Update: Larger charts:
We present this Flash Crash Summary Report using a time-line graph to
distinguish the events that caused the crash from those that were
effects of the crash. The main chart covers from 14:42:30 to 14:52:00 in
1 second intervals, and the inset covers from 14:42:43 to 14:42:46 in 25ms
intervals. Click on underlined blue text to view a supporting chart, table, or
document. To keep this document short, we will reference our
subsequent reports when possible. The color ribbon
lines are color coded to match the intensity of a data set. The scale of colors
are the colors of the rainbow and can be seen at bottom of the inset.
I. Causes ~ Negative news + quote
saturation + large (coordinated?) instantaneous sale leads to delayed price
The Greek Parliament's approval of austerity measures to avert debt
default, triggered riots in Athens, which led to speculation and fear that it
would ignite a string of defaults across Europe and the rest of the world. The
riots in Athens aired live on CNBC at the start of our summary chart. The
scaling to the left of the link indicates the elapsed time shown in the
that matches the time in our chart: an elapsed time of 4:50 in the video lines
up with 14:44 on our chart (the label 3m indicates the 3 minute elapsed time
Sale of eMini Futures and ETFs (see items 2 and 3 on main chart and
It appears that the event that sparked the rapid sell off at 14:42:44:075
was an immediate sale of approximately $125 million worth of June 2010 CME
eMini futures contracts and at the same time the immediate sale of over $100
million worth of the top
such as SPY, DIA, QQQQ, IVV, IWM, SDS, XLE, and EEM. Both the eMini and ETF
sales were sudden and executed at prevailing bid prices. The orders appeared to
hit the bids.
Quote Saturation (see item 1 on chart)
Approximately 400ms before the eMini sale, the quote traffic rate for all
NYSE, NYSE Arca, and Nasdaq stocks surged to saturation levels within 75ms.
This is a new and surprising discovery. Previouisly, when we looked at time
frames below 1 second, we thought the increase in quote traffic coincided with
the heavy sales, but we now know that the surge in quotes preceded the trades
by about 400ms. The discovery is surprising, because nearly all the trades in
the eMini and ETFs occurred at prevailing bid prices (a liquidity removing
While searching previous days for similarities to the
time period at the start of the May 6th drop, we found a very close match
starting at 11:27:46.100 on April 28, 2010 -- just a week and a day before May
6th. We observed it had the same pattern -- high, saturating quote traffic,
then approximately 500ms later a sudden burst of trades on the eMini and the
top ETF's at the prevailing bid prices, leading to a delay in the NYSE quote
and a sudden collapse in prices. The drop only lasted a minute, but the
parallels between the start of the drop and the one on May 6th are many.
Details on April 28, 2010
The quote traffic surged again during the ETF sell event and remained at
saturation levels for nearly 500ms. Additional selling waves began seconds
later sending quote traffic rates back to saturation levels. This tidal wave of
data caused delays in many feed processing systems and networks. We discovered
two notable delays: the NYSE network that feeds into CQS (the "NYSE-CQS
Delay"), and the calculation and dissemination of the Dow Jones Indexes
The delay of NYSE's quote to the CQS system has been covered extensively in
our original and
subsequent reports. However, there is
one new piece of information
which shows that some NYSE stocks were delayed more (24 seconds) than others (5
seconds). We found that stocks beginning with the letters A through M (except
for I and J) were delayed up to 24 seconds, while stocks beginning with I, J,
and N through Z were delayed up to approximately 5 seconds.
There were two separate
delays in the Dow Jones Indexes (the Dow Industrial Average, symbol DJI, is
shown in the main chart). The first delay arises from the delay in the input
data (NYSE-CQS delay) and the methodology used by Dow Jones in computing DJI -
they only use prices of trades from the NYSE for NYSE component stocks.
A second and larger delay appears to originate within the feed processors
that compute the index values. To find the second delay, we calculated DJI
using the same methodology as Dow Jones and compared the result to the value
disseminated in the feed. Our prices during the periods prior to and shortly
after the crash matched the prices disseminated by Dow Jones; however, during
the crash, we noticed significant delays. We confirmed the prices of DJI in the
feed matched or were ahead of other sources.
Financial news programs primarily used the Dow Jones Index to
convey to their audience the state of the market during the Flash Crash.
II. Effects ~ Delayed price
data leads to uncertainty, fear, and panic.
The main chart shows when significant stub quote trades occurred (light
blue dots) and the relative intensity of stub quote trades in the Stub Trades
Other reports point to Stub Quotes as a factor in the Flash Crash. This is
not supported by the data. Nearly all Stub Quote Trades, which are trades
executed at stub quotes, occur after the market bottomed and therefore
stub quotes could not have been a cause of the crash.
Stub quote trades may have been the result of a combination of a lack of
liquidity and panic induced market orders based on delayed quote data.
People watching television news sources showing the plunging Dow Index (which
we now know was delayed 2+ minutes) may have been one source of stub quote
trades. Also, the triggering of stop loss orders, which become market orders
when hit, may have resulted in stub quote trades.
However, it is interesting to note that a significant number of stub quote
trades appear immediately after the perceived bottom of the market -- when the
very much delayed Dow Index makes it's low. In fact, for those with timely
quote data, the market was actually rocketing higher when nearly all stub
quote trades executed.
