XTX Paper Takes Aim at “Invisible Tax” from Hold Times
Posted by Colin Lambert. Last updated: May 17, 2021
A new paper from XTX Markets, Last Look Myths Debunked, argues that additional hold times within last look windows and the use of last look generally for commercial, rather than risk, purposes, represents “an invisible tax on clients and a distortion of their true costs of execution”.
The paper uses historical data provided by a client to analyse the impact of additional hold times (AHT) as well as asymmetric hold times, as it seeks to put the case for a more prescriptive Principle 17 of the FX Global Code, which deals with last look generally. It argues the lack of a precise definition of last look means that various LPs in the FX market have created different interpretations of what is acceptable – some of which then utilise last look for commercial purposes.
The paper kicks off by highlighting the existence of two separate, but related, practices around last look – the credit, market and position risk check, and the AHT, or “latency buffer” as it is often described as. XTX highlights that an LP using a last look window may accept or reject a trade based upon market, credit and operation risk limit checks, however it confesses to “issues” with the use of AHT.
It argues that the Code “clearly states” that last look cannot be used for optimising profitability through the control of adverse selection (where the LP only accepts profitable trades) and market impact (where the LP takes an extended time to check on post-trade market moves before accepting the trade). The paper also observes that some LPs operate variable and dynamic hold times for clients, thus making it unclear to that client exactly under which last look regime their trade is being handled.
“How can holding transaction requests and offers-to-deal in a way that varies by client, varies month by month, varies between trade acceptances and trade rejects, and is for a disproportionately long period, be anything other than profitability optimisation – be it profit maximisation or loss mitigation (two sides of the same coin)?” the paper asks.
A fundamental question asked by the paper is why an LP would wait an additional period before conducting its price check? The primary FX venues update at 5ms for their fastest feeds and 25ms for some pairs, why then, the paper asks, do some LPs have additional hold times of 200ms?
Pertinently, it observes that the FX Global Code says that the check should be done against the current price, not “at some price in the future at a time determined by the LP”. It also notes that in a 200ms window there are an average to 50 price updates on CME alone, not to mention those from the OTC venues such as EBS Market, Refinitiv Matching, Cboe FX and LMAX Exchange (all of which are, or who operate, a no-last look venue). This means, XTX argues, the more updates the LP can see, the easier it is to predict the likely P&L of the trade.
More pointedly, it also observes, “An LP can even set itself fill rate targets (per client) and commercially optimise its logic so that it focuses on rejecting the most unprofitable trades within its reject “quota”. In other words, not all rejects are equal; it is quite possible for the LP to have a relatively low reject rate but absolutely optimise their outcome.”
The paper does acknowledge that LPs’ technology infrastructures can vary in quality, and this may mean some have inherent latency in their price checking functions, but it then asks, “why would any AHT applied in order to meet this current price check be different according to who the client is?”
Citing anonymised bank disclosures, the paper cites evidence that many LPs do indeed have different hold times for different clients, and then raises the oft-presented argument by proponents of longer hold times, the ability to handle toxic flow. Echoing an argument made in these pages (and those of Profit & Loss) over many years, the paper observes “there is a perfectly good, tried and trusted mechanism for that – it is called spread”. Toxic flow, XTX argues, should attract a wider spread, and better flow a tighter spread.
It also takes aim at the idea that disclosures alone can provide the necessary level of transparency around an LPs actions. By leaning upon disclosures from anonymised bank and non-bank LPs the paper highlights just how uninformative and unnecessarily complex some documents are and how some explicitly state the LP will use trade information and utilise pre-hedging “as permitted by the FX Global Code”. In yet another awkward question for some, the paper asks, “Would you as a client agree that pre-hedging, over-hedging, and use of unfilled order information is acceptable market practice so long as it is disclaimed to you in a disclosure document?”
