Tweet
May 6, 2010 flash crash (wiki)
The lords of finance constantly harp on the necessity of liquidity in financial markets. John Maynard Keynes begs to differ:
Of the maxims of orthodox finance none, surely, is more anti-social than the fetish of liquidity, the doctrine that it is a positive virtue on the part of investment institutions to concentrate their resources upon the holding of “liquid securities”. It forgets that there is no such thing as liquidity of investment for the community as a whole.*
The fetish for liquidity led to creation of many markets; there are nine exchanges on which stocks and related financial assets are traded. Each of them makes money only when it executes an actual trade. They compete among themselves to set the National Best Bid/Offer price, because that is the price at which trades actually occur.
Exchanges take great efforts not to “cross” their quotes for stocks. Prices are said to “cross” when the bid price on one exchange exceeds the ask price of another exchange. The ask price is the price at which a an owner of a security is willing to sell it. The bid price is the price a buyer is willing to pay. When prices cross, a High Frequency Trading computer will immediately buy at the lower ask price on the lower exchange, and sell at the higher bid price on the other, making a small profit.
The proliferation of exchanges creates opportunities to game the system. The market data analysis firm Nanex has produced a report explaining the mechanism of the sudden crash on May 6, when the stock market fell suddenly for no apparent reason, and recovered almost as quickly. This Flash Crash is an example of the obvious fact that there is no liquidity for the community as a whole.
The Nanex report says that the quotes from the New York Stock Exchange crossed the prices from other exchanges for a number of stocks for several minutes on May 6. Nanex suggests that this happened because quotes from the NYSE were delayed for some reason. This caused sell orders to flow to the NYSE, driven in part by HFT programs trying to make a quick profit. By the time the sell orders got there, they met lower prices which were then being issued. The other exchanges were not executing trades, so the NYSE trades were setting the prices for the stocks in question.
Nanex says that the delayed stocks were “high capitalization bellwether stocks” of many different industries. The sudden execution at lower prices caused the HFT programs to sell short, trying to capture profits on the downward slide. That exacerbated the problem. Some of the HFT programs decided there was a problem, and quit trading. That also exacerbated the problem. Prices fell drastically, scaring the heck out of everyone. Eventually people figured out that this was some kind of malfunction, and prices recovered.
Nanex backs up its claims with data showing the actual prices of the NYSE and other exchanges on May 6, and on other dates. The evidence is quite convincing.
The report also shows a number of quotation patterns from its massive data files. It refers to these as “crop circles,” possibly because they stick out from the usual random patterns of trading. You can see their data here. Take a close look at the Knife pattern, which shows quotes from BATS, a big exchange, for the stock of M & F Worldwide Corp. The x-axis is time, in this case, about 2 seconds. The y-axis is the price of bids and asks. There are four lines. The bright red line is the ask price from BATS, and the bright green line is its bid price. The two middle lines are the national best bid and ask prices. Obviously the BATS prices are far away from the best bids.
You can see why this is called a knife pattern. The handle is made up of high quotes, followed by another quote at a slightly lower price, repeated for 20 milliseconds. Then the price jumps back up to the initial point. This is repeated for 640 milliseconds. Then the same pattern repeats, only the quotes move downward twice as far before jumping back up. This lasts for 400 milliseconds, with the last drop, ending at 1080 milliseconds, not quite as far. Then there is a short burst of quotes with a much smaller increase followed by decreasing prices that lasts 60 milliseconds. Then the longer drop pattern returns, followed by a gradually decreasing pattern of drops. This lasts 400 milliseconds. Then the burst stops, and the pricing returns to what are called stub quotes, prices so far from the national best bid that they are not intended to be executed.
Nanex calls this quote-stuffing. It may have been the reason NYSE quotes were delayed. And that’s the price you pay to satisfy the liquidity fetish in badly regulated markets.
_________
* The General Theory of Employment, Interest, and Money, p. 155; Prometheus Books edition.
14 Comments
Support this site!
Subscribe to the newsletter
Advertise on Firedoglake
Send
us your tips
Make us your homepage
About Firedoglake
electronic trading programs need a 5 sec delay – indeed most folks trading from home get a 10 sec delay at best, making the whole system a scam to enrich the wealthy with direct access over the less wealthy – but so what is new about that?
Problem is to identify a trade as coming from a trading program. The Exchanges need some IT effort not sourced to those with litle QC – those off-shore worlds that now grad 7 out of 10 of the world’s engineers – and unlike the US actually give them jobs expecting to not out-source those jobs in a few years. But the QC is not great – the Met Life internal web had a dot one revision a few years back that was a disaster with loss of control of the PC and an image that was an overlay of all images connected to the site – but the Indian born head of IT did not suffer for it because he was saving money. – OK – end of rant.
Liquidity stuffing via instantaneous electronic trading versus the floor specialist making a market – I hope there is a third way to trade stocks.
Two comments:
1. It take a computer to really make a mistake.
2. This is a classic chaotic system behavior. Sudden huge change.
The system is chaotic. Something like this will happen again.
And it might be worse next time.
The whole damned system is a game. If the Stock Markets had any basis in reality, they wouldn’t be as high as they are and still gaining value.
it’s a racket.
“The proliferation of exchanges creates opportunities to game the system.”; you nailed it again Masaccio.
Amen. And the masking of insolvency is another.
“This is a classic chaotic system behavior.”
OK – a small change leads to massive change -
but I left serious math before information theory flowed back into the discussion of IT systems and physics
What does this this approach to the problem tell us about where to look for possible solutions?
If there had been an uptick rule, there could not have been short selling into the decline.
I’ve read that the knife and other patterns are meant to create background noise to delay or prevent competitors from seeing market moves before they are executed.
The markets are rigged. As I have said before, if you are not one of the riggers, you are one of the saps.
There is no solution that does not move the problem under some other rock.
It’s non linear feedback. It always causes chaos. That’s why markets have always crashed.
The addition of higher speeds increases the bandwidth (volume and amount) of the overall system, and exacerbates the extent of the inevitable sudden change.
At any time a noise spike can trigger the sudden change.
At one time in a chaotic system a control can be beneficial. However, because the system is non-linear, and chaotic, that same control can be deleterious.
The report says “bellwether” stocks.
[modnote: thank you. fixed.]
It’s non linear feedback. It always causes chaos” ??????????
seems we have been designing black boxes to kill feedback (or amplify it) for a 100 years. Chaos theory does indeed say there is no permanent solution although it allows for periods of stability. Perhaps Chaos theory is the wrong model? Perhaps the system itself can be changed to a model that is not a fit for chaos theory?
I do not know if something as simple as a 5 second delay will work – but the old out-call system did not have this much of a feedback problem – albeit psychological lack of information sell the stop order crowd panics were a problem.
These is an assumption we humans continually make, that a linear portion of a system’s behavior is the norm, and being on the linear portion of the behavior is attributed to “good management”, and “good management” gets paid much money.
This is not the case. Nor will introducing delays stabilize the system. Any change will change the system’s linear portion, but will not remove the non-linear feedback that causes crashes.
All that happens is the linear portion is changed in size, and the triggers, or tipping point, moved.
The crash of ’87 was another chaotic period. It’s just not true to state “we never had this behavior before”, because we have. All the changes we make move the linear (apparently stable) portion of the system behavior.
This make every crash look unique, and they are not. The crashes are an integral part of this chaotic system.
Nanex thinks a 50ms rule will do the trick: