Pragma attempts to shed light on the mysteries of the FX flash crash

While flash crashes are not exclusive to the FX market, the lack of any centralised market data in this market adds a layer of opacity to an already mysterious phenomenon. Some less liquid currency pairs have experienced flash crashes that have been barely commented on, notes Pragma.

The firm believes if the market can coalesce around a single definition it will encourage more comparative studies and lead to a greater understanding of why they occur and perhaps how they can be prevented.

Pragma defines a flash crash as an instance where volatility spikes to 13x normal levels, with a 70% reversion, occurring at twice the speed of average market moves. It published a study that examines FX market moves in 2015 and 2016 across a range of currency pairs, identifying dozens of instances of flash crashes per year based on its proposed definition. 

Curtis Pfeiffer, chief business officer at Pragma Securities, says: “Given what a liquid and generally well-functioning market FX is, we were quite surprised by the sheer number of flash crashes we observed. We also noticed they are often not discreet events - there is evidence of flash crashes in one pair triggering similar moves in other pairs.”

There are differences of opinion about how exclusive the definition for flash crashes should be. Pragma believes the number of crashes its definition identifies is a strength. “Because it identifies dozens rather than a small handful of events per year, it escapes the realm of anecdote and provides a useful statistic,” it says in its report.

However, others argue that lowering the bar to include so many events risks trivialising the phenomenon, and making flash crashes harder to distinguish from mere upticks in volatility.

Brad Bailey, research director in the securities and investments group at Celent, says: “‘Flash crash’ is an overused term. If you strip out a few specific events, such as the Brexit vote, the [Swiss National Bank] peg removal in 2015 and maybe the period around the US election, volatility has been very low and when you get used to that, any spike in volatility feels like a flash crash.”

David Clark, chairman of the Wholesale Markets Brokers Association, also emphasises the distinction between a flash crash and other forms of market dislocation.

Clark says: “Recent flash crashes do not compare with the market gapping experienced in former times, which were usually caused by major political events and were generally unexpected. They usually resulted in a significant rebasing of a currency such as the dollar-deutschemark after the Louvre and Plaza accords. This occurred with sterling after the Brexit vote – but that was not seen as a flash crash.”

Clark argues flash crashes are characterised by being technical in nature, typically caused by systems or transactions problems, rather than by a market response to a political or other news event. The problem then manifests as sudden illiquidity and loss of depth of book.

Put another way, flash crashes are a form of operational risk, as distinct from bouts of volatility that derive from market risk, which means the latter do not qualify as flash crashes.

But flash crashes almost certainly have numerous potential causes. Bailey cites many contributing factors, including the fragmentation of FX liquidity, non-bank electronic market making and the rise of algos. Time zones also appear to play a role.

“When the price moves away from these market makers they pull back from the market, leaving these air pockets of illiquidity – and volatility. These look particularly dramatic because most of the time volatility in FX is so low,” says Bailey.

The role played by algos is particularly controversial. Russell Dinnage, lead consultant at GreySpark, says: “A lot of algos are well run and well maintained, but others are not. It really depends on who is responsible for managing the machine, ultimately it is still humans that are responsible.”

"However, it is how algos respond to other algos that can become unpredictable and potentially disruptive," says Dinnage. When one machine malfunctions, all the other algos in the market react to it in concert, and the speed at which it happens makes it difficult to control. That makes it more important than ever that traders have the right pre-trade risk controls in place, to ensure they have the necessary kill switches so they can react quickly,” he adds.

Dinnage notes that flash crashes are more common in low volatility markets – something that is borne out by the number Pragma identifies in the relatively subdued FX market in recent years. “Algos are always hunting for inefficiencies and discrepancies. Even if there is nothing out there, they are still looking. Flash crashes often occur when machines detect some random event and interpret it as a meaningful macroeconomic move,” he says.

However, Pragma believes the growth in algo-use is a red herring in terms of explaining the occurrence of flash crashes.

Pfeiffer says: “We do not see any evidence that flash crashes are worse now than they have been in the past, it is just we talk about it more now. In the days when traders had to call their broker to get a price, five calls would have generated five different prices. The fact there is more electronic trading now than in the past means market data is better recorded and analysed than it was 10-15 years ago. But algos do not cause flash crashes. In fact, some algos allow traders to spread orders over time, which actually minimises the impact of flash crashes.”

Bailey believes that while there are likely many complex and interrelated factors contributing to instances of flash crashes, overall they are best understood as a symptom of the dramatic change – particularly technological change – experienced in the market in recent years.

“These kinds of market dislocations are more likely during periods of transition, like we have today in terms of market technology and capital scarcity, and the sheer growth of FX trading in recent years,” says Bailey. “But over time these changes are creating a better, fairer market.”

All agree that flash crashes warrant further study, and that markets will be better off if and when their causes are better understood – especially if that leads to them becoming easier to predict. Better understanding their causes will help ensure the market can develop next generation risk controls to manage the growth of next generation algos and new forms of electronic trading.

Dinnage says: “It is really important we study and understand the implications of algo trading so we can better prepare ourselves for the future, when quantum computing will take this to the next level. The vastly greater analytical capacity and highly interconnected AI-driven trading will lead to strategies that humans today would never be able to dream up, and while that is very exciting, it also means new types of risks for client money.”