If You're Dumb You Gotta be Fast

The FESX-FDAX Trade

The simplest, most obvious strategies also tend to be the most competitive - exactly because they are simple and obvious. The prime example on Eurex is the FESX (Euro STOXX 50 future) / FDAX (DAX future) arbitrage.

It relies on the fact that (a) these two instruments are highly correlated, (b) the FESX has a bigger tick size compared to the FDAX (2.5 bps vs 0.6 bps), and (c) the FESX leads the FDAX.

So, when the price of the FESX changes, it is very likely that the FDAX will follow suit. And because of the tick size difference, even crossing the spread in the FDAX will be profitable if that happens.

Trigger Threshold

Typically, a price change is triggered by a trade or a burst of trades. Because the latter scenario is more common, one does not wait for the trade which eventually clears the level. Instead, one triggers off a sufficiently large trade which in itself only partially trades the level.

Still, the larger the trade, the higher its information value and the more likely that additional subsequent trades will fully trade out the level. There usually is a threshold value for the trade size below which no action is taken and above which an aggressive order is sent. This threshold need not be static but could be a function of the current state of the book or some other model.

The likelihood of a subsequent FDAX move (and hence the profitability) increases with the trade quantity of the triggering FESX trade. For large enough trades, no other signal is required. A strategy "target the FDAX for every FESX trade above 500 lots" would likely be profitable - assuming a sufficiently high fill rate. For smaller triggering trades, the strategy needs a more complex model with additional inputs.

Composition of Reactions

The simpler (dumber) the strategy, the more competitive is the trade. This is shown in Fig. 1 below. The top panel shows the composition of what happens in the FDAX within 1 ms (this accounts for the known cable lengths and hence represents the pure reaction time of the participants) after a trade in the FESX of a given size (horizontal axis).

Following small FESX trades, there is usually no subsequent trade in the FDAX, i.e., nobody reacts (grey area). After large trades, however, there is almost always a reaction in the FDAX (blue shades). The shades of blue distinguish between the estimated reaction times of the participant with the fastest reaction.

The bottom panel shows the same data as the top panel except it excludes no-reaction samples. It only shows the composition of reaction times conditioned on there being a reaction. We see that the fastest participants dominate for larger trigger trade sizes. All reaction times below 1 μs most likely require a FPGA or ASIC. Reactions to smaller triggers are exclusively pure software reactions (≥1 μs reaction time).

Consequences

So, either you have a dumb strategy and are very fast, or are smarter than the competition so that you can better pick the good trades among the remaining achievable opportunities. Unfortunately, this is not an exclusive or. Participants can be fast and smart.

#hft #algotrading #fpga #asic #eurex #futures #lowlatency #microstructure #quantitativeresearch #arbitrage #latency

Footnotes

1. Another example (and the corresponding chart looks almost identical except for a different scale of the x-axis) is the FGBL/FGBX pair.
2. Data is exclusively based on HPT Executions files.