What does a full-day average spread establish about a strategy that trades only around an announcement? Very little unless the observations cover those actual trading conditions. Making the average larger does not establish that the missing regime has been measured.
This research note is a method for interrogating cost assumptions, not an execution study or a recommendation to trade an event.
Price the decision that would be implemented
A useful cost specification starts with the instrument, venue, order type, size, decision time and holding period. It then separates spread crossing, explicit fees, financing or borrow, and price differences attributable to delay or impact. Almgren and Chriss model execution as a trade-off between transaction costs and uncertainty while a position is being executed, including temporary and permanent impact.[10] Their framework explains why execution speed is part of the problem; it is not a ready-made calibration for every instrument.
Choose a reference price before measuring slippage. A difference from the decision midpoint is not the same component as a difference from an executable quote at order arrival. Adding a full spread charge to a fill-based shortfall that already includes spread can count the same friction twice.
Turnover needs units too. Specify whether it counts each traded leg or divides purchases plus sales by two, and whether the cost rate is per side or per round trip. A cost formula with an ambiguous turnover definition cannot be audited. GIPS provision 2.A.13 requires transaction costs to be deducted when calculating returns, including its defined gross-of-fees returns.[3] This is a reporting reference, not a claim of compliance or proof that simulated deductions are realistic.
Conceptual example
Imagine two identical hypothetical orders. One is scheduled in the liquid part of a session; the other is triggered only when the available quote is sparse and changing quickly. Assigning both the same all-day cost estimate suppresses the very condition that distinguishes them. No assumed price or return is needed to see the identification problem.
Now freeze a synthetic order ledger and raise a non-negative fee charged to every executed side. Total net P&L must not improve. If it does, inspect signs, units and duplicated or missing legs before concluding anything about the strategy.
Cheap check
First run that fixed-ledger fee fixture and reconcile each gross-to-net adjustment independently. Then inspect cost-data coverage at the proposed order times, grouped by instrument, session, order type and relevant event conditions. Report unsupported groups rather than filling them with the unconditional mean.
Pass: the fee fixture is monotone in total net P&L, entry and exit reconcile once each, and every group underpinning an execution-realism claim has relevant evidence or an explicitly bounded assumption.
Fail: higher fees improve fixed-ledger net P&L, components overlap, or a supposedly measured regime has no matching observations. Withdraw the execution-realism claim. An unsupported group is an evidence gap, not automatically a dead strategy.
Limitations
If costs alter sizing, stopping or order choice, the order path changes; fixed-ledger monotonicity no longer describes the whole adaptive system. Quotes alone do not establish actual fills, queue position or capacity. Stress scenarios measure sensitivity under assumptions, not the probability of execution outcomes. Report a break-even cost as a research hurdle, never as evidence that the hurdle is attainable.
The availability-clock note precedes this check; fail-closed safeguards make its missing-evidence boundary enforceable. The gate keeps accounting and validation separate.
Sources
[3] https://www.gipsstandards.org/wp-content/uploads/2021/03/2020_gips_standards_firms.pdf — CFA Institute — 2020 GIPS Standards for Firms [10] https://smallake.kr/wp-content/uploads/2016/03/optliq.pdf — Almgren and Chriss — Optimal Execution of Portfolio Transactions (2000; mirrored primary paper)