Most Trades are Uncontested
One interesting tidbit regarding the much-hyped latency race to zero is how little of the market it affects. On Xetra, almost 90% of the trades are uncontested. Eurex is more competitive, but even there around 80% of trades are uncontested.
Within a 1 ms (an eternity) after a trade we count the number of IOC/FOK (immediate-or-cancel, fill-or-kill) orders which targeted the traded price(s) of the winner but missed. We exclude misses with the same owner as the winning order. We also only consider a single miss per participant. If a miss falls within multiple 1 ms windows, it is assigned to the most recent trade (no double-counting). Also, only on-book trades executed during continuous trading are considered. Finally, only trades stemming from IOC orders are used. Strictly speaking, the last condition is not necessary, but next week's post builds on this week's results and the rationale will become apparent then.
Without further ado, Fig. 1 shows the distribution for the number of IOC misses per trade for all Eurex (dark blue) and all Xetra (light blue) trades during April and May 2023. Note the log scale of the y-axis.
领英推荐
Frankly, I was somewhat surprised how far the histogram extends to the right. There are indeed rare instances where more than ten distinct(!) participants go for the same opportunity. But this is consistent with with Eurex's official statement in their latest (May 2023) "Insights into Trading System Dynamics" presentation (highly recommended, by the way) that "around 10 trading participants" have reaction times below 50 ns (we only knew there are at least 4). Note that the length of the observation window used here means that the vast majority of participants are included in the analysis and not just the elite HFTs.
In hindsight, one could have also counted late cancels to the IOC misses since these in a sense also participate in the race. But based on earlier work, their number is an order of magnitude lower than that of IOC misses.
Next week, I will look at the predictability of these contested events. Given participants A & B participate in the race, what is the distribution of the odds to win - for all (A, B) combinations? For a given pair, does always the same participant win?
Head Of Research bei SSW-Trading GmbH
1 年Great work as always Stefan, thank you very much! I know this is a slightly older post but here are my two cents: Maybe I am not that surprised that aggressive trades are mostly uncontested for equities. This reads to me as "a sginificant fraction of marketable orders are sent by someone's meta order handler who is trying to keep up with his participation-of-volume schedule and not some HFT/arbitrageur/otherwise informed party". If this is somebody working a large (multi day) order, these trades would probably even look profitable a large amount of the time since the meta order will be driving the price for some time after the trade.
Business Development and Partnerships - Unizen
1 年Super informative !
Research
1 年"(there are around 10 trading participants with reaction times < 2770 ns" why do you interpret it as 50ns?
Derivatives Trader
1 年Loving these articles Stefan Schlamp