Does the Market Detect Icebergs?

TLDR

At least some market participants seem to detect iceberg orders and make use of the hidden liquidity when sizing their aggressive orders.

My apologies, this will be very niche and technical.

Iceberg Orders

Adding a large order to the book signals to other participants that you want to buy (sell) a lot of shares. This has the effect of pushing the price up (down) and it is why execution algos break down such large "parent orders" into smaller "child orders" which are inserted & modified over time until the desired quantity has been traded (in addition to partially trading aggressively).

Another approach which predates such execution algos is to use a so-called iceberg order . Say you want to buy 1,000 shares. By using the iceberg order only, say, 100 shares will show up in the book. Once the displayed 100 lots have been filled, the matching engine will release the next 100 shares and so on until the full quantity has been filled.

The idea is to hide the true size of the order. Unfortunately, only based on the public market data it is possible to identify and track an iceberg order - at least once it has replenished the displayed quantity for the first time. No heuristic or model is needed. There is (almost, see technical note at the end) no ambiguity.

Iceberg Fingerprints

The clearest pattern is to look for aggressive trades that were sent for more volume than was seemingly available on the level. If an iceberg was present on the level, then its displayed quantity (tip) would be fully filled, the next slice(s) of the iceberg would be released, some of which would be filled immediately and the remainder of the slice would be added to the book.

This scenario has the following fingerprint in the public data: Within an event (no change in timestamp, but possibly spread over multiple packets) we observe an (a) ExecutionSummary with RestingHiddenQty > 0 and (b) an OrderAdd on the passive side of the book and at the traded price. This added order is then the newly released slice of the iceberg order after the previously displayed quantity had been fully filled in addition to some non-displayed volume at the time.

From this point onwards, we know that an iceberg is present on the level (but not how much if any hidden volume is still open) and which order on the level represents its tip. This latter point allows us to track it over time, e.g. when its price is modified.

Note that there are other fingerprints that are also consistent with icebergs but which have other equally likely causes.

Data and Methodology

The question is: is this being done by participants and, if so, do participants modify their behavior? To answer this question, I extract a list of all trades from A7 and annotate each one with whether or not the market could have known that the traded level had an iceberg order on it prior to the trade (plus some other information to enable the filtering described next).

I then apply the following filter

• only single-level trades, i.e., no trade events where more than one level was traded
• only trades without a book update within the previous 10 ms; this is to ensure that the market should have been aware of the displayed volume on the level when the aggressive order was sent; no race conditions
• continuous trading
• DAX/MDAX/SDAX constituents

Finally, and based on proprietary data,

• only IOC orders
• with limit price = trade price

We then analyze what fraction of the trades had an order quantity less than, equal to, or larger than the visible volume on the targeted level.

Results

The instances where icebergs are revealed potentially introduce a bias: the revelation typically happens during large, full-level trades. Trades following such a revelation are likely statistically different from trades without such a preceding event.

Nevertheless, we see that the sum of the fractions of at least full-level trades (bars labeled "=" + ">" in Fig. 1) is about the same for both category. This is what matters here. If the market is aware of hidden volume then we would expect more participants to make use of this liquidity by sizing their aggressive order beyond the visible volume on the level.

And indeed we see that orders targeting levels with a detectable iceberg order were more often oversized than those targeting other levels; not by a large margin, but consistently for all segments and over time.

This would be consistent with some participants applying such iceberg detection and using the information for sizing their orders.

As a side note: I am surprised that even for levels without detectable hidden volume, 15% of orders have a quantity in excess of the visible volume (note that the volume on the level has not changed over at least 10 ms). Maybe these are participants wanting to probe for icebergs? Or maybe participants also looked for other patters which give a hint at the presence of icebergs.

Technical Note

The pattern described above would also be consistent with a stop limit order which was triggered by a trade which was fully traded, i.e., the inserted order is not part of the iceberg but instead a stop order which happened to be triggered in a trade involving an iceberg order. However, given that iceberg orders are relatively rare and that stop orders are relatively rare, the probability of both being seen in combination is small.

This pattern does not capture all iceberg events. Trades which only target the visible volume on the level (but with an iceberg on it) would also yield an OrderAdd in the same event and at the same price but without any hidden volume being traded. However, a stop order triggering would exhibit the same pattern.

#iceberg #xetra #equitytrading #microstructure #marketdata #quantitativeresearch #hft #algotrading #tca