On the 15th January 2015, the euro crashed 20% against the Swiss franc in a matter of moments, before recovering rapidly. Similarly, on 7th October 2016, sterling plunged in value by over 9% against the dollar, again regaining most of its value minutes later.
These are amongst the most famous examples of the market phenomena know as the ‘flash crash’, but they are by no means the only examples. In fact, according to a study by algorithmic trading technology provider, Pragma, which aims to help monitor and track the prevalence of flash crashes, there were some 69 flash crashes in 2015 and 2016. Almost one a fortnight.
The causes of these market phenomena are unknown. It has been suggested that flash crashes are the result of ‘fat-fingered traders’ or lapses of human judgement. After the pound sterling incident, the Bank of International Settlements (BIS) released a report which suggested technical error as a possible cause.
However, as most research has considered these events as of one-off incidences, drawing generalised conclusions has been difficult. Without other flash crashes to compare, it is not possible to tell which variables in a complex market are contributing to the crash and which were incidental. For example, some commentators have suggested that a principal cause is algorithms overreacting to news events, but further study has found no particular correlation between other flash crashes and news events.
This is where Pragma’s research is vital. It has analysed two years of tick by tick foreign exchange data to identify and catalogue all instances of flash crashes across numerous major currencies between 2015 and 2016. To do this, it has developed a precise, quantitative definition of the flash crash.
Previously, the BIS described a flash crash as a ‘large, fast, V-shaped price move and a sudden widening of bid-offer spreads,’ the V-shape implies a reversion of the price after the initial price move. Pragma’s definition builds on the BIS’s and defines a flash crash as having a:
- Large price move ( 13x than normal price volatility)
- Widening bid-offer spread 2x normal)
- Strong price reversion ( 70% price reversion)
Using this standard, the examined time period had 69 instances of what would be considered a flash crash.
This dataset allows industry analysts and academics to more accurately examine the causes of flash crashes and what effects such as changing technology, regulation and industry practices are having on market quality going into the future.
For now, the causes of flash crashes remain unclear. But Pragma’s research provides an important foundational step in moving the market towards a more full understanding of this market phenomena.