The Last Look…

Historically, there has always been an imbalance in fortunes in FX markets between market makers and takers, and the pendulum is rarely stable thanks to one or the other getting an upper hand, usually through technology. AI is a buzzword at the moment, and FX is not immune to the chatter, so how, if at all, will this technology alter the balance in our markets?

Over the past 50 years, it could be argued that the market makers had the advantage for the best part of three decades thanks to the interbank market being the only true source of price discovery and the relatively slow advance in hedging techniques by customers.

Over the past two decades that has changed, however, and I think it is fair to argue that the consumers have had the upper hand. Yes, they are still (sometimes) paying spreads, but those spreads have been crushed as competition for their business has grown and new techniques, such as algos, have helped them take more control of their execution, while still having the option of risk transfer if markets are a little hairy.

This should not, incidentally, be taken as an expression of sympathy for the market makers – last time I checked, they were largely billion-dollar businesses so are still doing nicely, perhaps not as nicely in some cases, as they were before.

Can AI, however, alter the balance again? I suspect it can thanks to changes in the market structure over recent years, driven by regulation or best practice.

There is little doubt that AI will help market makers better understand the flow they are quoting for, and react accordingly. It will also allow the makers to process the information from the trade quicker – although the reality is, if a customer gave a manual trader 10 out of 200, they would process that info pretty quickly as well!

I don’t think it will be about speed, more it will be, no pun intended, about intelligence – AI models should be able to better differentiate, for example, between flow that is “toxic” for 200ms, but anything but over, again for example, two minutes. This will allow the maker to better handle what flow they win and perhaps make them “smart brokers” rather than the churn and burn shops so many have become. There could be a higher degree of risk warehousing, thanks to the benefits delivered by AI.

For the takers, however, it could be a very different story, because this intelligence brings with it increased signalling risk, especially if the taker is spreading the business around – simply because it will understand how the taker usually executes. AI will also probably introduce defensive skews for flow that is, and will remain, tricky, after all, that is the natural course of things – if you know the counterparty is a seller why stand in the way voluntarily? At least try to protect your book a little.

In many ways, a lot of this intelligence is already gathered by the makers and utilised where possible, it’s just with AI is will become more systematised (and probably proactive at some stage).

So how does the taker react to this? It’s actually a tricky one because one way could be seen as a step back, while the other transits into tricky areas conduct-wise. I should point out that some, laughably, think the Fix is the answer, as it’s dealing by appointment, but that only works if you are counter-flow – as we have proved for over two years now with our monthly data on the month-end fixes.

The first real option is the retro one – go back to risk transfer because it is the sure-fire way of keeping your intentions out of the public eye. Equally, the use of algos could increase, but the taker still has the market risk, and the same challenges I outlined above for a taker around leakage are just the same for an agency algo.

This is anathema to many takers, however, because they have developed sophisticated models to help them eke out a few more performance points from their FX hedging activities. Equally, there are fewer makers willing to price competitively for larger amounts – too many are content to hand the order off to their algo team and take the fee without the risk.

This leaves the taker with a second option – try to “fool” the AI and disguise their business, perhaps deploy AI to deliberately confuse the other AI? (at which point my head explodes!)

It will be interesting to see how tactics deploy as AI comes more into play, if a taker is hitting a public market, or is dealing in parcels rather than (proper) full amount, there will be information leakage and it could increase in an AI era

Will takers consider trading, for example, against their order by calling lower grade makers? There are plenty of “liquidity providers” out there who do nothing but turn the risk over, sometimes with a bit for themselves in terms of a position (always short-term of course), but always exacerbating market impact. Does the “smart” taker hit these makers, before turning to an algo or a risk transfer with more solid LPs who will hold some of the risk for the “real” trade?

In some corners this will be called spoofing, but is it? There is an intention to deal (proven by the fact that the taker did actually trade with the “churners”, and the firm is taking on market risk. Previously a bank may have done something like this in an attempt to improve the client’s fill, and I understand why that has been stamped out, but it is a different thing doing it as an agent as opposed to a principal. Would such activity by a taker be thought of in the same fashion?

It has been proven that AI can be confused, even duped, and other AI will know that and look at it as a perfectly sound tactic (heads still exploding) and deploy it. If this happens, what does the oversight at the taker firm do? Is it more important to get better execution outcomes, or stay well away from what some see as a conduct line in the sand?

It will be interesting to see how tactics deploy as AI comes more and more into play, but the fact is, if a taker is hitting a public market, or is dealing in parcels rather than (proper) full amount, there will be information leakage and it could increase in an AI era. If this is the case, and execution quality is degraded, what do these firms do? More importantly, if they do try to mask their intentions and mix up their signals, will the regulators take an interest?

It is interesting that some regulators have been amongst those expressing concerns over the impact of AI in markets. The irony is that the market structure driven by these same regulators is likely to be exploited by AI to the detriment of those the authorities wanted to protect. They have created a perfect landscape for AI to shift the balance of power yet again – the question is, will it be allowed to?

colin@thefullfx.com