The Gateway to Algorithmic and Automated Trading

Flash Crash - The Hunt for Weapons of Market Destruction

Bob Giffords scours September's CFTC/SEC "Flash Crash" report for the regulators' conclusions, lessons and remedies

After nearly five months and up to ten terabytes worth of analysis later, the US financial regulators, the SEC and CFTC, have come up with their findings of what happened on 6 May 2010, when the US financial markets suffered their worst shock since the great crash of '29. On the last day of September they published their long awaited report. However, those who hoped for closure will be disappointed, because their conclusions and recommendations are apparently still to come and more investigations around market data issues continue. Moreover, the focus was on 'how' the crash unfolded rather than 'why', although some 'explanations' were suggested.

They found a smoking gun of sorts: the large $4.1 billion futures trade that algorithmically sold 75,000 of the June e-mini S&P 500 contracts in 20 minutes on the CME. Within minutes of this 'fundamental' trade kicking off at 2.32 pm Eastern the Report describes how the various indices had plummeted to their daily lows, some major blue chips traded for a penny a share and many participants had withdrawn wounded or frightened from the fray. Post hoc ergo propter hoc is the regulators' refrain. If it followed, it must have been due to the event that preceded it. The Report helpfully explains how the collapse in liquidity spread from the futures market, first to the major ETFs, and then to the stocks underlying those ETFs in successive waves and in that order. Over 20,000 trades had subsequently to be broken by the market authorities as being clearly irregular.

Unfortunately the regulators had to admit that the futures trade in question was large but legitimate, and performed in the normal course of business as a hedge to an existing equity position. Indeed it was only "the largest net change in daily position of any trader in the E-Mini since the beginning of the year" [my emphasis], given that the fundamental trader in question had executed two larger programmes in the past year without problem. That does not really sound so earth-shattering.

Therefore, to stir up our outrage the regulators spend some considerable time impugning the execution strategy used by the mutual fund involved, if not the trade itself. An in-line 9% participation sell algorithm was apparently deployed that only looked at volume, but not at time or price. This resulted in the rapid unwinding of the trade because of the volume spike that then occurred. Next, the authorities note that on the earlier occasions the same fundamental seller had used a mix of human traders and a more cautious algorithm driven apparently by volume, time and price that consequently took over 5 hours to trade a similar size. How dare they throw caution to the winds, the regulators seem to ask and plough on regardless to sell their contracts in 20 minutes. However, the so-called fundamental seller of the e-mini did not determine the timing, but only the participation rate as so many do. The rest of the Report then relates how the markets reacted to these events and documents the subsequent collapse in liquidity in some gaudy if not gory graphics. They confirm, as we knew, that as markets disappeared into the abyss, spreads widened and liquidity disappeared, particularly on the buy side as floods of sell orders rained down on the markets. They also confirmed but did not comment on the fact that the markets were already starting to recover when the large trade completed. So the real issue is what turned the markets, not why they fell. This seems of less concern to the Regulators.

The Report is fairly obsessed with liquidity and its vanishing act. Perhaps someone should have reminded the regulators of the dry observation by Michael Milken, the 1980s junk bond king: "Liquidity is an illusion. It is always there when you don't need it, and rarely there when you do."

They usefully reminded us of the factors that drove participants to withdraw their liquidity, namely:

  • Price driven integrity pauses
  • Data feed driven integrity pauses
  • Internal system technical capacity issues
  • Fears over cancelled trades
  • Internal risk limits

The first two issues made traders question the validity of their instruments and therefore pause to verify. The Report notes that some returned to the market in seconds while others took much longer or never returned that session. Again, no surprises here.

The regulators documented in particular how during the critical period most of the 'liquidity' involved high frequency traders, only 16 of them, who at first bought and then subsequently joined the fundamental futures seller to square their positions. This progressively eroded the so-called fundamental buy side demand in the market. The Report calls this 'hot potato' selling, but is that not what market makers do? They smooth out the time difference of demand between fundamental sellers and buyers.

The detail is riveting but of course irrelevant to the real issues. Nevertheless it is fascinating to read that at the height of the crash in 14 seconds 16 high frequency jockeys traded 27,000 e-mini contracts at the gallop representing 49% of the market. Oscar winning performances must, however, go to the "one large internalizer (as a seller) and one large market maker (as a buyer) [who] were party to over 50% of the share volume of broken trades, and for more than half of this volume they were counterparties to each other (i.e., 25% of the broken trade share volume was between this particular seller and buyer)." Now that is impressive and rather imprudent of the internalizer.

One curiosity of the preliminary report is however finally explained. During the critical 10 minutes of the flash crash trough, the overwhelming majority of trades were flagged as short sells (70 to 90%). Why would anyone short sell for derisory bids of less than $1 dollar? The Report confirms, "In the case of an internalizing broker-dealer that is facilitating a customer sell order where the customer is net long and the broker-dealer is net short, but is effecting the sale as principal or riskless principal, the broker-dealer must mark the principal leg of the transaction as short." Elsewhere the regulators confirm that these internalizing broker-dealers had essentially ceased to take the risk themselves and were routing all retail and other sell orders to market. Moreover, they were converting market orders to limit orders, but then chasing the prices down until they traded. 90% of broken trades were apparently limit orders.

