Quantifying Adverse Selection Apples-to-Apples

Bid-Ons, Offer-Unders

When comparing aggressive and passive markouts, one faces a compositional bias; participants trading passively could be different from those trading aggressively - in terms of technical capabilities, alpha, objectives, etc. There is one subset of orders which provide their own (almost perfectly, see note at end) matched control group, so to say: Bid-Ons (& Offer-Unders).

Suppose the ask level shows 100 lots of liquidity. A participant then sends a buy order (but not an IOC) for 130 lots targeting this level. 100 lots will fill aggressively, but the remaining 30 lots will enter the book and establish the new best bid. Assuming this resting part is not canceled and that the price does not move away, it will also be filled at some later time.

The key is that the aggressive and the passive fills came from the same participant, based on the same information and intentions. Hence, comparing the markouts of the aggressive trade to the markouts of the subsequent passive trades represents a good (apples to apples) metric for the adverse selection encountered (or implicit optionality provided) by resting orders.

Data and Methodology

We use A7 to extract all bid-on (and offer-under) events incl. individual fills for inserted volume and markout prices for the initial aggressive trade and each passive fill. We pick eight of the most liquid equity index and fixed-income futures trading on Eurex: DAX40 (FDAX), Euro STOXX 50 (FESX), Bund (FGBL), Bobl (FGBM), Schatz (FGBS), Buxl (FGBX), OAT (FOAT), and OAT (FOAT) futures. We do this for the most liquid contract of each product for each trading day of 2023 up to and including 20230531. We then compute the volume-weighted average aggressive and passive markout and normalize each by the tick size.

Results

Fig. 1 shows the results. We see that

• with the exception of FOAT, the aggressive trades mark out positive for all instruments
• the aggressive trades consistently outperform the passive fills
• the passive fills have negative markouts for all instruments
• the gap of 0.2 ticks is surprisingly consistent across instruments
• the split between aggressive and passive volume is very balanced; except for the FESX, it is between 40/60 and 60/40.
• for some instruments the markout of the aggressive portion of bid-ons deviates significantly from that of non-bid-on full-level trades. All else being equal, one would expect bid-ons to mark out better than non-mark-outs but the data shows the opposite for the majority of instruments.

Notes:

1. If one wanted to nitpick, one could point out a remaining bias. Suppose bid-ons where only a small remainder enters the book (compared to the aggressive volume) came from a different population than orders with little aggressive volume compared to the much larger passive insert. This could skew the markouts. One could analyze this by comparing orders with similar aggressive/passive splits or by only considering the minimum of the aggressive and passive volume from each order and bucket.
2. We ignore trades which trade more than just the top level.
3. We ignore trades which interact with hidden iceberg volume.

????