How can new data-driven approaches to risk management help trading firms better understand black swan events such as the global pandemic?

The idea behind any quantitative or algorithmic trading is that there’s certain patterns that repeat themselves, and if we can train our algorithms to identify those patterns and their indicators then we are better prepared to take advantage. The challenge with the current pandemic is that this is not a pattern that we’ve seen for a very long time. It is not something we have trained our algorithms for, and it’s not something we would expect those algorithms to know how to deal with.

So in that sense, COVID-19 is very much a black swan event. Nonetheless, we have to be careful not to rethink our trading algorithms for things that will happen every 30-40 years because that’s not how these things work. There won’t necessarily be enough data points to train the models. The markets might react differently each time. Circumstances will change.

What we can do is look at how the algorithms have performed in terms of risk management. And we can learn from that. It’s not about preparing for exactly the same event again, but making sure that our systems have the ability to adjust and respond to disruption. When these crazy events happen and there’s huge movement in the markets, this is the time where you have to be more systematic in the way you react. And that is especially true when it comes to risk management.

A good trader follows good rules, and a good fund has a good risk management strategy. Essentially, you’re trying to take out emotion. Emotions are good when you are thinking creatively, when you’re trying to design a system, when you are coming up with new ideas. But when you’re executing those systems, you want to be like a machine. Because firms that perform least well are those that panic. Panic can often cost a fund more than the hit they take in the first couple of days of a crisis – they close the position, they lose the rebound, they don’t follow the rules. Computers aren’t subject to those kind of emotions: they are good at executing rules and not panicking.

They are also very good at executing in real-time, which allows trading firms to take advantage of techniques such as ‘nowcasting’: the prediction of the present and the very near future. Trading is a game of trying to see what’s coming next, so in principle any kind of information can contribute to that effort. Both traditional datasets and so-called ‘alternative data’ – social media sentiment, analysis of consumer trends and behaviours, and other niche information beyond standard financial market indicators or statistical releases – are invaluable in feeding algorithms and making sense of volatile and uncertain conditions.

Of course, that presents a significant data management challenge, and this is where technology has a big role to play. The fact that we have storage capacity that we didn’t have before, the computing power to analyse all that data, and the algorithms that come with that offers a huge advantage. Whether you’re talking about the neural net, machine learning or traditional trading algorithms, they’re only able to run because of the computing power that is available to analyse the data in real-time.

You still need a strong underlying trading strategy – for example, something like momentum-type trading or mean reversion, statistical arbitrage – to provide a sound base. All these strategies still have a place. But when you combine those with the risk management capabilities that approaches such as nowcasting offer, that’s when you start to see real benefits. Bringing those elements together and training your systems to react appropriately based on known data offers huge advantages.

It’s like sailing your boat through a really terrible storm. In an ideal world you wouldn’t do it; you would look at the forecasts, look at the data, and plan accordingly. But sometimes you get caught. And as long as you make it back to port, that experience can be a great learning curve. You now know what works in extreme conditions and what does not. You can go and fix the bits of the boat that broke. And you have more information to work from in order to be better prepared for when the next big storm hits.

Nir Vulkan is an Associate Professor of Economics at the Said Business School and the Oxford Man Institute. He is also a Fellow of Worcester College. His research focuses on fintech, algorithmic trading and AI in finance. He is the Director of the Oxford Programmes on Algorithmic Trading, Fintech and on Blockchain Strategy, and in addition to his academic work, he developed algorithms that are used widely by hedge funds and ecommerce platforms today.