Faith in the machine

The coronavirus crisis could be a defining moment for machine learning in finance

The machines have had their ups and downs of late. 

Fraud detection systems that rely on machine learning have not coped at all well with the turbulence caused by the coronavirus lockdowns. But other automated systems are having a comparatively good war.

Contrary to expectations, algorithmic execution is seeing a boom. Ralf Donner, head of fixed income, currencies and commodities execution solutions at Goldman Sachs, estimates a 50% increase in algo foreign exchange volumes since markets began to crash in March. Customers grew comfortable with algorithmic execution during quieter times, he speculates, and shifted more volume their way during a crisis to free up more time to cover the more volatile equity and commodity markets.

And machine learning has had some notable successes in the fund management sector. Not every fund using machine learning methods has produced good returns, but an index of the sector shows decent performance, and some managers report that machine learning has made a significant – and lucrative – difference.

In particular, the machine learning advantage showed itself early on in the pandemic, spotting the first signs of pandemic risk and recommending derisking or even a stop to trading.

Some have turned the machines on the virus itself. Quants from various parts of the financial industry are repurposing models designed for financial analysis and options pricing to simulate the spread of the virus and the effectiveness of policy responses. And while the goal of many is to predict the market response, some believe their work could help shape policy responses as the crisis continues. 

The crisis was also a reminder that models are only as good as the data that goes into them. When faced with an unprecedented event such as the coronavirus crisis, models based on historical data are of limited use. To address this, some quants are experimenting with machine learning techniques to create ‘synthetic’ datasets. These so-called market generators can spit out vast quantities of realistic but fictional data, which could be used to train the machines of the future.

The combination of machine learning and unlimited data can be a powerful one. If the pandemic serves as a stress test to cull the worst-performing machine learning applications and move the state of the art on, 2021 could be a revolutionary year for finance.