[Essay 03] ·
Notes from the desk
Measuring execution quality in 24/7 digital-asset markets
Most established execution benchmarks are artefacts of a session-based world. Arrival-price assumes a meaningful arrival; VWAP assumes a representative volume profile; closing-auction-anchored measures assume a closing auction at all. None of these survive contact with markets that trade continuously, span multiple time zones of dominant flow, and exhibit volume profiles that are themselves a function of what other algorithms are doing on the same calendar tick.
A defensible approach is to measure against a rolling fair-value reference constructed from a wider information set than any single venue exposes. Slippage is then defined as the signed gap between realised fill price and that reference at the moment of submission, not at some session anchor that did not exist. Fill rate is reported against quoted size at the same instant, which captures the cost of liquidity that evaporates between decision and acknowledgement. Realised cost nets fees against rebates, and is tracked at the level of individual child orders rather than smoothed across a parent.
None of this is novel in spirit. What changes in a 24/7 setting is that the reference itself has to be recomputed continuously, and that the distributions of slippage and fill rate are heavy-tailed in ways that point estimates obscure. Instrumenting every order with the inputs needed to reconstruct the reference offline allows post-trade analysis to be replayed under different definitions without re-running the trade.
Without that scaffolding, "good execution" reduces to a feeling, defended by anecdote and contested by the next bad week. With it, the question becomes tractable: a desk can ask whether a given strategy is paying its costs, whether a venue is consistently more expensive than the reference suggests, and whether routing changes are improvements or merely changes.