Flock of Swans
by wjw on February 14, 2012
On 6 May 2o1o, the Flash Crash hit at 2:45pm and the stock market lost 1000 points in only a few minutes, spurred by frantic sell orders from automated trading systems.
Because we haven’t seen anything as radical as the Flash Crash before or since, it’s been assumed that this sort of thing is fairly rare.
A new study by Neil Johnson and his cohorts at Florida University shows that flash crashes— “black swan events”— occur more than once a day on average. 18,520 times between 2006 and 2011.
They just happen over a fairly short time scale. Crashes tend to last no longer than 650 milliseconds and spikes no longer than 950 milliseconds— which happens to be the reaction time of the programs’ human operators. (It’s assumed that crashes attract more attention from the humans than spikes do.) When the operators see a crash happening, they intervene.
These trading programs know nothing about actual economics— the programs that carry out 73% of all trades know nothing about goods and services, cost of production, labor services, capital, productive efficiency, supply, demand, efficiency of management, industrial organization, demand for labor, externalities, value, and other factors that professional traders are supposed to understand. They only know how markets have reacted in the past, although they know that very, very well.
The study assumes that black swan events are a common, inevitable result of programmed trading. And the entities most susceptible to these kinds of flaws are . . . wait for it . . . major international banks— which, they say, “hints at a hidden relationship between these ultrafast ‘fractures’ and the slow ‘breaking’ of the global financial system post-2006.”
And of course the banks are creating even faster trading programs, and installing faster infrastructure to get their trades to the market just that few milliseconds faster than the other guy, which might just well lead to more of these “fractures” and “breaks.”
Woo-hoo. Another brave new world of finance, because the last one just isn’t broken enough.
And let’s just point out that the study’s thesis is even more relevant as of today, when the CME Globex Crude market went offline for 75 minutes, due to, ahem, “a blast of quotes corrupted a memory queue causing the software to believe the queue was full all the time.” One of our major markets just, y’know, went away, cuz their buffer was full . . .
Algorithm error? Quote stuffing? Enemy Action? Who knows?
(And if anyone does, we won’t find out. It might undermine our confidence.)
The plot of Hardwired features a massive stock fraud framed so as to take advantage of gullible trading bots. For dramatic reasons, I had to make the trading action last the length of one air battle. Maybe the battle should have been over within 650 milliseconds.
Hardwired still works because it was based on human action, not computers. Most stock trading done by actual humans involves specific pricing decisions, not “buy/sell at whatever the market price happens to be at the instant I push the button”.
And you don’t need to invoke arcane trading activity to explain how repeatedly underpricing risk could lead to financial trouble. It may well be that high-fructose corn syrup is metabolized differently than other carbohydrates, but that isn’t what makes a 4500-calorie-a-day eater fat.
I have loeokd at clique and am very impressed by the granularity and multiple identity support . I think it’s on the right track.Your points about the challenges of decentralization are valid, but if you don’t decentralize you are forever depending on whoever operates the central server to behave themselves (Facebook being a prime example of why this might not be a good idea). If protecting privacy is among the key goals, you have to build a decentralized ecosystem. You are also correct that most people won’t bother running their own node. However, I can’t see how one gets real privacy and real control over one’s data without doing that.
Comments on this entry are closed.