Distribution of Markouts

I came up with a nice visualization which combines a lot of pieces of information into a single chart: profitability, volumes, latencies, and aggressive-vs-passive.

Fig. 1 shows the distribution of volume vs. markouts. The horizontal axis is the markout vs. the T+10 s microprice (to get a continuous variable) of trades in DAX40 constituents on Xetra during continuous trading. Against the x-axis it shows stacked histograms of the traded notionals for a range of different estimated reaction times.

The reaction time is measured with respect to the most recent (physically possible) trade trigger on Eurex or Xetra. The known fixed latency has been subtracted - the reaction time is the pure wire-to-wire latency on the side of the trading participant. For aggressive trades (notional shown along the positive y-axis), it refers to the incoming aggressive order. For passive trades (shown along the negative y-axis) it refers to the order insert event. Trades without an identifiable trade trigger within a 100 μs window are denoted as "non-competitive".

The fastest category (< 140 ns) comprises participants employing the practice of speculative triggering on an FPGA. The second category (140 ns ... 1 μs) includes non-spec triggers on FPGAs. The next three categories describe different classes of pure software implementations.

The chart demonstrates a few things very nicely:

• Only a small fraction of the volume is attributable to FPGA-triggered orders (sub-1 μs reaction time) - only 6.3% of aggressive trades and close to zero of passive volume.
• In aggregate, aggressive (passive) trades have positive (negative) markouts. This reflects the asymmetric fees for a significant subset of participants. Participants in the liquidity provider scheme who meet certain KPIs benefit from reduced or waived fees for their passive trades. No such rebate exists for aggressive orders.
• The standard deviation of the markouts far exceeds the average; this hold for all subgroup including the fastest participants. Hence the saying that one only has to be right 51% of the time. And while 51% might be an exaggeration, even the fastest participants are wrong very often.
• Having said that, one clearly sees the asymmetric distribution for the faster participant categories. A big chunk of the profitability comes from the tails.
• There is room for alpha capture which does not rely on pure speed. Even reasonably fast software platforms can capture sufficient edge.