The Gateway to Algorithmic and Automated Trading

What just happened?

Published in Automated Trader Magazine Issue 18 Q3 2010

There's high-frequency trading, and now there's high-frequency crashing. But identifying the causes of the 6th May 2010 'flash crash' has proved to be a relatively slow process. Bob Giffords goes in search of causes, outcomes, explanations and lessons to be learned.

On Thursday, 6 May, at just after 2.30pm Eastern time, the US capital markets dipped, crashed suddenly, but then recovered almost as quickly in the space of a few minutes. Ever since regulators, market organizers, participants and the media have been debating the 'flash crash' and what caused it. Some 20,761 trades were busted or unwound by the market authorities, for being more than 60% away from their last print price prior to 2.40 pm. This left some firms nursing significant losses, while others enjoyed substantial windfall gains. The financial crisis just took another bizarre twist.

Mark Holt

Mark Holt

In their joint preliminary report the relevant US regulators, the SEC and CFTC, confirmed that 98% of trades executed during the critical 20 minutes remained within 10% of their pre 2.40pm price, but that 11,510 trades were executed at negligible prices of one cent or so, against so-called stub quotes at derisory prices posted typically by market makers to meet a quoting obligation. ETFs made up 70% of the securities with broken trades. However, NASDAQ stocks made up over 59% of the broken trades themselves, while ETFs only had 4,903 trades that were broken. NYSE had no broken trades. Curiously 70.1% of the stub quote trades in the five minutes after 2.45 pm were short sales, while 90.1% of the busted stubs in the following five minutes were shorts -- why would anyone go short for so little?

"Obviously it makes no sense to short sell at a few cents a share," says Mark Holt, head of systematic implementation at BlueCrest Capital Management. "So either traders were selling at market and their systems could not keep up or they were trading relative value with a shorter and shorter alpha horizon. If technology has reduced the costs of entry, perhaps the level of expertise has fallen as well."The CME futures market also claimed no broken trades. Overall the major stock indices including the Dow and the S&P 500 apparently never fell more than 10%.

The Nomura analysis of 6 May, according to Andrew Bowley, their head of electronic trading product management, identified early market declines of 3% over Greek debt and the withdrawal of two large, high-frequency market makers as sufficient cause for the New York Stock Exchange's slow market that pushed it into manual auctions and separated its prices from the other ATSs. This combined with a $4 billion option trade on the S&P 500 to trigger more selling until it all reached a tipping point.

Formally speaking, the NYSE manual auctions were called liquidity replenishment points or LRP and were intended to give the market a breather on a particular stock to regroup. NYSE invoked LRPs on 1000 stocks that day, compared to a typical quota of around 50.

Andrew Bowley

Andrew Bowley

"We've seen no evidence of any erroneous trades," says Paul O'Donnell, COO at BATS Europe, referencing their US platform. "There was a large futures sale going on at the time, but this does not appear to be erroneous." For O'Donnell the crash appears to have been a series of events - Greece, the oil rig spill in the Gulf, etc. - that caused growing selling pressure and a downward trend in prices, combined with a variety of market structure factors. NYSE LRPs kicked in, taking out 20% of the market, just when people were desperate for liquidity, then NYSE Arca had problems, with Nasdaq and others declaring self-help. BATS too apparently had problems with Arca but also at two other ATSs. "Then we saw some market makers pull back and a flood of sell orders from two large broker dealers at 2.40 pm Eastern," continues O'Donnell. "It just all came together at the wrong time. Prices plummeted on many stocks but came back equally quickly."

Under the Reg NMS rules, market centres can declare self-help and cease to forward orders under the best execution rules, if the receiving platform fails to respond correctly within the required latency.

"When the big board went slow due to hybrid auctions and others went into self help, traders had to move quickly to grab what liquidity was there," says Jeff Wecker, President and CEO at Lime Brokerage, an independent agency broker focusing on the high performance and algo trading community. "That might have made matters worse than they otherwise would have been."

