The Last Look…

There has been a growing, but still relatively subdued, debate in FX circles about the value – or otherwise – of AI in the trading function; we all understand its value in workflow, but are proponents being handed a lesson about its limitations in the current trading conditions?

I need to preface this piece by acknowledging that a systematic trading firm may not be using AI at all, but at a high level, what I want to discuss is the value of an automated decision system through the prism of AI, and its potential. I don’t think anyone can doubt the value of AI (or machine learning) in the pricing function, it seems very good at identifying micro trends, thus helping the pricing engine do what manual traders often had to guess at – where to skew.

When it comes to reading the market for anything longer that a few milliseconds, however, it is clearly a very different matter, especially in highly volatile conditions. While I have not been surprised by the stream of results from banks (and other LPs), showing they benefited from Q4 2024 volatility (and will, no doubt, be feeling it even more in the last month or two), I was a little surprised at the gulf in performance between systematic and discretionary hedge funds.

We reported this week that the HFR indices indicated that the gulf is growing, namely systematic macro hedge funds were underperforming their discretionary brethren by 8% year-to-date. In the world of CTAs it’s similar – the BarclayHedge CTA indices show that discretionary traders are approaching 4.7% better off after March than the systematic group.

Whichever way you look at it, systematic just isn’t handling the conditions too well at the moment, probably because they are too data driven. This is an event-driven market, if you wait for the data to tell you what to do, you’re a day late and a few million bucks short to paraphrase. A lot of systematic firms are also trend following, which means they don’t tend to be over-reactive – this works well in other conditions, but when the swings are large, they clearly suffer.

So what does this bode for AI-driven trading? The strength of the technology is that it learns, and as such, it may well learn to react better in conditions such as we are currently witnessing. The downside is that by its very nature, AI-trading has to be data-driven, and as such, the current woes from the systematic sector are a decent proxy for how things would go.

It is important to note, of course, that these conditions won’t last forever…probably!…but they are likely to be around for quite some time, prompting another paraphrase – the volatility could stay lethal longer than many systematic traders can stay solvent.

What is needed, therefore, is pure old-fashioned common sense – if you have the ability then you have to deploy different strategies for different conditions. There are market regimes where discretionary traders suffer because they tend to over-think things, what we have seen for the last six months or so, is how systematic traders suffer with extreme swings – the solution should surely be a blend?

This raises the issue for investors of managers jumping outside their box – style drift is something few allocators are happy with because it breaks up their data-driven strategy – but for managers it probably signals the need for a more flexible, or defensive, approach.

It will be fascinating to see results from the relatively few AI-driven investment managers (many jumped on the publicity bandwagon, few actually have deployed it), especially over a period of time. If nothing else, this will tell us whether the uber-intelligence is clever enough to recognise it needs to pull its horns in and live to fight another day, or if it will simply keep plugging away (and losing money) until the market regime changes?

Investors are unlikely to be that patient, which leaves a question mark of the near-term future of AI as a trading mechanism. It also highlights one of the market structure changes that have occurred over a long period of time, especially in FX, in that the price making and decision-making functions, which were historically the domain of one person (or machine), are now very much being separated.

I still don’t think AI will dramatically change how people trade markets, but it is clearly changing how they price them, and how they manage all the ancillary functions pre-and post-trade. We will, no doubt, continue to hear about the benefits and revolutionary impact of AI on markets from advocates, but perhaps what is being shown is that AI has a future in markets, but it is limited? Not least because for all its undoubted benefits, making trading decisions anytime we are event-driven is not one of them.

colin@thefullfx.com