This content was sponsored by Deutsche Bank
Several high profile flash crash episodes have been reported in recent years. Their frequent occurrence has raised concerns around the stability of the market, and has attracted attention from both regulators and market participants alike. In this article we highlight two of DB’s market-leading research pieces in this sphere.
The first develops a methodology to systematically identify flash crashes while the second investigates the dynamics of the trade and order books around three recent, high profile flash crash episodes in FX. We draw two main conclusions from our research. Firstly, flash crashes are not market glitches but rather are regular stylised facts across all markets. Secondly, contrary to popular perception trading conditions are more active and liquid in the run up to a flash crash, not less. Flash crashes are not simply the outcome of market illiquidity, but a more complex phenomenon where a bout of unusually high market activity depletes liquidity at a time of day that it cannot easily be replenished.
Our first piece (Christensen et al., 2018) makes an important contribution in an area where there isn’t yet an academic consensus – what defines a flash crash? Using a comprehensive set of tick data, we develop a simple measure that systematically identifies flash crashes by using unusually strong short-term and one-directional price movements. This measure can then be applied historically across FX, rates, equities, and commodities markets…
Our second piece investigates the dynamics of a few high profile flash crash episodes in FX (Grover et al., 2019). It includes the study of flash crashes in EUR (2015), GBP (2016) and JPY (2019). Below we discuss the latter crash in detail, with the other two events following a similar pattern.
On January 2nd between 22:35:00 and 22:39:10 GMT, USD/JPY fell over 3% from 108.18 to 104.92, with the bulk of this move happening in a 25-second period between 22:35:45 and 22:36:10 despite an absence of material news. Stretched short-positioning was likely one factor behind this flash crash. In our report we focus on the broader liquidity conditions by using our high-frequency dataset to understand the trade and order book dynamics during the yen rally that preceded the flash crash, as well as around the crash itself. Our key findings can be summarised as follows:
Firstly, we found abnormally high levels of trading before the flash crash. In the preceding minutes there was an unusual rise in number of trades on the EBS platform. Given that Japanese markets were still closed and other markets had just opened after the holidays, we would have expected less-than-average market activity. But, in the 15 minute period before the crash there were a total of 367 trades, versus an average of just five trades at this time of the day in December 2018.
Second, liquidity from a volume perspective was actually very good before the crash, though quite onesided. Over $200 million was generally available across the 10 best bids between 22:15 and 22:35. Price ranges, measured by the spread between the best and the 10th best price, were also very tight for most of the pre-crash period. Overall, liquidity conditions compared favourably to those observed during news-driven rallies in the yen, though both price and volume liquidity did both dramatically deteriorate in the seconds before the crash.
Finally, we found large imbalances between the buy and sell side of the order book in the precrash period. Specifically, the 0.8% drop in USD/JPY between 22:15:22 and 22:35:44 was accompanied by a total selling pressure of $675 mln. A similar move on Brexit day, in contrast, witnessed a selling pressure of only $122 mln. This suggests merit in monitoring this measure to identify any sudden build up in buying/selling pressure that could preclude a flash crash.
To take advantage of our research, some of the flash crash indicators described above are now available to clients in real-time, via the Autobahn Mobile app.