Indeed matters were pretty bad. "If we look at Procter & Gamble it ticked up 55% and down 36% tick to tick over four or five minutes and the fail-safe rules weren't able to prevent people being left on the wrong end of some trades," says Andrew Morgan, head of Autobahn Equity Europe at Deutsche Bank. "Those swings couldn't have happened in London where you have the 5% dynamic price tolerance at LSE. While such rules are replicated across most venues in Europe, they're not consistent, which means you might have a diluted version of the crash happening in Europe. Still it's safe to say the US extremes couldn't happen here."

"Self-help declarations and LRPs occur on a regular basis," notes O'Donnell at BATS. "Normally it's not a problem, but when everyone is selling and with so much capacity out of the market, the available liquidity was depleted on the remaining venues."

While some piled into the market, others pulled out and cancelled their orders. "Traders who make markets in single equity stocks without a quoting obligation, normally have basis as a significant component to their model," says Nick Nielsen, head of trading at Marshall Wace Asset Management. "They timestamp their orders and then compare when the exchange responds. Their alpha models depend on normal latencies. So when response times grew beyond N standard deviations, they stopped trading; cash equity and futures prices then quickly diverged."

"What we saw was a jump down in the e-mini futures price of around 5% just before 2.40 pm," says Wecker at Lime Brokerage, "followed by disruption first among ETFs and then their underlying stocks. It seems as if the futures traders tried to hedge in the ETF market and when those traders got run over they all shifted to single stocks. This happened electronically in great waves of small orders that swept out the bid sides leaving prices to fall through to the stub quotes."

"Some market makers shut down on all venues," continues O'Donnell at BATS, "just as a lot of stop loss orders flooded the markets. While the overall S&P 500 index only dropped 7%, individual stocks fell through to stub quotes at a penny. A lot of reversion traders then came in and the market recovered."

"All but two of our clients traded through the incident," says Wecker at Lime Brokerage, "although some may have widened their margins or paused occasionally. The two that withdrew apparently had technical problems given the extraordinary data volumes. Overall our clients traded a couple of hundred million shares during the crucial 20 minutes."

"Macro high frequency traders, are often driven by the short term momentum in the futures," notes Nielsen at Marshall Wace. "They continued to trade through the crisis without real problems. Here high frequency traders had their best day ever. On the other hand, retail flow internalisers with quoting obligations had their worst day ever. They didn't want to trade but had to."

Clearly NYSE Arca was experiencing some problems as several self help declarations against them confirm. "We actually declared self help against Arca after the crisis was passing at 2.49 pm," says O'Donnell at BATS. "We had experienced over 1,800 reject messages on marketable orders routed to them during the preceding 5 minute period. Five minutes later we resumed trading with them under the trade through rules."

"Liquidity tends to cluster around the last traded price, plus or minus 7% or so," says Dennis Lohfert, fund manager at Ion Asset Architecture. "Normally there's no liquidity around 50% of the expected price. So when liquidity dries up, as it did on May 6, any market orders will fall through to whatever is still resting around the very bottom of the order book." And that seems to be what happened on the day.

Jamil Nazarali

Jamil Nazarali

The pressure was clearly on. "When you're on the roller coaster you really start to question how far
you trust the model," says Holt at BlueCrest.
"We have a strong culture of trusting the system, believing in our research and technology. When you've back-tested a model, you don't want to make discretionary adjustments.

You either take the hit or shut it down. There's no tweaking."

Holt points out that what happened on 6 May was part of a larger process of euro zone instability that still continues. "We stayed in the market on Thursday, betting on reversion, and the inventory we took on pushed our risk limits beyond our comfort zone," says Holt. "As the markets bounced back on Friday and then on Monday after the euro bail out, our model's strategy was vindicated, so we were OK."

As fast as the markets fell, suddenly they rebounded. "Some traders clearly pulled their orders because they couldn't trust the pricing feeds," says Jamil Nazarali, global head of the electronic trading group at Knight Capital Group. "Once they realized the feeds were OK, they rushed back." He argues that with lots of large cap stocks on the floor, traders saw little risk and so started buying. It then snowballed and prices recovered as traders took their profits. "It was only late that evening when regulators bust their trades, that they found themselves with short positions," he concludes.

"Cancelled trades happen in some markets more than others," says Lohfert at Ion, "but they happen, and systematic traders need to take that into account.

