London/New York: The increasing frequency of flash crashes in the $5.1 trillion-a-day (£4 trillion) foreign exchange market has regulators scrambling for answers.
Sudden, violent and often quickly reversed price moves are now a regular occurrence in world currency markets - often during the so-called 'witching hour', a period of thin trading between 5-6 pm in New York when currency dealers there have powered off and colleagues in Tokyo have yet to sign on.
Two big crashes this year separately pummelled the yen and the Swiss franc and, given the importance of currency pricing for trade, investment flows and the global economy, policymakers are concerned a major fracturing could threaten financial stability.
"The question is, is this a new normal, or is it a canary in the coalmine sort of thing?" said Fabio Natalucci, Deputy Director of the Monetary and Capital Markets Department at the International Monetary Fund (IMF).
"We have seen the frequency of these events increase so this may be pointing to a major liquidity stress event coming at some point in the future."
Natalucci said liquidity strains - market lingo for an insufficient number of buy and sell orders - were evident days ahead of a big crash and the IMF was creating a monitoring tool that might be able to predict when the next one was coming.
Reflecting official disquiet, flash crashes have been a regular topic of discussion this year at the Federal Reserve Bank of New York's FX market liaison committee, a forum for central bankers and market players.
Bankers and policymakers agree that an industry-wide switch to machine-trading in FX markets is behind the frequency and severity of the price moves, meaning that further crashes are likely.
"Our pessimistic view is that this technology is going to become an increasing part of the FX market and we need to step up our monitoring," a G10 central bank official said, declining to be named because he is not authorised to speak publicly.
Regulators aren't pressing the panic button yet. Natalucci said there was no evidence that flash crashes so far had raised funding costs for firms or households and it made sense to study the problem before "rushing into enacting any regulatory responses".
Mini-crashes already occur roughly every two weeks in the FX market according to a study by Pragma, a company which creates computer trading models. In these incidences, a currency's price will shift dramatically followed by a swift reversal, along with a sudden and significant widening of the spread between prices quoted to buy and sell it.
The spread usually narrows after a few minutes.
Computer models known as algorithms, or algos, have largely replaced humans in currency trading, helping banks to cut costs and boost the speed at which deals are done.
The models are designed to execute trades smoothly by breaking down orders into small pieces and searching for platforms where liquidity is plentiful.
But problems arise when market conditions change, for instance, when trading volumes suddenly collapse or volatility spikes as has been the during Britain's protracted attempt to extricate itself from the European Union. At such times, algos are often programmed to shut down.
Two senior central banking officials, speaking on condition of anonymity, said such "kill switches" drained liquidity.
And, because fragmented forex markets depend on algos for a constant stream of price quotes - by one estimate there are 70-odd trading platforms - a widespread shutdown causes volumes to nosedive, making the price moves more dramatic.
The first of this year's notable crashes came on January 3 when the yen spiked suddenly against the dollar after Tokyo markets closed. It jumped 8 per cent within the space of seven minutes against the Australian dollar and 10 per cent to the Turkish lira.
The second was on February 11 when the Swiss franc gyrated frantically, with an unexplained and brief jump against the euro and dollar.
A Reserve Bank of Australia (RBA) report noted that several flash episodes have been recorded during the witching hour. It was also during this illiquid period on October 7, 2016 that sterling collapsed 9 per cent in early Asian trading, falling to around $1.14 from $1.26 within minutes.
The RBA's analysis of all these flash crashes concluded algorithmic trading strategies likely acted as "amplifiers".
Human traders would be able to spot an opportunity from the market turmoil - buying a currency in free fall - which would help to defuse it. But these days there are far less of them around.
Upto 70 per cent of all FX orders on platform EBS, one of two top venues for currency trading, now originate from algorithms. In 2004, all trading was undertaken by humans.
With banks under constant pressure to cut costs and post-financial-crisis rules making it ever more expensive to trade, there is no sign of firms hiring extra staff or deploying existing employees onto a graveyard shift.
Instead, some try to avoid trading when they know volumes will be light such as major holidays.
Machines, meanwhile, are expected to become even more dominant.
Pragma has just launched an algorithm to trade non-deliverable forwards, derivatives used to hedge exposure to illiquid currencies, especially in emerging markets, according to Curtis Pfeiffer, Chief Business Officer at the firm.
Trading in illiquid, emerging market currencies was previously the mainstay of voice traders.
"FX trading in banks is a tough business because spot trading is so commoditized and revenues are squeezed," said John Marley, a Senior Currency Consultant at Smart Currency Business.