Brief description of Flash Crash and the purpose of this demonstration.

The reason for monitoring autocorrelation of equity time series.

March 2010: Evidence of a raging storm.

May 6, 2010: Landfall and subsequent return to normal.

Multi-year chart and its prominent points.

Flash Crash is the enormous intra-day market move that took place on Thursday, May 6, 2010. In the afternoon of that day equity prices in the U.S. were dropping with unprecedented speed. Before the day ended, the plunge was followed by just as rapid recovery. Overall, the first half of May 2010 was rather eventful for equity markets. The next day after the Flash Crash, Friday, May 7, 2010, saw broad declines in equity prices associated with the European debt crisis. Monday, May 10, 2010 saw one of the biggest rallies in years in response to the measures announced by the European governments intended to contain the debt crisis.

Is the Flash Crash really an intra-day event? Has it been brought about by trading activities of certain individuals on the day of the crash or on the days leading to the crash, in April and May 2010? We'll try to answer these questions by taking publicly available end-of-day prices of a single large bank, JP Morgan Chase, and showing that the evidence of abnormal market conditions could be seen in that data starting in March 2010.

If the abnormal market conditions that had been observed starting in March 2010 could be compared to a hurricane, the Flash Crash on May 6, 2010 would be the day of the hurricane's landfall. After that, the calm returned.

Is the stock price more likely to move up than down today if it moved down yesterday? If we assume that equity price reflects all available news up to the current time, there should be no material correlation between yesterday's and today's price moves. That, of course, would not be true had a piece of good news been likely to be followed by another piece of good news, for example.

Correlation of a time series to a copy of itself shifted by 1 day is called "autocorrelation with one day lag." For actively trading stocks under normal market conditions the autocorrelation should be close to zero. Material deviations from zero are usually due to data errors. It is therefore natural to monitor autocorrelations of equity prices as a data quality check.

It is customary to report autocorrelation every day by calculating it over 1-year period ending on that day. The points on the chart below are autocorrelations of JP Morgan Chase daily closing equity prices (JPM Equity), calculated over the last 264 business days.

In the second half of March, 2010 the autocorrelation had become alarmingly large. By March 24, 2010, the autocorrelation crossed -20%, and by the end of the month, -30%. The negative sign of the autocorrelation meant that the daily closing prices were moving in an oscillating fashion: up and down, up and down.

If we now expand the time frame of the chart by including the day of the Flash Crash and beyond, we can see that the worst point was the day of the Flash Crash, May 6, 2010. After that, the conditions returned to normal.

Interestingly, the next prominent point on the chart corresponds to another major market event: the downgrade of the U.S. debt on August 8, 2011. Unlike the Flash Crash, the U.S. downgrade, according to the chart, was an unexpected, sudden event.

By further expanding the time frame, we see that the third prominent point on the chart is yet another important market event: the Swiss franc de-peg from the Euro on January 15, 2015.

It appears, therefore, that the publicly available market data contained sufficient information to anticipate the coming May 6, 2010 Flash Crash. The evidence of abnormal market conditions was there since the second half of March, 2010.