"We only had 500 broken trades at the end of day," notes O'Donnell at BATS, "and probably for two reasons. First we convert all market orders to limit orders with a collar at 5% or 50 cents. So those orders didn't chase prices down. Then we don't automatically refresh stub quotes. Other market centres with the automatic refresh functionality for stub quotes suffered many more broken trades."

"In the end from the millions of trades we executed during the crash we only busted 230," says Wecker at Lime Brokerage, "and of those approximately 200 were ETFs. That gives an idea of the size of the crowd focusing on ETFs."

"We were in the office late into the night, first waiting for news and then reconciling lists of busted trades," says Nazarali at Knight. "We had to hedge our risks in after hours trading wherever we could; we even resorted to partial hedges in the Asian futures markets."

"Just around 100,000 sides were cancelled out of a daily volume of about 100 million that the US has every day - proportionately it was not very large," says Diana Chan, Chief Executive Officer of EuroCCP. "Cancellations are all captured automatically by the central counterparty, as part of our business as usual process. The hard work is on the trading venues - they need to select the trades to be cancelled."

Speed Kills

It had been a rough ride. Nasdaq reported its highest trading volumes on record, while NYSE reported its second highest. The SEC and CFTC conclusion was that total volumes were up by a factor of 2 over recent normal days, but peak second rates were up rather less.

"May 6 was the highest day ever in US markets for pricing data, reaching 2.808 million messages per second at the peak around 2.43 pm," says Jeff Wells, VP product marketing for Exegy, the hardware accelerated ticker plant supplier. "That includes all the major equity cash, futures and especially the options markets. Previous highs had only been around 2.4 million."

Wells confirms that all the electronic order book exchanges experienced sudden sharp peaks around 2.43 pm that afternoon after building steadily from 2 o'clock. All the major platforms, such as ARCA, BATS, NASDAQ and Direct Edge, had, however, individually experienced higher peaks previously in terms of messages per second. "ARCA for example peaked at over 277,000 during the crash," observes Wells, "but had experienced over 300,000 back in December 2009 and even more in April 2010. Nevertheless, when NYSE went into manual mode, that left other exchange systems quite exposed as orders were routed to the remaining centres."

He goes on to explain how peak message rates may, however, disguise what's happening over smaller timeframes. "Platforms with market data peaks of 200,000 messages per second might show instantaneous rates in any given millisecond equivalent to over 3 million messages per second," says Wells. "Our data doesn't track these microbursts, but they can give trading engines and data feed recipients a hard time."

High data rates can be a challenge even for advanced players. "We had some issues with high data rates," says Holt at BlueCrest. "One of our brokers slowed down significantly, and the cross Atlantic link to our decision support systems for traders in Geneva slowed. Normally we filter out insignificant ticks but the size and speed of the changes meant that data rates soared. We'll need to adjust those limits, but the trading engines in the States were OK. Capacity is key if you want to stay in the market."

Nielsen at Marshall Wace suggests that part of the problem was probably due to penny quoting. "This was brought in to help retail investors," says Nielsen, "but on highly priced stocks it has drawbacks with regards to market data processing - there are many more price updates than necessary due to small minimum bid / offer spreads. So there was an order of magnitude increase in the numbers of cancel-replace transactions, in fast markets people can't keep up, so the ticker falls behind. May 6 was simply a market data problem." He notes that in Europe or Tokyo tick sizes are banded by price. "It avoids excessive updates and makes market data much more manageable," says Nielsen.

Speed also potentially creates problems for the post trade clearing. "If during the day there is large volatility in the market, the central counterparty could face extraordinarily large exposures to certain clearing participants," says Chan at EuroCCP. "Monitoring the amount of exposure to a clearing participant is fine, but what really counts is the central counterparty having sufficient margin on hand if a firm defaults". Chan explains that trading is so fast that by the time an intraday margin is collected, trading positions and the central counterparty's exposures could have completely changed. If the central counterparty keeps collecting large amounts of margin during the trading day, it could be very disruptive to the liquidity management of the clearing participant. "Whether to collect intraday margin should require human judgment," says Chan. "That old concept of 'know your customer' is really important. The central counterparty needs to know the normal trading volumes and be aware at all times of the financial health of its participants, so that it can take the appropriate actions during a stressful market."

Andrew Lo

Andrew Lo

Rooting around for causes

"The financial media are always trying to explain why things happen and to attach a simple label," says Lohfert at Ion, "but markets aren't like that. Often there is no explanation, or many, and it's very difficult to tell why a particular move has just occurred." Indeed many systemic failures can be traced to a series of things going wrong. At least five different factors, a confluence of events, combined apparently to trigger the crash:

  • Economic failure of the global financial crisis and the worries over sovereign debt
  • Technical failures, apparently, of some market centres leading to the self help declarations and some traders leaving the markets
  • Regulatory failure: the lack of robust, automated mechanisms in the market microstructure to deal with the volatility
  • Human inexperience: many, especially retail, traders used market orders to protect themselves against such eventualities, but these proved to make things worse and lead to unexpected results
  • Quant design errors which led to crowded trades in the extreme conditions of the flash crash

The last point is made particularly by Andrew Lo, director of the MIT Laboratory for Financial Engineering and chief investment strategist for AlphaSimplex Group, an absolute return specialist owned by Natixis Global Asset Management. He finds parallels between the flash crash and the quant meltdown in August 2007 when a group of independent, long-short equity-market-neutral portfolios all suddenly suffered significant losses, and then all recovered at the same time. "My simulations with Amir Khandani suggested that this was a 'crowded trade' phenomenon where the rapid unwinding of a large equity-market-neutral portfolio may have caused losses in other portfolios using similar strategies," says Lo, "leading to a deadly feedback loop of more losses, increased collateral calls, and cascades of forced deleveraging. Both events occurred in the midst of growing market pessimism and flight to safety: in 2007 it was the subprime crisis, in 2010 it was worries over Greece."

Some people of course try to blame the quants and their high frequency trading friends for the whole of the May 6 crash.

"High frequency traders are the good guys," says Rob Flatley, CEO and president of Netik, a market data provider for ETFs and market indices. "They keep the market efficient and everything aligned. When some of them bail out, while others pile in, that's when things come unstuck. We need to learn from this."

Hirander Misra, co-founder and chief executive of Algo Technologies, puts it quite bluntly: "These high frequency traders are some of the most risk averse guys in the world. They're trading with their own money, so they're very conservative. It's all hard coded into the algorithms."

It is important however to understand trader behaviours. "There is a tight link between futures, options and cash trading," says Jesper Alfredsson, VP of product management, Orc Software, recent merger partner with agency broker Neonet. "Options traders and market makers like volatile markets, but given their exposure, they can find it very hard to unwind positions if the markets behave erratically like at the time of the flash crash. Inevitably they will need to cover some of the unwanted risk in the cash markets and that most probably happened on 6 May."

Regarding the high number of ETF breaks, Flatley at Netik argues that you have to remember there are two communities here: "The high frequency players know the risks of busted trades and will have been quick to pull their liquidity," he notes. "There are also a lot of retail traders, who like to use market stop loss orders well below the bid to protect themselves. This time they got caught out, as market orders fell through to the bottom."

"High frequency traders arbitrage ETFs in two ways," continues Flatley. "They arb the index and its underlying reference stocks, but also ETF pricing efficiency, the difference between the ETF price and its net asset value. If ETF market makers pull out over liquidity worries, ETFs will quickly become disconnected from the underlying market. That appeared to be what happened on 6 May."

"Whatever caused the initial shock, the fragmented, non-uniform market structure with an overlay of burdensome Reg NMS self help rules only increased the speed and dysfunction around the response," concludes Flatley.

Finding the Way Forward

Wecker at Lime Brokerage believes the SEC has come up with a couple of "tremendous" stock-specific circuit breakers that should avoid another free fall like the flash crash. "First," he says, "there is the new short-selling rule, yet to take effect, that would require a bid test once a threshold fall was reached; and then, there is the maximum fall of 10% in five minutes on any S&P500 stock that will halt trading in that stock on all market destinations to give everyone a breather. I think they are trying to strike the right balance with minimum interference in price discovery."

Others are not so sure. "Markets have now started to use the new SEC circuit breakers on a pilot basis," says Nazarali. "Already one manually miscoded price in an over-the-counter trade print stopped all the markets. This should have been foreseen with any reasonable consultation process, but it was rushed through regardless."

Indeed Nazarali argues that many of the proposed regulatory remedies could wind up making things worse. "Some markets, like Nasdaq have introduced new volatility rules like the NYSE LRP, but all the exchanges and ATSs are different," he says. "With fragmented responses, liquidity could still flow to the weakest link. The SEC has now introduced this circuit breaker and a new short selling rule, but it could just deter market makers from supporting the market if they worry their trades might get busted. Even worse since markets can fall so quickly, traders will withhold their risk capital even as we start to approach the thresholds, accelerating the decline."

Alfredsson at Orc Software warns that the SEC short selling proposals could also make life more difficult for option traders, who use short selling as a hedging technique, not as a directional play."

Meanwhile, Lohfert at Ion argues that the proposed circuit breakers are "a really bad idea". "Markets abhor a vacuum," he says, "so they'll trade OTC, related instruments or synthetics, anything where they can get a price. If you lock down the futures, people will trade forwards or trade synthetics via options. Regulators just don't understand that it's not the trading that needs to be stopped." Essentially Lohfert argues this disenfranchises a portion of the market that is restricted from trading these other avenues, concluding, "It just creates black holes and uncertainty."

O'Donnell at BATS believes that circuit breakers are part of the solution, but insists, "We can't let one rogue print shut down the whole market as it recently did". He thinks it would be better to impose a price collar up and down so no trades ever have to be broken. Trading could continue within those limits. "In Europe," says O'Donnell, "we have volatility halts similar in concept to LRPs, but the price collars are all geared to the primary exchanges since we don't have a consolidated tape."

Bowley at Nomura adds that a key difference in Europe is the broker convention that auctions are only held on the primary exchanges: "If a volatility auction is called, we all shift liquidity from the MTFs to the primary market auction. There's no regulatory requirement, but brokers have done so to re-establish robust price discovery on that stock. In the US the ATS platforms tried to trade through, which dislocated prices, drained liquidity and the market fell through to stub quotes."

Not everyone may move their liquidity however. "If people want to trade through a volatility auction we shall do it for them," says Richard Balarkas, President and CEO of Instinet Europe, "but we make sure that they realize how risky that can be. I don't think we should restrict that. We need to allow for competition and choice, but sometimes it will also be sensible to insist on the market taking a breather."

"In the past when LSE had technical problems," says Bowley, "they put the market into a pause mode which many firms treated as being an auction interruption, so the market transferred liquidity to LSE and everything stopped. Consequently, brokers are recoding their algos to trade through the primary market auctions on the MTFs. Now, perversely, Europe may be moving towards the American model. Perhaps this ought to be reviewed."

O'Donnell agrees that a "pan-European solution" is also needed here.

Alasdair Haynes

Alasdair Haynes

However, remedies may require more than merely volatility interrupt auctions. "At Chi-X we simply do not accept market orders, only limit orders, so traders know what prices they will accept," says Alasdair Haynes, CEO of Chi-X Europe. "Moreover, we don't accept prices that are more than 20% away from the last print in any market. That helps to catch fat-finger problems." He further argues that in Europe the lack of a trade through rules like Reg NMS is actually an advantage. "The brokers route orders and thus can cater for unique conditions," says Haynes.

Balarkas at Instinet is also concerned about the Reg NMS rules. "I don't think exchanges should be required to onward route. I don't think anyone is better than we are at smart routing. Stats I've seen demonstrate that very few firms are able to do this properly. Besides looking at onward routing in the US, we should look at what kinds of orders we allow on the exchange. What about snake-in-the-grass orders? Should we allow these and stop loss orders so far away from the prevailing price? This is the approach we should be taking."

"There are some things I think we shouldn't compete on, such as tick sizes, for example," says Morgan at Deutsche Bank. "Consistency is vitally important. We're striving to get a model in place that encourages innovation and competition and efficiency, but at the same time has the appropriate level of consistency to make sure that the worst consequences don't play out."

