforts to detect and deter fraudulent and manipulative acts and practices in the marketplace, and generally to regulate their markets and members. In addition, such an audit trail would benefit the Commission in its market analysis efforts, such as investigating and preparing market reconstructions and understanding causes of unusual market activity." If the events of May 6 don't qualify for unusual market activity, I'm not sure what does. If a consolidated audit trail had been in place prior to May 6, I could have delivered this speech a few months ago.

It should be obvious how a consolidated audit trail will help regulators, but so what. What's in it for you, and for the markets in general?

This is the point in my remarks that I could really use a soap box.

I believe that a market structure that can support the requirements of a consolidated audit trail will necessarily be more robust and provide participants with more confidence, even during extreme events. Furthermore, I maintain that the process of implementing such a system will in itself lead to direct and indirect improvements as existing systems are revisited, order types are re-examined, and the industry as a whole confronts some of the issues that make a consolidated audit trail necessary.

Why am I confident this will occur? Because for the past decade I have been involved in the design and implementation of risk management frameworks for many hundreds of the world's largest and most sophisticated banks, brokerages, hedge funds, and asset managers. Rolling out a risk management framework is hugely challenging because it requires the collection and tracking of data across a wide variety of disparate systems. And in nearly every case, the risk managers and technology officers who were ultimately responsible for the system reported that the knock-on effects of implementing such a framework led to intangibles such as better confidence in their position data, an ability to track holdings over time, and methods for reviewing past actions. More so, firms that had the hardest time implementing such a framework were those that needed it the most, and those that ultimately benefited the most from intangibles that were unrelated to the end-use of the system itself. I appreciate that many of you will be skeptical of this line of reasoning, but ask around — I think you will be pleasantly surprised at what you find. And I think the market place as a whole will greatly benefit from initiatives like the consolidated audit trail.

With that said, I will now get off my soap box so we can finish the story.

We knew that the E-Mini suffered a liquidity crisis and that this translated to rapidly declining prices in both the futures and the equities markets. We also knew that market participants on the equity side created their own liquidity crisis by pulling back for all of the reasons discussed, and with lots of implications for future policy related to the obligations of liquidity providers. But what caused the initial liquidity crisis in the E-Mini?

We all know that the markets are highly inter-related and trading is replete with many types of feedback mechanisms where the actions of one participant trigger those of another. These mechanisms can be formal (such as stop-loss orders) or informal (such as momentum trading).

In a terrific piece of work that was just published, the CFTC analyzed trading in the E-Mini market on a second-by-second basis to determine what affect an order to sell 75,000 contracts had — not on market prices — but on market participants. Over the last week a lot has been written about this large order, but some articles may have missed the point. We already knew that individual component executions of the large sell order did not directly coincide with all of the moves in the price of the E-Mini. If they had, our job would have been a lot easier.

But markets are more complicated than that. And what the CFTC analysis showed is that the large sell order placed at 2:32 dynamically changed the behavior of other market participants. The impact of this sell order was not immediate but took a few minutes to propagate through the E-Mini market, much in the same way that resulting price declines took a few minutes to propagate through to the equity markets. In the equity markets we saw how market participants reacted to fast-moving prices. In the futures market we can see how the algorithmic strategies of high frequency traders and other market intermediaries reacted to this large sell order. The net result was that, in the absence of price or time throttles on the sell algorithm, the order was executed much faster than might have been anticipated, and fundamental buy-side interest was not able to keep up.

Perhaps even more importantly, the analysis demonstrates that volume and spreads are not necessarily good indicators of liquidity in all cases. Since the SEC issued a concept release on market structure at the beginning of the year, there has been a very lively debate in the industry, the public, and the regulatory community, regarding the role of high frequency trading and liquidity. In fact, last Sunday night 60 Minutes aired a piece on just this topic.

Our report was not intended to address the general topic of high-frequency trading. But I do think it sheds some light on an issue that is often characterized by somewhat extreme views. On the one hand, we have those who believe that high-frequency traders directly caused the flash crash by forcing down the market and creating an unstoppable wave of selling, perhaps through something called quote stuffing. Our analysis of May 6 does not support this view. At the other extreme, we have those who believe high frequency traders provide tremendous liquidity to the markets and act as a stabilizing force for prices. This too, is not supported by our analysis of the flash crash.

What we did find was that on May 6, a group of 17 HFTs accounted for between 40% and 50% of total dollar volume, which is roughly in line with the trading patterns of these firms in the days preceding, and the days after, the flash crash. However, we also know that about half of all these HFT orders on May 6 were executed passively, whereas the other half were executed aggressively. It would therefore seem, at least on May 6, that HFTs took as much liquidity from the equity markets as they provided. Furthermore, data show that in the 15 minutes leading up the flash crash at 2:45, HFT activity increased proportionately with the market, as HFT algorithms transitioned from being aggressively neutral to being net aggressive sellers. This transition was roughly in line with the market as a whole, though HFT activity after 2:45 dropped off considerably. So while it does not seem that HFTs directly caused a wave of selling, HFTs did ride that wave down as prices declined.

In conclusion, what I take away from the events of May 6 is that the markets are perhaps even more complicated than we might have imagined. Interactions between market participants can be subtle, and the actions by some can cause unexpected reactions in others. Along the way we collected an enormous amount of data, and performed many analyses, that will hopefully aid and inform the continued dialogue on market structure. And though the afternoon of May 6 was quite harrowing, I am confident that the lessons of that day will ultimately lead to lasting improvements that will benefit all market participants.


Modified: 10/22/2010