NYSE Slow Mode or Liquidity Replenishment Points (LRPs)
The NYSE employs a circuit breaker known as the
Liquidity Replenishment Point or LRP. Each NYSE stock
has a LRP which is essentially a predetermined price above and below the trade
price of that stock. If the price of a trade in that stock exceeds the upper or
lower LRP price, software at the NYSE automatically turns off NYSE's
auto-execution system and places the stock in Slow Quote Mode, which means it's
handled differently by that stock's Designated Market Maker.
When an LRP is triggered, the NYSE will send out a quote with a special tag
indicating that stock is now in Slow Quote Mode. That quote is sent out the
same way that other quotes are disseminated, which means if the quotation
system is overloaded and delayed (recall the NYSE-CQS Delay), the quote with
the special tag will be delayed too. This means a lot of orders can get sent
between the time the switchover to LRP mode occurs on the NYSE floor and the
reception of the quote indicating the switchover occurred.
We have noticed that trade executions in a stock immediately after it hits
an LRP are often at substantially lower prices than bid prices on other
markets. It was this reason why we began our flash crash by first investigating
LRP's role in the flash crash. However we found that the
number of stocks switching to
Slow Mode did not become significant until well after the market had already
Anemic Recovery Volume
The Trade Rate color ribbon shows trading activity first surged at 14:42:44
(bright red), again at 14:43:20, 14:43:45, 14:44:05, and then at 14:44:30 and
remained high until about 14:46:20, where it began slowing considerably
(changing to green and cyan) and except for a brief surge ar 14:49:25, remained
low. The Quote Rate ribbon (directly above the Trade Rate ribbon) shows a
strong correlation and also begins to slow around 14:46:20 (note this is also
about the same time the NYSE-CQS Delay peaked).
This drop in quote and trading volume also occurs shortly after the market
bottomed (though the perceived bottom for those using the DJI as a proxy, would
not come for another 40 seconds at 14:47). Yet prices of stocks rocketed
higher. By 14:48, the SPY, and many other stocks, had regained all they had
lost from the drop at 14:42:44 and on significantly lower volume.
Applying simple supply and demand logic to these events might suggest the
recovery was not driven by stong demand, but rather a lack of supply; perhaps
there were few sellers left.
Large eMini Futures Discount (ES.M10 - SPY)
Beginning at 14:46:35 the difference between the price of the
SPY equity and the eMini futures contract grew to about 3% at 14:48:50, and
returned to normal at 14:50:40. A considerable arbitrage opportunity, selling
SPY and buying the future contract, existed for nearly 4 minutes. Some of this
difference could have been due to the Waddell 75,000 contract eMini sell
program that reportedly took place between 14:32 and 14:52.
III. Further Research
Here are some ideas for further research.
Nanex does not have access to brokerage account data and, therefore cannot
research those questions requiring such access.
- Were the sales of eMini's and ETF's at 14:42:44.075 coordinated?
- Were the quotes that occurred prior to the eMini/ETF sales (in question 1)
- Were the initial sales and quotes on 4/28 also coordinated?
- Can the initial events of 4/28 and 5/6 be tied together?
- Why did what appears to have been a very profitable SPY/eMini Futures
arbitrage opportunity, go unexploited for almost 4 minutes? Were there too few
traders with quote data deemed reliable enough to execute the strategy?
- Could the peak in the NYSE-CQS quote delay have been an indication that
many trading systems and networks were stressed to the point of collapse?
Is that why the quote and trade rates dropped off and remained low for the rest
of the recovery?
- Further investigate the 'flickering quote' exception of Reg NMS Rule 611
and how it might relate to many of the sequences we see
In our original flash crash report, we made 3 simple and straightforward
recommendations. We've clarified and reduced them down
to 2 (a rule banning quote stuffing is not required because the minimum quote
life rule would effectively stop this practice):
- Replace the existing time stamp in consolidated quotation feeds
with one generated immediately before that data becomes available to any
- Add a minimum quote life rule, whereby a quote must remain active
until executed, improved, or a specific amount of time (50ms) elapses.
A minimum quote life rule has far less chance for unintended consequences,
because as early as 5 years ago, it was essentially in force due to the limits
of technology at that time.
This chart shows a conservative estimate of the amount
of quotes that would have been eliminated if the minimum quote life rule was
50ms. We believe the reduction in quote traffic from this rule would have
prevented quote delays, making access to this data more fair to everyone.
Require exchanges to route orders based on data from the
Security Information Processor or SIP (e.g. CQS/CTS, UQDF/UTDF). This would
level the playing field and provide a much better audit trail for determining
if a trade received the best possible prices. Currently, each exchange
essentially must recreate their own internal SIP in order to determine whether
to route an order. The data from this internal SIP is not available (as far as
we can tell) from any exchange. We created a
video of an application we wrote
that illustrates exchange routing and the separate internal SIP.
Ban exchanges from providing direct feeds that contain core quote
and trade information. Current regulations (Reg NMS, rule 603(a)) prohibits
direct exchange feeds from transmitting data this core data to a vendor any
sooner than it transmits the data to a Network processor (e.g. CQS).
Finally, we think a solution exists without new rules and regulations: if
only the SEC would enforce, or at the very least provide guidance to,
existing rules and regulations.