Pragma Securities has released a new report that analyses FX spot market data in order to provide a more accurate definition of exactly what constitutes a “flash crash”.
Arguing that previous reports on flash crashes “have tended to look at individual events in isolation” and that “discussion of the recent trend at industry events has been correspondingly anecdotal”, the paper attempts to define what a flash crash is and then systematically track the incidences of flash crashes using this definition.
“No one has really looked at this from a systematic viewpoint before,” says Curtis Pfeiffer, chief business officer at Pragma. “What we wanted to see was if flash crashes were occurring more frequently, and whether they are always news related or whether they sometimes occur when there’s no news at all. So the value of this report lies in the fact that we’ve proposed a definition of what a flash crash is that we hope will allow market participants and regulators to use systematic data to look at the underlying market and see whether these types of events are occurring more or less based on changes in technology, regulation or industry practices.”
The core data set for the analysis is the tick-by-tick quotes of the more liquid currencies – AUD, EUR, GBP, CAD, CHF and JPY – across 2015 and 2016.
Explaining the methodology used to analyse the data, the report says: “From the tick data, we compute five-minute bars of high, low, open and close mid prices, number of quote updates, etc. Bar volatility of the close-to-close return is computed specific to the pair and hour of the day, as volatility varies over the course of the day in a characteristic pattern. Characteristic volatility is estimated over the past seven days.
“Average spreads within a bar are time-weighted. Finally, we filter out data in the 4pm-7pm New York time, to avoid the disproportionate number of data problems occurring during this time frame, for example, resulting from market participants’ daily operational cycle.”
Examining this data, Pragma identifies three main criteria that define a flash crash: a large price movement in a very short amount of time, a significant and quick reversion of that movement and a widening out of the bid-offer spread.
Specifically, the report concludes that to be a flash crash, there has to be a price movement that is more than 13 times the normal volatility, followed by a price reversion of more than 70% and that spreads must be at least twice as wide as normal.
The Pragma report found that, out of more than 313,000 possible flash-type events over the two-year period being studied, only 69 – or 0.02% – met its definition of a flash crash. The report also indicated that there was no clear trend – up or down – in the incidence of flash events over this period of time.
It is worth noting that the specific thresholds used to define a flash are somewhat arbitrary, something that the report openly acknowledges. However, Pfeiffer emphasises the criteria being looked at are not.
“You could say that the price reversion has to be 75% instead, or that the volatility actually needs to be 15 times normal, and obviously tweaking those numbers would have an impact on whether more or less events qualify as a flash crash. But our goal was to try and provide a good starting point for analysing these events,” he says.
The essential point made here by the report is that “these criteria do not define a discrete cluster of events, but rather a continuous range of extreme events”.
Looking at “Figure 7” from the report, the blue dots in the upper right quadrant represent market events that meet the report’s definition of a flash crash. Should a market participant wish to use different thresholds for each criteria, the dotted lines can be moved and the spread requirement can be changed, meaning that the number of blue dots in that top right quadrant would change, but the underlying picture would remain the same.
Figure 7: Extreme price move vs reversion. Blue indicates spread more than 2× the mean; Red indicates less than 2×. The vertical dotted line is at price move of 13× volatility. The horizontal dotted line is at 70% reversion
The report claims that the value of the definition proposed by this research is that it can “provide a useful quantitative index for regulators and industry participants interested in understanding how ongoing changes to technology, regulation and industry practices affect market quality over time”.
Or as Pfeiffer puts it: “This research report takes a systematic approach to try and put a stake in the ground that market participants can use to measure market quality going forward.”