Back-Level Orders - What Have They Ever Done For Us?

One sees a lot of order inserts away from the top level, sometimes pretty far away. Why?

The hope is that (a) orders ahead of the inserted order will get canceled and/or (b) that additional volume will have joined behind our order by the time the order's price level becomes the top level.

Are these hopes justified? To what degree?

Fig. 1 below shows the distribution of queue position gains between the time of the order insert away from the top-level and when the order's level becomes the top-of-book for the first time. The different lines denote different distances from the top-level of the order's insert level (0 = top-level). The circles are the medians, the squares are the averages. The percentages in the legend are the fraction of these orders of all traded orders (not volume-weighted).

The queue position improvement is the difference in actual volume ahead once the level becomes top-of-book and the counterfactual of only inserting on this level as it becomes the top level. For example, assume the top level volume is 100 lots (excl. this order). So, if we inserted the order immediately after the level becomes the new BBO, it would have 100 lots ahead in the queue. But because we inserted the order earlier we only have 80 lots ahead but 20 lots behind. The improvement is hence (100 - 80) / 100 = 0.2, i.e., the order is 20% further up in the queue than it would otherwise have been.

Observations:

• We see from the percentages that the vast majority of volume comes from orders which were either price-setting or which joined the top level (95% or so).
• Our sample is heavily biased because we only consider orders which traded. We can safely assume that orders which are either canceled or never traded experience smaller queue position improvements.
• The improvement increases the further from the touch the order was inserted (as expected), but the marginal gain by inserting an additional level further out decreases with distance.
• Most orders inserted "touch adjacent" only exhibit minor gains.
• Even for orders inserted 10 ticks from the touch, the majority is only in the middle of the queue by the time they reach the top level; only around 15% have reached the first quintile.
• Compared to the latency race for queue position following price changes (e.g. see here), these gains seem fairly modest.

Back in the day when level-3 market data was not available, one did not know even one's own queue position. Obviously traded volume was removed from the front, but when an order was simply canceled (volume on level became smaller) one could not tell if this was one of the orders ahead or behind. Usually, the conservative assumption was made, i.e., assume that all cancels are for orders at the back of the queue - unless incompatible with the data. Fig. 1 demonstrates why this was a pretty good approach.

Speaking of orders far away from the touch, the banner image is a visualization of the FESX order book over a few days in 2020. Note the GTC (good-til-cancel) orders patiently waiting in the outer reaches of the solar system for their chance to shine (contiguous, long horizontal lines near the top and the bottom); for more on GTCs, see here.

#eurex #marketmaking #microstructure #quantitativetrading #quantitativeresearch #hft