| 3.9°C Dublin

Robo-traders prove less prone to panic but fears of flash crashes persist


Events such as the coronavirus in China haven’t spooked the markets to the same extent. Photo: AFP via Getty Images

Events such as the coronavirus in China haven’t spooked the markets to the same extent. Photo: AFP via Getty Images

AFP via Getty Images

Events such as the coronavirus in China haven’t spooked the markets to the same extent. Photo: AFP via Getty Images

Widely blamed for volatile 'flash crashes' in currencies and equities, high-frequency algorithms may also be why shock global events, including the current coronavirus, seem to have lost their power to spook markets for any length of time.

Whether stocks, bonds, currencies or commodities, asset prices seem less prone to any sell-off for very long. The US killing of an Iranian general and Iran's retaliatory missile attack are among potential catastrophes that triggered violent but surprisingly short-lived reactions.

Now, even as China's coronavirus threatens to throttle economic growth, global stocks are not far off all-time highs.

Certainly, many factors are shaping the resilience, not least central bank money-printing and rising global savings, which boosted the value of world stocks by $25trn (€23trn) in the past decade. Yet it is hard not to link the shift in reaction by financial markets to the rise of automated trading strategies.

In the past six years, the share of algo-trading in the $6.6trn-a-day FX market has more than doubled to 27pc among fund managers, a survey by Greenwich Associates found.

There is some reason to believe algos cause volatility, especially when trading thins and the humans overseeing them vanish, for instance during public holidays. That is what likely happened during the Wall Street flash crash of 2010 and dramatic but fleeting yen moves in 2019.

But they also offer the advantage of being able to transact at lightning speed at any hour of the day or night, with razor-sharp accuracy and lower overall costs. Being machines, they are also alien to the common human impulses of fear and greed.

Algorithmic trading is dispassionate, said Scott Wacker, global head of fixed income, currency and commodity e-sales at JP Morgan.

"As a result, the reaction function in currency markets to even major geopolitical news has considerably shortened which enables stability to return more quickly," he said.

In short, when left-field events hit, not only can algos scan and react swiftly to news, but many now can gauge the potential asset price impact. The most sophisticated can be 'trained' to learn from the experience before the next shock.

One currency trader familiar with algo use said a machine reading coronavirus cases would typically buy stocks if informed of "500 new cases, 10 deaths". "If it's '3,000 new cases, 200 deaths', they might sell. The point being that as soon as a headline is out, the machine-led market is trading on it," the trader said.

But the machines have 'vol triggers', he said, meaning they can stop trading when the market moves beyond specified limits.

Algos can now be hooked up to sophisticated language processing technology, to "read and analyse" news feeds, then react accordingly, all in the space of seconds, said Antony Foster, head of G10 FX trading at Nomura.

However, this can "lead to over-reaction in the first instance", Mr Foster warned.

The impact in fast-moving markets can be outsized if the models rapidly push prices toward existing buy/sell order levels, trip them and trigger other orders.

That is what happened when news broke of Iran's January attack, according to a quant fund manager, who said algos had bought yen with the aim of triggering larger buy orders once a key option barrier was tripped.

A plausible comparison may be the 0.6pc plunge in the S&P 500 within the space of half an hour on January 29.

The move came after American Airlines and Lufthansa said they were suspending their China flights, but an hour later, the losses had been recouped.

Stephane Malrait, ING Bank's head of market structure and innovation, said in such instances, algos are programmed to check if moves are in line with price trends.

If the swings are in response to an incident, human traders can step in to smooth out the trade, Mr Malrait added.

Next, the algos may gauge the seriousness of the incident based on patterns of investor behaviour and economic consequences that followed previous such episodes.

After the Iran missile attacks, for instance, the message from the specialised data-crunchers to the algos was: stand down.