Analysis of the technical causes of the fall of the NYSE on May 6, 2010

    Since 2004, the American company Nanex has been processing information flows from stock exchanges in near real-time mode (they sell the NxCore system through which any company can receive a common data feed and use a simple API for its own manipulations). For six years, the company has accumulated a base of approximately 2.5 trillion quotas, including 7.6 billion transactions on the “tragic” day of May 6, 2010.

    A week ago, Nanex published on its website a detailed analysis of the events of May 6 from a technical point of view. They give their explanation of the reasons that could cause the Dow Jones index to fall by 600 points (5.7%) in just four minutes: from 14:42:46 to 14:47:02.

    Let's try to figure out what happened.

    There are nine trading floors that handle NYSE transactions: these are NYSE, Nasdaq, ISE, BATS, Boston, Cincinnati (National Stock Exchange), CBOE, ARCA and Chicago. All of them generate their purchase and sale applications, and the best applications are assigned the status of National Best Bid (best selling price) and National Best Ask (best purchase price) - they go for execution.

    Exchange systems constantly monitor each other for price compliance so that their own application for a purchase or sale does not go beyond the National Best Bid / Ask, otherwise high-speed trading systems will instantly intercept it within a few milliseconds.

    The fact is that on May 6, 2010, starting at 14:42:46, the NYSE started broadcasting a stream of sales orders at prices that were lower than National Best Ask immediately for 100 securities, and two minutes later for 250 securities ( see graphs ). The number of such “incorrect applications with the NYSE is shown in green.



    Prices immediately fell slightly, but it was only a starting trigger for further events.

    Naturally, high-speed trading systems instantly grabbed the “bait” and started working to the fullest. At the same time, they began to use special technology to “litter” the market with unnecessary applications in order to complicate the life of competitors. The technology is called “quote stuffing”and provides for the generation of up to 5000 applications per second for each paper with slightly different prices - a change in price five times per millisecond confuses any system. For 4,000 traded securities at 9 sites, this theoretically gives a flow of 180 million applications per second (at least 58 bytes each). Among other things, even T1 fiber can overwhelm such a stream.

    The flow of applications was so large that the system simply did not have time to process them on time - as a result, quotes on the NYSE site began to lag a few milliseconds from quotes on other sites. This caused a massive migration of applications from all sites on the NYSE, because at each point in time prices here were slightly higher than at other sites where they had already fallen slightly. NYSE quotes lag shown in these charts. In fact, at other sites, the sale of securities ceased completely, because all requests for sale were sent to the NYSE. But when they came here, here the purchase price was already de facto lower, because the previous orders placed in the queue were sent for execution.

    Nanex experts show that similar cases, albeit on a smaller scale, have already occurred on the stock exchange on October 30, 2009 and January 28, 2010.

    In order to prevent cases when automatic trading systems completely flood the exchange, Nanex offers several protective measures: prohibit “quote stuffing”, put time-stamps on orders at the time of generating orders (to detect cases when the site is behind the market) and set the order lifetime to 50 milliseconds.

    Also popular now: