Sterling’s Flash Crash was long overdue—and there will be many more

Researchers at Sapience.org foresee market instability intensifying by the computer trading ‘arms race’


Last Friday the sterling has experienced a dramatic, ultrafast crash. It lost 10% of its value in minutes after the Asian markets opened — a decline usually reserved to declarations of war, major earthquakes and global catastrophes — and bounced right back. Although the affected exchanges are yet to release the details, computer trading algorithms almost certainly played a key role. Just like the 2010 Flash Crash, yesterday’s event is characteristic to Ultrafast Extreme Events[1]: split-second spikes in trade caused by ever smarter algorithms razor-focused on making ever-quicker profits. But the arms race is only likely to intensify as computing speed accelerates and AI algorithms become more intelligent.

Stock exchanges have become war zones where algotraders compete over pennies millions of times a second

Few realise that high-speed trading has all but overtaken the world’s stock markets. Human floors trades have been replaced by the matching engine and human securities traders gave place to algotraders — lightning-fast computer trading algorithms. Investment banks and hedge funds pay hefty sums to connect their servers directly to the stock exchange’s computers. Algotraders hoover vast amount of data and execute over 50 buy/sell decisions per second, each. Transactions now largely appear to take place virtually instantaneously in human terms.

The result is that “Wall Street is no longer Wall Street”[2]. The ticker tape running at the bottom of the screen is an illusion. Rather, stock exchanges have become war zones where algotraders compete over pennies millions of times a second [3]. Stock exchanges argue that algotrading “improves liquidity”[4] but analysts have long been warning about the instability they cause.

Time is money and algotraders bring profit in the billions. In a bid to outpace their rivals, algotraders parse news attempting to foresee their effect on market prices. They attempt to react to news releases, twits, even blogs[5] within milliseconds, which can leads to trigger-happy tactics. On the 2013 anniversary of its 1973 October war the Israeli Defence Forces commemorated the 40 years event with a tweet commemorating the war. The innocent tweet jolted oil trading[6]. It is unlikely that human trader misinterpreted the tweet as a cause for panic, but such mistakes are typical to primitive Natural Language Processing algorithms— NLP being a rapidly advancing field of artificial intelligence. Friday’s Flash Crash followed news reports of the French president’s mildly negative statement about the British exit negotiations.

The SEC’s analysis[7] of the events leading to the 2010 flash crash in the NYSE shows a similar pattern of ultrafast spiral. Analysis shows that the 2010 crash was triggered by a single, large but not uncommon sale. The Asian exchanges have not released yet the details but experience proves that Friday’s event was triggered by very few — perhaps even a single algotrader — who ‘panicked’ by a minor event such as Hollande’s comment. Given its speed and intensity, it is even more likely that Friday’s downward spiral in the Sterling’s value was almost entirely the work of algotraders. Greedy and trigger-happy algotrader(s) evidently moved volume large enough to cause a chain reaction and an explosive positive feedback process of sell orders.

Exchanges experience an ever larger number of ultrafast spikes brought about by a ecology of ever faster and greedier algorithms

Firms seek to benefit using algotraders that predict their rival’s next move and profit from it milliseconds earlier. Since investment strategies are trade secrets, their tactics and how exactly algotraders attempt to outsmart each other is anyone’s guess. Regardless, the drive for an ever-quicker profit led an ever expanding jungle of selfish, competing computer processes evolving to act ever faster. A Nature article [1] explains how stock markets have become ‘ecologies’ where ‘dumb’ algotraders fall prey of predatory algotraders. But in the race to outbid each other, predators too fall prey to ever more cunning algotraders. Cleverer algotraders feed on less clever, trigger-happy algotraders, and so on. Stock exchanges globally have been taken over by a superfast ecology of rapidly evolving processes.

Stock markets are known to behave chaotically. But algotraders have led to the emergence of a new phenomenon: Superfast Extreme Events, meaning spikes and crashes which occur for no good reason and lasting under 1/50 of a second. Most often the market corrects itself as fast as a downward spiral is followed by an equally superfast rise. As they last only about 25 milliseconds, superfast events are virtually invisible to human investors, and firms are unlikely to disclose their trade secrets even if their a malfunction in their own algotraders cost them millions.

