The Twitter flash crash: A black swan event?
Recently financial markets endured a flash crash social media-style. The very first Twitter flash crash. For those of you who don’t know what I’m talking about: A few weeks ago, the Twitter account of the American Press association was hacked. The hackers used the account to send out the following fake tweet:
Afterward, the Dow Jones plunged 150 points before bouncing back up to “normal” levels. All within half an hour:
As a side note, I think it deserves a mention here that the media (as usual) did its utter best to exaggerate what happened, stimulating panic all around. The Dow Jones plunged 150 points in only a few minutes, which of course is a big deal. But in most images featured in the media it looked like the Dow had a total meltdown, because of the choices made with respect to the axes of the graph shown. For instance, if you were to draw a five-year horizon chart (below), axis would be at quite different levels. From such a perspective, our flash crash would have been nothing but a small bump.
After the crash, the same media exaggerating the scene were the first to scream blue murder about how financial markets have changed for the worse with the rise of social media. All kind of experts were quick to start jabbering away about the risks of Twitter as a user-generated news agency.
Now that most of the dust has settled, I thought I’d share my take on the matter. Please note that I’m not a financial expert. Neither do I have any intent (or the skill set) to become one. My knowledge of financial markets is limited to my finance Masters and my daily work as Founder of SNTMNT (which also means that I’m slightly biased). From my humble point of view though, I don’t believe that the problem here is Twitter. I fully agree with Mark Gongloff that the actual problem is way more fundamental than that: the algorithmic trading systems that are using Twitter as an input source.
I believe that a lot of the buzz and confusion around the events of April 23 can be related to a theory I recently learned about. It’s an interesting theory from 2007 by Nassim Nicholas Taleb, called black swan theory:
Black swan theory
Black swan events are unexpected events of large magnitude and consequence and that have or will have a dominant role in history. According to Taleb’s black swan theory, many major scientific discoveries, historical events, and artistic accomplishments can be qualified as “black swans”. Taleb gives the rise of the Internet, the personal computer, World War I and September 2011 as examples of black swan events.
I thought it would be interesting to see whether the recent flash crash can be considered such a black swan event. According to Taleb, an event qualifies as a black swan when:
1) The event is a surprise (to the observer).
2) The event has a major effect.
3) After the first recorded instance of the event, it is rationalized by hindsight, as if it could have been expected.
I don’t believe there will be much call for debate when it comes to the first and second qualifications of black swan theory, even though the media exaggerated point two. The third qualification on the other hand, allows for a lot more discussion in my opinion. I understand that many investors would argue that the Twitter Flash Crash was highly unexpected. The technologies behind the flash crash, mostly text algorithms mining the firehose for specific keyword pairs are all based on AI related concepts, which is in itself a very new science.
Although I do agree that the size of the black swan was unexpected, I strongly disagree (without rationalizing of course :) ) that nobody could have seen it coming though. This is because of two reasons:
1) Back in 2011, the Atlantic wrote a great article about how the buzz surrounding Anne Hathaway’s Oscar was influencing Berkshire Hathaway shares. Although the effects of it were smaller and the methodology behind it is slightly different (sentiment driven versus keyword-pair driven), I believe this was already a perfect example how signals from social media can be misinterpreted by a computer.
2) In a more general sense, history has proven some of the shortcomings of algorithmic trading. Both the flash crash and the above example happened because of shortcomings in computational logic, not human logic. A good example of this were major news outlets like CNN and CNBC who did in fact check their source and quickly learned that the AP tweet was fake.
I think that the history of algorithmic trading is a main reason why the Twitter flash crash is most definitely not a black swan. After all, I believe there’s no false rationalization here that algorithmic trading involves specific risks. Of course most of the time systematic trading strategies will do well and outperform the market. And of course proponents of algorithmic trading will say that it provides more liquidity to the markets.
Until it doesn’t.
Maybe it’s because I am only an outsider on the subject. But when I’m reading the many comments criticizing Twitter after the crash, I can’t help but wonder: Is sending out a fake tweet really that much worse than creating false liquidity by firing off fake bid / ask quotes all day long?
I’m not so sure. I believe that as long as we allow computer systems access to financial markets, there will always be vulnerabilities to extreme events like this. But don’t tell me that nobody could have seen it coming. I believe this crash was no different than in May 2010. Or on the day of the Facebook IPO…
Image credit: Nevit Dilmen (Wikimedia Creative Commons)