Most people would seem to agree with Nazarali at Knight, when he says, "Busted trades are really bad news." He argues that wherever you draw the line, one trader wins and another loses, so there's huge uncertainty about positions. People will hard code their risk aversion into their algos. Instead, the exchanges need to intervene before the trades occur, if they might get busted, like the CME did on the e-mini [futures contract]. That will encourage traders to support the market."

"If you're running a portfolio strategy," says Holt at BlueCrest, "you're hedging approach is pretty intolerant to busted trades. Actually they worked out pretty even for us. We lost profit but it didn't increase our risks. Still it might change our view in future. Trade breaks are never the right answer. Even though we're using technology to execute, our attitude is that, if we do a deal, we do a deal."

According to Wells at Exegy US markets have already exceeded 3.2 million messages per second on the 29 June, so something needs to be done.

Moreover, Misra at Algo Technologies confirms that they have just launched the fastest exchange-matching engine in the world with 16 microseconds of round trip latency. "That compares to 300 to 400 micro seconds on other existing exchanges, but we think we can get it down even more." So the race goes on. "From a market structure view we need to introduce more checks and balances to ensure fair and orderly markets," says Misra, "but it has to be consistent. The good news is that as technology advances you can put in more validation and still not impact latency; multi-core technology makes that possible as we saw when sponsored access came in. There was very negligible latency impact. We can clearly do much more, but it has to be consistently deployed."

There appear to be a lot of ideas for change and some technology solutions. However, as Bob Fuller, non-executive director of Fixnetix, the front-office outsourcing specialist, puts it, "The big problem here is the pace of change. Regulation coming from both sides of the Atlantic is greater than people have ever seen before. It's absolutely unreal. Any one of these changes might structurally impact the whole environment yet it's all happening at the same time. Anything moving this fast can create instability." Fuller also notes the structural changes in the marketplace: in Europe

NYSE Euronext will move away from LCH Clearnet, as may LSE. LSE is proposing a new derivatives market. Two new bond initiatives have appeared.

There are also a large number of MTF applications pending, as regulators insist on greater transparency and controls, yet MTF margins are paper-thin. "In the US," says Fuller, "a number of prop traders have left the larger firms fearing greater regulation, so we could see even more competition. This could be the growth of the hedge fund industry to the nth degree. Regulation always triggers market change and firms will have to adapt to that as well."

So, what should we do? Perhaps Bowley at Nomura has the best answer. "The balance of technology and rules is comparable to Formula 1, where we see increasing rules intended to slow the cars down to make them safer, yet in reality the growth of speed reduces, but the absolute speed itself relentlessly increases." With F1 at least they review the safety rules every year to cope with innovation.

"When we put more powerful engines into cars, we need to upgrade the brakes, put in air bags etc., to maintain safety," says Morgan from Deutsche Bank. "We need the same approach in financial markets to give supervisors red-amber-green type signals. There's a lot more potential here for momentum signals, volatility, directional moves or relative spreads, not just price thresholds, to make the process more robust. As long as they're applied consistently and communicated properly, it would get us to a place more appropriate to the reality of markets in 2010."

Besides more regular reviews, crash scenario planning may also be called for. "Some people believe central counterparties got away lucky with Lehman Brothers," says Chan at EuroCCP. "The fact is that being able to manage defaults without loss requires preparedness. The ability to deal with very big problems lies in the depth of experience of a CCP. Months before Lehman Brothers collapsed, DTCC actually ran a simulation of all that needed to be done to close out the positions of a very large investment bank, and picked Lehman Brothers as the example. Even though DTCC has dealt with over 30 participant bankruptcies without loss over three decades of experience, it is not complacent. Here in Europe we are also constantly looking at how we can take risk out of the system and doing business continuity drills."

The world has indeed changed. "What took a few days in August 2007 can apparently unfold in a few minutes today given the amount of high frequency trading that now exists, making it easier for prices to become temporarily disconnected from fundamental values," says Lo from MIT and AlphaSimplex. "Yet while such short-term dislocations may seem more likely now, we also have to acknowledge that high frequency trading has significantly reduced trading costs and improved the speed with which investors can get in and out of positions. We should try to preserve these benefits, but can no longer ignore the potential risks of technical glitches causing market turmoil in the presence of too much liquidity."

