On a regulated platform, a cluster of eight accounts that behaved as one participant generated 49 self-trading events across a set of related markets on a single fixture. In the largest of those markets that activity was 34 per cent of all volume, and roughly 31 per cent in two adjacent markets on the same event, simultaneously.
Read one way, that is a serious anti-money-laundering finding: a third of a market's apparent depth was one participant trading with itself. Read another way, it is a market maker doing its job, quoting both sides so that members of the public have something to trade against.
The two readings are not distinguishable from the public record. They are distinguishable from the operator's own records, and only from those.
Signature
The quantitative fingerprint, so the pattern can be recognised again on any platform.
- Cluster size
- 7 to 8 accounts, densely interlocked
- Pairwise fills
- 3 to 16 per pair, most pairs connected
- Self-trading events
- 34 to 50 per market in the scan window
- Individual fill size
- $5 to $250
- Share of market volume
- 31 to 34 per cent, across three or more related markets at once
- Market conditions
- 24 to 36 hours before close, two-sided crossing at complementary prices
- Net position
- No directional accumulation
This is why reconciliation exists as a separate capability rather than a refinement of detection. No amount of better rule-writing resolves it. The operator names the accounts that hold a market-making mandate, and the same measurement becomes either evidence that the mandate is being honoured, or a finding about accounts doing the same thing without one.
What this cannot show
- Because only matched fills reach the settlement record on this kind of platform, order-book indicators had no visibility here at all. That is a standing disclosure, not a gap in this case.
- Volume shares are measured over the observed window, not the market's full history.