The Last Look…

For this column, I want to tick another issue off my list of things to talk about early in the life of The Full FX – the complacency around the micro issues of last look (and in case you missed the first public column, which took aim at equity market structure, conduct and lawyers, you can find it here.)

In particular, it seems to me that too few participants in FX markets worry about the cost of rejects, and if I am right, why is that?

Last week was a busy week and inevitably when reconnecting with regular acquaintances and friends the subject matter is both current and traditional. There is a lot of interest in what is going on in the crypto world, but inevitably when I am involved in the conversation, things like last look come up.

What was interesting to me was how so many people I spoke to, both on the provider and customer side, thought that last look was a “dead issue” as one put it. I think in terms of a general framework for behaviours it is, but what doesn’t appear to be settled are some of the micro issues within the subject, most pertinently, what is it costing firms with reject rates in the 20%-plus range? As I asked a couple of people last week, does anyone actually bother to look at the cost of dealing with an LP rejecting 5% versus one rejecting 15%?

Absolutely some participants do, and well done to them, but the majority still do not, it seems. I mentioned to a buy side portfolio manager (someone not involved in the FX execution process I should stress) that while the TCA report might show LP1 has a tighter spread than LP2, the cumulative cost of LP1 rejecting three times the number of trades should be taken into account. Had I been in the same room I suspect the response would have been a blank look, and further discussion elicited the view that Random Walk would even it out. All I can say to that is if Random Walk actually works why do all of my sports teams have a long and proud history of non-achievement?

It is not as if the analysis isn’t there, plenty of firms can provide it. One acquaintance last week suggested that in addition to random walk, taken in isolation the numbers are not that large, but that, to me, misses the point. First of all, it adds up over time, and secondly, if you take the cost of rejects into account the dynamics change and the effective spread looks very different. It may be that when cost of rejects is taken into account the LP seemingly with the tightest price and a 15% reject rate is actually offering the exact same spread as the LP rejecting 5%. In that case, where should the flow go, especially if one is talking about trust and a relationship?

My view is that firms ignoring the cost of rejects are highlighting their lack of concern over their FX execution more generally, which leads me to question whether they are actually bothering to check on hold times, or whether there are patterns across their LPs. We talk about the relationship a lot in FX markets, but so many aspects of that are moving online, so are customers looking at which of their LPs are costing them the most?

How can one LP can hold a deal request for 5ms while another holds the same party for 100ms? Both are receiving market updates at 5ms from the primaries and in real time from other venues, so why the delay?

If there are new readers scanning this, I should point out that I have had an aversion to last look that is now 15 years old and counting, but I accept there are some risk mechanisms required when quoting via so many channels. What I don’t understand is how one LP can hold a deal request for 5ms (some have zero hold time via certain channels and with certain counterparties), and another holds the same party for 100ms? Both are receiving market updates at 5ms from the primaries and in real time from other venues, so why the delay?

I have never bought the argument that the LP is giving the trade the longest possible time to come back into court, to me a longer hold time naturally equates to a higher reject rate. Also, where does it say that an LP has to make money on every trade?

One has to question why the LP is even bothering to quote the customer on a 100ms window because “the flow is generally toxic”. So what? If it’s toxic either stop quoting that customer and try to deal with the trauma of having, for example, a 4.5% share in the Euromoney survey compared to 4.6%; or widen the spread to that customer to take into account the toxicity of the flow.

I do sometimes wonder exactly how much of this issue is driven by ego – the need to have the tightest spread and the highest volume in the market – as opposed to good business sense. I understand that a portfolio of counterparties often works well, but if certain customers are particularly tricky, deal with them on different terms to those you don’t have issues with.

Conversely, if, as a customer, your reject rate is high, don’t be surprised or upset if your trading style is questioned. It bemuses me sometimes how the foreign exchange industry effectively created a structure that made it hard for latency arbitragers to thrive, but continues to let other participants machine gun away across a spread of LPs. Is it just the fact that the former was originally the archetypical four people in a garage that has now become a competitor of sorts, while the latter is a “name” that still has some kudos attached to it?

Overall then, I think the debate we have been having in this industry over last look has been healthy and delivered a good framework in the form of the FX Global Code – but that is all it has delivered, an outline. What needs to happen now is people having more discussions around the details of the various last look policies. I sense that most have come to terms with separating internalisers and externalisers in their liquidity pools (although I argue there needs to be a clearer definition of internalisation if this is to be really effective), but have largely ignored the rest of the issue.

That’s all well and good, but the devil really is in the detail here. The concept of last look does not necessarily increase the cost of trading on a universal basis, but its application in certain circumstances most certainly does – and I think more needs to be said about that.

colin@thefullfx.com