"Now we actually see market structures that reflect buyer power," says Balarkas at Instinet, referring to this as a 'democratisation' [see also page 52]. "The people calling the shots in terms of costs, latency and innovation are investors, and some high frequency traders are the purest form of those investors."

Many agree that high frequency trading is worth protecting. "While we trade high frequency, we also look at the six month, one month and 5 day trends, deploying significant levels of capital," says Holt at BlueCrest. Although we aim for market neutrality, we're not market makers always trading flat." However, he notes that since high frequency has reduced technology costs, many traders are not as well capitalized as they use to be, so their risk tolerance is lower. They pull their liquidity more quickly. If markets want to rely on them, they'll need to be protected with more liquidity controls. "The game where bulge-bracket firms used client flow and wide spreads to fund the risks is over," says Holt.

However we certainly need to do something. As Haynes from Chi-X says, "No exchange in the world can ignore what happened in the US on May 6."

Anatomy of a Crash

Drawing on official documents and news reports, Bob Giffords charts the course of one day's trading in New York - 6th May 2010.

Market open

Market open British general election takes place with growing worries over a hung parliament

News bulletins feature rioting in Athens on the previous day over the government's austerity programme, including the firebombing of a bank. Greece is facing a €300 bn debt rollover on 19 May.

The CBOE Volatility Index (VIX), known as the fear gauge, opens at 25.88 and stays level during the morning, 25% higher than recent levels

8.30 am

ECB press conference makes no mention of a widely expected bail out for Greece; Greek CDS spreads widen to 937.9 from 844.2 the previous day and under 400 in February

11.00 am

Dow Jones Industrial Average (Dow) is down 60 points, with heavy selling; the percentage of stocks falling without first ticking up reported to be at its highest since the World Trade Centre disaster on 11 September 2001

1.16 pm

NYSE securities in reserve mode, i.e. paused for manual auctions called Liquidity Replenishment Points (LRP), starts to rise

A market maker stops quoting on BATS

1.30 pm

The VIX starts to rise, COMEX gold market closes with prices having risen from $1180 to $1210 / troy ounce, US Treasury yields are falling as there is a flight to quality

1.46 pm

NYSE securities in reserve mode, (LRP) reaches around 100, compared to a normal daily value around or below 50

BATS notes a spike in price alerts

2.00 pm

The VIX reaches 28.6 signalling increased volatility in the S&P

Dow down -1.5%, S&P500 down -2.9%, June 2010 e-mini S&P500 down -1.3%

BATS reports a large increase in sell orders routed to them from another exchange

2.02 pm

Eurodollar & euroyen exchange rates fall suddenly due to sovereign debt worries

NYSE securities in reserve mode, (LRP) rises to over 200 during the next half-hour

2.15 pm

A hedge fund, apparently, places a $7.5m bet buying 50,000 options on a fall in the S&P 500 ending June at 800, paying $4 billion, as the markets continue to fall the counterparties start to lay off their risks

2.23 pm

Nasdaq surveillance system starts issuing alerts in numerous securities due to abnormal behaviour

2.24 pm

First stock hits a daily low more than 80% down on previous night's close against a market maker's stub quote; most of these derisory daily lows occur after 2.40 P.M.

2.30 pm

The VIX reaches 31.71 and keeps rising

A trader starts a major hedging transaction selling the June 2010 e-mini S&P 500, representing 9% of the market and completing the trade around 2.50 p.m.; the e-mini starts to fall leading the cash ETF on the S&P down

Volumes traded on e-mini S&P 500 futures is 3-400% higher than on the S&P cash ETF (SPDR), price ranges were however similar CBOE declares self help and stops forwarding trades to NYSE Arca - essentially declaring that NYSE Arca were not responding promptly and were outside NBBO rules

2.37 pm

Nasdaq declares self-help against NYSE Arca

2.38 pm

Nasdaq OMS BX declares self-help against NYSE Arca , although this was published to the market a couple of minutes later

2.40 pm

Two major, high frequency liquidity providers, seeing too much chaos and apparently worrying over cancelled trades, try to pull their trades; instead of taking a few seconds, one reports taking 2 minutes to close out.