The paper also confronts the objection raised by some LPs, that removing AHT would trigger a technological race to zero. This is, in its purest sense, a real possibility of course, but XTX suggests, whilst reiterating that it believes AHT should be prohibited, that a market consensus to allow a short AHT would prevent this. In terms of how this generally agreed AHT could be calculated, the paper suggests it be explicitly linked to an LPs “tick to trade” time, which it defines as the time it takes the LP to receive a relevant market update and make its next price. This would cover those LPs who do not pay for the top-end 5ms data feeds.
“This ‘median tick to trade’ should be a reasonable benchmark for all LP’s,” the paper asserts. “We would like to see public disclosures of these times. It is not an inherently difficult calculation for any LP engaged in electronic market making and does not represent a significant overhead.”
The Challenge of Client Behaviour
Using data provided by a client, the XTX paper highlights changing LP behaviour over a three month period in 2020. February to April. This is an insightful time horizon as it took in the start of the pandemic, when markets were volatile but not excessively so; March, when volatility spiked tremendously; and April, when activity levels dropped significantly.
The paper contains tables that highlight the challenge XTX has in trying to win hearts and minds in this debate, for in spite of what could be termed worse behaviour by certain LPs, who increased hold times as well as asymmetricity, those LPs won more market share. One set of tables looks at LP behaviour in a “full amount” book over the three months, the other in a “sweep” book over the same period.
All clients and LPs were cross-connected, latency variations are therefore under 2ms and while round trip times could be impacted by a single day’s anomaly the data is “broadly representative”.
Behaviour in the “full amount” book stays largely consistent in terms of round trip times, especially for acceptances, however one LP shifts accept time to 75 ms in April from 10ms in February and March. The same LP spikes the reject round trip time to 125ms in March, however, and keeps it there for April. Most of the other LPs stay consistent at 10 or 20ms for round trip times (both accepts and rejects) throughout the three months, aside from one which has reject times of 40ms in February and 30ms in March.
The striking aspect of the data is how “LP D”, the firm to extend round trip times most dramatically, saw the lowest reject rates and the highest market share in the ‘full amount” book.
The data from the sweepable book, which typically attracts more aggressive flow, actually indicates more stability in round trip times (albeit in many cases much longer than in the “full amount” book), but reject rates spike sharply in March from the aforementioned “LP D” (and “LP A”).
Reject rates from the other four LPs analysed indicates reject rates that are largely stable – there are inevitably, some changes – whereas “LP D” has reject rates of 3.1% in February, 60.6% in March, and 2.3% in April (“LP A” was 12%, 57.6% and 8.2% respectively). Again, however, in market share terms, “LP D” leads the way with a 27% market share in February and 25% in March – it was, however, surpassed in April, but still had a 24% share.
The obvious conclusion to make from this data is that the client is not closely studying the cost of rejects, rather it is still focused on spread. Noting that it finds the behaviour laid out in the tables as “inconsistent with the FX Code” XTX observes it is difficult, even for sophisticated clients, to compare service levels and the cost of trading across the panel of LPs.
Acknowledging that it is almost impossible to calculate the true total cost of AHT in the market, (“The very fact that LP’s try so hard to retain variable AHT as a mechanism, and spend so much time obfuscating clients around the cost suggests it is considerable.”) the paper argues that the mechanism distorts true price competition.
It also refutes the argument that what the data actually shows is merely a redistribution of yield across the panel of LPs, mainly because “real” LPs, with a genuine alpha-based skew, risk capital and interest to deal, can be crowded out, and excluded from the mass of the client’s flow. This would, it argues, affect their ability to price the client in the future or when really needed. “That is neither good nor fair for the client as it will ultimately result in wider spreads,” XTX states.
It also provides analysis that indicates that the value bound for the client is a choice price (minimal in a world in which spreads are so tight), whereas for the LP, the AHT provides unbounded value (conversely, unbounded slippage costs for the client).
The paper also takes aim at LPs use of asymmetric hold times, typically where the round trip time to reject a trade is longer than to accept. XTX states, “There is zero justification for using default asymmetric acceptance logic for what the FX Code defines as a ‘risk control mechanism’. Asymmetric logic simply means that the LP will fill the client at the original (now off-market) rate if the market moves in the LP’s favour; but reject the client if the market moves in the client’s favour.”
Certain top-tier LPs” still use asymmetric logic as their default setting – in other words, a client has to explicitly opt out.
It adds, cuttingly, “The reason asymmetric logic exists is that it is more profitable for an LP than symmetric logic on the same flow. This is zero-sum: the client loses precisely whatever the LP gains.”
Whilst accepting that, with “crystal clear” disclosures in place and on the understanding the client can quantify the cost of asymmetric last look, there may be a place for the practice, the paper claims that “certain top-tier LPs” still use asymmetric logic as their default setting – in other words, a client has to explicitly opt out”.
It also observes that even in a symmetric environment, the balance of rejects goes against the client, therefore, the longer the AHT, the higher the cost of rejects to that client becomes. It adds that the “reality” is, “the longer the AHT, the more likely to LP is to reject you”.
Any Other Business…
The paper also takes time to dispel four “myths”.
The first is that regional banks are crowded out because they can’t price as quickly, calling this a “red herring”. In fact, XTX argues, it is the “AHT wielding players” that are squeezing the local banks by showing (in competition) much tighter spreads even though they, to quote the paper, “only have interest in trading on one side of the market” (they just can’t tell you which side yet……).
The second “myth” argues that AHT is needed for those regional customers in less developed areas who are trading on a GUI. There is indeed natural latency in this flow but, the paper argues, “The AHT clock only starts ticking once the offer to deal has been received by the LP.” Thus, while round trip times will be longer, the actual risk checking process should be the same, and based upon the median tick to trade of that LP (the paper also observes that HSBC and JP Morgan service their clients in connectivity challenged geographies without AHT).
The necessity to use AHT on anonymous venues is the third “myth” debunked in the paper, although this does require the cooperation of the ECN owners to a degree. Accepting there is “some justification” for AHT in this environment, the paper argues that removing certain tags from a price stream (or pricing them wider) offers a solution. It also calls for all LPs on a platform to be held to the same AHT, if it is deemed necessary on that venue, and, while praising those platforms that have shortened the maximum hold time in recent years, “those times could be significantly shorter”.
The paper also backs the idea, first initiated by Euronext FX (when it was Fastmatch) of the ECN applying the price check independently. This methodology means the LP does not see rejected trades (which are judged on a symmetric basis of course) and therefore cannot use the information.
The final “myth” acknowledges the minefield that price improvement has become in equity markets, by observing that some LPs claim AHT allows them to improve client execution. XTX argues that such a price improvement suggests that the market has indeed moved in the client’s favour, but that the original LP may no longer be best bid or offer, and as such the trade should be re-priced to ensure the client gets best execution. It also observes that it becomes “all too easy” for large players to use their scale advantage across a portfolio of trades, meaning downstream clients will have less visibility on individual trade performance. “This is a slippery slope that we think would be well avoided in the FX market,” the paper states.
The paper concludes by suggesting adjustments to the language in the FX Global Code – something that is unlikely to happen given the GFXC’s apparent determination to provide further guidance on last look rather than redrafting Principle 17. While that is the bad news for XTX, the more positive is the GFXC’s drive for more uniformity in disclosures through the use of templates.
The content of the disclosures will still be a matter for further debate, of course, but the very fact the GFXC wants clients to be able to compare and contrast LPs is a step forward. Ultimately, the paper is the latest step in a long campaign by XTX to generate more transparency around last look processes and, more pertinently, to eliminate its use for commercial purposes. As the conclusion states, “The FX Code should make a clear statement that profit-related issues of adverse selection and managing market impact should be handled using spread.”
There, in a nutshell, is the argument.
The full paper can be read at https://www.xtxmarkets.com/leadership/last-look-myths-debunked/