The Last Look…

Our industry is becoming increasingly focused, obsessed even, with workflow efficiency, something that has fed into the “AI is a real game-changer” narrative that is also building. Both are missing an important factor.

I absolutely understand the push for improved workflow efficiency, as spreads remain tight – and even tighten – cost control becomes an important part of the business, but I do sometimes wonder if we are forgetting the best way to actually save investors or shareholders money. Yes, we can make savings with greater efficiency (creating those savings comes at an initial cost of course), and AI will make a difference in helping manage and control workflow, but the best way to actually save money is still through genuinely better execution (and by that I don’t mean a tick-box exercise in TCA).

The evangelists argue that AI will also change the trading business, I am not so sure. Beyond helping with the aforementioned workflow (which was already heavily automated anyway), my problem with AI is that it remains data-dependent, and by that nature, reactive, rather than proactive. It might only be by milliseconds, but the chances are the market will already have reacted by the time the AI analyses the data and decides to trade, and in this business, unless you are really getting into longer-term trading decisions (a whole five minutes or more!), a few milliseconds is often too late.

Chris Churchman, Goldman’s head of Marquee, the bank’s single-dealer platform, highlighted this issue in a recent internal podcast, first reported by CNBC, when he warned of the danger of over-reliance on AI. “There’s a huge danger here that in the era of AI, we outsource our reasoning to these models, and we have cognitive atrophy that stops us being able to reason from first principles ourselves,” CNBC quotes him as saying, and, of course, he is right.

The next generation of traders absolutely need to understand the capabilities of, and be able to work with, AI, but believing it can take over the critical thinking required by a trader, not to mention the predictive nature of a trader’s process is a dangerous path to take. This was a point made by my friend Steve Flanagan in a recent 360T Podcast, when asked if AI will replace humans his response was “are you kidding me?”

It is imperative that the coming generations are appropriately schooled and are given hands-on experience when it comes to trading and risk

I was reminded of a conversation I had about 10 years ago, post-Brexit. I was with two people from the hedge fund world, one systematic, one discretionary as it turned out. Discussing the events a few months after the event, it turned out one had a much better day than the other (although both did well). The revelation, to me at least, was how they responded to the Sunderland result, which was when, for the first time, people started to seriously think the UK could leave the EU.

The discretionary trader, who had been in the business for a couple of decades and earned their stripes, said they saw the result and “smacked everything I could find”, going limit short in terms of UK assets. In FX terms (true to form I wasn’t interested in, or have forgotten, the other stuff), they said they got short of “a decent chunk of Cable” around 1.48/1.49, and watched it collapse down.

The systematic trader, however, saw the result, but reported the model only sold a small amount, not wishing to go all-in on one poll, presumably. As more data came in, it got progressively shorter, but, if I recall correctly, around the 1.43/1.44 level.

Clearly both had a good day, but if this was a smaller, less Black Swan event, the delay in trading by the model could have been expensive (it reminded me of a high-profile strategist I once worked with at a bank who was expert at hitting highs and lows the wrong way round because they always wanted one more piece of data).

Another example is SNB-Day. It was not humans who were selling EUR/CHF into virtual extinction, it was algorithms. The humans were trying to buy anywhere under 0.80 and were rewarded for their efforts by having to re-paper!

There is always the risk that as a trader AI will either be too noisy, or too late

Of course, AI can be trained not to over-react, but what if it is a genuine Black Swan? Does it second guess itself and delay a critical decision? This is a point I think Churchman was making in the Goldman podcast when he observed that systems must be designed so that “employees still call the shots in high-stakes, high-uncertainty decisions rather than becoming passive operators,” according to the CNBC report.

To me, this is the crux of the issue. Yes, AI will help with more routine assessments and processes – and it can help build workflow efficiency (are firms factoring in the energy required I wonder?) – but there is always the risk that as a trader it will either be too noisy (active), or too late.

At the moment there are few fears that AI will dominate FX markets at least, mainly because there is still a generation that grew up in a semi-manual world, who are used to, and understand, genuine risk and the need to make decisions for themselves. The concern for the industry starts when this generation disappears – therefore it is imperative that the coming generations are appropriately schooled and are given hands-on experience when it comes to trading and risk.

Frankly, my 91-year old mother could probably provide a decent price in most currency pairs (it would be wide and slow, she’s not as quick as she used to be, bless her), so yes, well done to the AI if it makes an LP’s pricing slightly smarter, but it’s not exactly changing the world is it?

What my mother probably couldn’t do, is manage the risk associated with pricing, or pick a direction of the market (and yes, before you all dive in, she would probably still be better than me…) and that is where the real skill is in trading. And it’s a skill I remain unconvinced AI has…

colin@thefullfx.com