BATS reports receiving an unusually large volume of sell limit orders from two large broker-dealers

Dow, e-mini S&P500, and S&P500 cash index all fall approximately 5% over the next 5 minutes

2.44 pm

A very few extreme highs begin to occur, matching against very high stub quotes, and continue until 3.00 p.m. NASDAQ-100 Cash index spikes down hard relative to the NASDAQ-100 E-mini which is down relative to the S&P 500 E-Mini; at this point NASDAQ shares lead the S&P500 shares down

2.45 pm

June 2010 e-mini S&P 500 hits its daily low of 1056.00 (-9%) and is so volatile that CME issues a 5 second lock, the e-mini keeps trading but not below its price at the time of the lock, since that would trigger stop loss trades and further falls which might be have to be cancelled under NASDAQ rules; the lock enables participants to build liquidity, executions only take place at the end of the lock, and e-mini starts to recover, followed closely by the S&P 500 ETF (SPDR), see below

Volumes traded on e-mini S&P 500 futures is 580% to 800% higher than on the S&P cash ETF (SPDR) until 2.55 pm, price volatility on the SPDR is 5x that of the e-mini

Over 200 securities fall by 50% from their 2.00 pm levels

NYSE puts Procter & Gamble shares into reserve mode (LRP), NYSE securities with LRPs hit dramatic high of 1000

3M prints low of $67.98 (-20% from 5 May close, -18.39% from price at 2.40 pm)

The NASDAQ-100 cash index relative to the NASDAQ-100 E-Mini stops falling and reverses. The S&P 500 cash index spikes lower relative to the S&P 500 E-Mini but the NASDAQ-100 E-mini remains weaker than the S&P 500 E-mini

70.1% of the stub quote executions below 6 cents per share until 2.50 pm are short sales,

2:46 pm

NASDAQ-100 E-Mini moves higher relative to the S&P 500 E-mini that is still suffering from the S&P 500 cash index sell off.

S&P cash index stops and reverses relative to the E-mini.

VIX hits 40.26 and then oscillates between 40 and 35 until 3.36 pm at which point it starts to fall

S&P cash ETF (SPDR) hits low (-10%) and then starts to build, following the e-mini

Liquidity dries up on the iShares Russel 1000 Growth Index ETF with only 4 bids above $14 vs. $51 a few minutes earlier NYSE puts Accenture shares into reserve mode (LRP)

Dow hits its daily low of 9,872.57

NASDAQ 100 hits its low of 1752.31

Dow, the e-mini S&P500, and the S&P500 all start to recover by 5% over next 5 minutes

2.47 pm

Accenture prints low of $0.01 on other venues (-460 from 5 May close) for a period of 7 seconds

Procter & Gamble prints low of $39.37 on other venues (-36% from 5 May close)

Eurodollar and euroyen exchange rates spike down to daily lows and then recover back to the levels at 2.02 pm, CME declares locks on Yen and Sterling - can still trade but not at a lower price for 5 seconds to allow liquidity to build

2.49 pm

BATS declares self help against NYSE Arca

2.50 pm

NYSE puts 3M shares into reserve mode (LRP)

Dow recovers a bit, trading at 10,232; e-mini S&P500 recovers, trading at 1,096 and continues to strengthen

90.1% of executions against stub quotes below 6 cents per share until 2.55 pm are short sales

2.54 pm

BATS revokes self-help against NYSE Arca

2.57 pm

Trading ranges of e-mini and cash ETF on the S&P (SPDR) return to similar normal ranges

2.59 pm

June 2010 e-mini S&P 500 recovers to a little over -4%; S&P cash ETF (SPDR) also recovers to little over -4%

3.00 pm

Selling pressure starts to abate, S&P recovers

3.15 pm

500 NYSE securities declaring LRPs still dramatically higher than normal but about half the 2.45 pm peak, LRPs declines slowly through the afternoon