The prevalence of penny stub quotes is also explained. Some markets will automatically post such derisory bids and similar maximum offers at ludicrous prices when their market makers withdraw in order to fulfill the latter's obligations to quote. Such stub quotes are automatically renewed when they are hit. In other cases the market makers posted their own stub quotes.

So the market microstructure was specifically engineered to generate broken trades. It was a feature, not a failure of the market! And it would appear one internalizing broker-dealer fell straight into the trap. No wonder the authorities acted quickly to bust the trades.

Despite this enlightening entertainment, the events of 6 May were very serious indeed. It is a pity therefore that the regulators did not spend more time looking at how the markets and the various rules either encouraged or limited the damage, rather than exploring the celebrated vagaries of liquidity. For example, they did not assess whether best execution and the trade through rules worked during the crisis.

Indeed regulators spent some time defending the 'system'. For example, they argued that failures in prompt market data feed distribution in general and the consolidated quote and tape systems in particular, the impact of the self-help declarations of Nasdaq and BATS against NYSE Arca, or the use of liquidity replenishment points by NYSE were not material factors in the collapse of prices. This conclusion was based on the proportion of trades on NYSE and NYSE Arca that held up during the affected periods. However, this does not really prove anything, since it was likely that in a panic many traders would in any case revert to the home exchange. They do admit that these factors all worked to disquiet the market and motivate some participants' decisions to withdraw liquidity, thereby admitting there could have been an indirect effect.

That is all the report essentially says. Clearly the regulators spent a lot of time and effort documenting the disappearing liquidity on 6 May. They gave us an excellent description of how the markets collapsed and then recovered, but no real sense of why. They seem to suggest that if the one large, fundamental futures trade had not happened, or had been programmed to trade more slowly, it all might have been different. We shall never know. They also admit that nothing they have done so far will truly fix the problems in future although they hope that their circuit breakers will limit the damage next time.

The regulators appear to understand rather better the dilemmas faced by all market participants, including high frequency traders, as events unfolded, and they have explained in some detail how various aspects of the market worked in practice, and the ETF market in particular.

However, they failed to assess many, curious behaviors on the day. For example, the Report does not address or even mention the impact of the following:

  • Gyrating FX prices and the dynamics of the VIX and other fear gauges during the crisis and how they added to the tension
  • The large options purchase on the S&P500 index at 2.15 pm Eastern by an unnamed hedge fund potentially paying out another $4 billion if the index hit 800 by end June. As the markets fell the counterparties were busy laying off their risk. This might not have been classified as fundamental selling by the counterparties but was a directional move triggered by the hedge fund, not the highlighted futures trader.
  • Other so-called fundamental sellers in the market that day, whose combined size comprised around 63% of the so-called fundamental sales (excluding flat trading 'market makers') between 2.45 pm and 3.08 pm and possibly more earlier since bid volumes were declining, but we are not really told enough to judge.
  • The impact of bids on the NYSE quote distribution that started at 2.42 pm Eastern to cross the National Best Ask prices in around 100 NYSE listed stocks as reported by Nanex. This was due apparently to queuing delays and the presumed NYSE practice of time stamping quotes when they emerged from the queue rather than when they went in. Nanex argues that this could have led to a potential misperception that the market was falling faster than it in fact was.
  • Crowded algorithmic trades as correlations tightened across many different instruments and asset classes in the panic. Increasingly common signals were used to drive decisions of those that remained.
  • The net effect of retail stop loss or market orders on the day. The Report implicates them indirectly but never really assesses their overall impact.

The regulators conclude that the crash was primarily due to the collapse in fundamental liquidity on the day, and not to other factors. Well, most crashes are due to this ultimately. Whether this is helpful, remains to be seen. It might have been more helpful for the regulators to assess the relative contributions of ETF driven hedging on underlying stocks versus other natural liquidity for those stocks. From the Report it appears that probably all the activity in the stocks was driven by the ETFs and the futures prices. More comment might have been helpful. What role did leveraged ETFs play in the crash, for example? [See "Perfect Storm?" - my article dealing with market correlations from the Q4-2010 edition of Automated Trader]

Moreover, as for the regulators' methodology of classifying traders as fundamental buyers or sellers only by their behaviour on the day, whether this is ideal needs to be debated. Many people who intended to buy or sell as a directional move may have subsequently sold or bought due solely to the volatility rather than their overall intention. If all the liquidity is taken into account, the large future trader's contribution was his planned 9%. Since so-called fundamental liquidity is not visible, it cannot be taken as a decision variable by active traders. In many futures markets, market makers will make up a high proportion of the turnover, often over 50%. To highlight the HFT volumes that day out of context is perhaps tendentious.

So how much have we really learned by this Report from the SEC and CFTC? Quite a lot of detail but not a lot more about why it happened or how we can protect all market participants - retail, institutional, intermediaries and high frequency traders - from such clearly unfair and disorderly markets. The regulators pretty much confirmed my earlier analysis of the flash crash [click here for "What Just Happened" from the Q3-2010 issue of Automated Trader], but did not really answer the questions posed. Essentially broken trades imply regulatory failure, so how are we going to avoid such a collapse in liquidity in future when no financial WMDs have so far been found? Clearly the devil will always be in the detail, but the wrong sort of detail can be the devil's best advocate.