How accurately can complex option trades be signed? First grading against exchange truth
To build a dealer book from the options tape you must decide, print by print, who bought. For simple orders the quote rule does that. But on the day we measured, 40.2% of all SPX option prints were legs of multi-leg packages — and on those legs the quote rule is invalid by construction: an exchange fills a spread at a net price and allocates it across legs by convention, so a leg can print…
To construct a dealer book from the options tape, one must determine, print by print, who purchased the options. On a particular day, 40.2% of all SPX option prints were parts of multi-leg packages, but in these cases, the quote rule is not applicable by design. An exchange executes a spread at a net price and divides it among legs based on a convention, meaning a leg can appear anywhere within (or outside) its own market, irrespective of who initiated the trade.
The terminal signs the package, reassembles the legs printing at the same millisecond, compares the package's net price to a net bid/ask derived from the leg quotes, and then applies the quote rule at that level.
There has never been a quantifiable measure of accuracy for this complex trade signing process. Cboe provides a free trade-by-trade sample carrying details such as the side, capacity (customer / market maker / firm), trade type, and an identifier linking legs of one complex execution. This data represents ground truth at the per-leg level.
When joined with the trading tape, 12,422 complex legs are evaluated, and the result shows that the rule, when graded on customer legs per print volume-weighted package signer, achieves an accuracy of 80.4%. When excluding legged-in contras (which are unrecoverable in principle), the accuracy rises to 87.9%. However, when considering single-leg quote rule applications on the same legs, the accuracy drops to 75.1%.
Even the category of "legged-in" (which Cboe itself does not attribute) performs at 69%, which is worse than the coin flip (50%) accuracy that the rule was budgeted for. The reverse check reveals that a rule claiming to identify the customer should be reliably wrong about the market maker on the opposite side. Indeed, it is: on legs where the sampled participant is a market maker, the rule aligns with that side only 10.9% of the time.
The semantics reveal that the rule fails against time-priority truth (16.9% against true liquidity removers) but scores 90.4% against auction initiators. This discrepancy is due to the package quote rule identifying the side that paid the spread against the derived net mid, rather than the side that arrived second. In the context of a customer-versus-dealer book, this is not a defect but the very reason the rule functions successfully.
During one truth day, the binomial standard error was approximately 0.7 points. The signer abstains from making a determination on approximately 45% of complex volume, rather than guessing.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