Exchanges experience an ever larger number of ultrafast extreme events. Occasionally, an unusual number of algotraders get sucked into the spiral to send entire markets haywire. But profits of high-speed firms such as Goldman Sachs and Bank of America are in the billions. And stock exchanges profit too by selling real-time access. Despite Michael Lewis’ warning that computer trading created “a class system rooted in speed” whose biggest players “you’ve never even heard of”, few stock markets attempted to make trading fairer or more stable.

Sapience.org sees more flash crashes in the next few decades as the computer trading ‘arms race’[8] will intensify. In the next few decades we expect AIs to get smarter, computers to get exponentially faster, and investment decisions being increasingly handed over to algorithms. More flash crashes are not unlikely. As financial markets become increasingly interdependent, instability may affect many more than before. Financial authorities would be wise to consider various measures for preventing flash crashes from destabilising stock markets.

Dr Sandberg of Oxford’s Future of Humanity Institute and a researcher in Sapience.org says: “Very fast processes can go haywire faster than humans can control them. We can put in circuit breakers, but these only respond to known problem outcomes and might hence be blindsided by emergent behaviour of new kinds. Hence there is always going to be a gap between human intention and controllability of fast, complex systems. This is sometimes acceptable, but there is a second gap in many fast domains between how fast the technology develops and how fast human discourse can reach the moral consensus to accept this: this forms a second, ethical gap.”

Dr Eden, principal at Sapience.org, adds: “There is financial and military incentive to delegate increasingly more important decisions to superfast machines. High-speed firms now oversee almost all stocks at NYSE[9]. US’s Department of Defence is funding the development of autonomous lethal weapons (“killer robots”) which will make superfast decisions to (literally) pull the trigger without human intervention.  As the algotrading arms race created ‘flash crashes’, the robotic arms race could lead to Flash Wars[10]. By handing power over to superfast processes we may lose control.”


[1]          N. Johnson et al., ‘Abrupt rise of new machine ecology beyond human response time’, Nat. Sci. Rep., vol. 3, Sep. 2013.

[2]          M. Lewis, Flash Boys. Penguin, 2014.

[3]          F. Salmon, ‘Chart of the day, HFT edition’, Reuters Blogs, 06-Aug-2012. .

[4]          T. Hendershott, C. M. Jones, and A. J. Menkveld, ‘Does Algorithmic Trading Improve Liquidity?’, J. Finance, vol. 66, no. 1, pp. 1–33, Feb. 2011.

[5]          Bloomberg, ‘Ex-Goldman Banker Starts Hedge Fund Analyzing Japanese Blogs’, Bloomberg.com, 21-Apr-2011.

[6]          Reuters India, ‘Tweet recalling Yom Kippur war, 40 years on, jolts oil traders’, Reuters India, 10-Oct-2013.

[7]          CFTC and SEC, ‘Findings Regarding the Market Events of May 6, 2010’, Sep. 2010.

[8]          A. Sandberg and A. H. Eden, ‘There is Plenty of Time at the bottom: ???’ 08-Oct-2016.

[9]          Bloomberg Business, ‘High-Speed Firms Now Oversee Almost All Stocks at NYSE Floor’, Bloomberg Business, 26-Jan-2016.

[10]        P. Scharre, ‘Robotics on the Battlefield Part II: The Coming Swarm’, Center for a New American Security, 2014.


Related posts:

Five More Lessons About Documentation: A follow-up to Mark Birch’s “Developer Documentation”

Last week’s DevBizOps blog entry (“Developer Documentation: Developers don’t like writing docs, what’s the alternative?“) asked: How programmers can get their answers from documentation, and are there alternatives? As ever Mark’s post advises on questions every developer has to ask in understanding software. Open source or proprietary, documentation is necessary either for using, extending, or changing software. The costs of searching for answers are known to developers and project managers. Empirical literature and decades of research in software comprehension and reverse engineering show that the costs of understanding software are significant. What, then, can be done? We offer five more lessons in addition to Mark’s post, presented as “formulas”:

Can AI be creative?

Last month Deep Mind’s latest AI broke another record, winning 5-0 against one of the world’s strongest professional players. This victory forces us to revisit our preconceptions and recognise how creative machine learning has become.

1 Comment

  1. “I certainly agree that there are many more to come. The fact that people are still arguing about what exactly caused the Flash Crash of 2010, makes this discussion about ultrafast instabilities very important.”

    — Neil Johnson, Professor of Physics, Miami University, author: ‘Abrupt rise of new machine ecology beyond human response time’

Leave a Reply