Queue-Preserving Amend vs Cancel/Replace Slippage Playbook

Date: 2026-03-12
Category: research
Scope: Limit-order execution on venues where some modifications preserve queue priority while others reset it

1) Why this matters

Most execution engines still treat "modify" as operational plumbing, not a first-class cost decision.

But in modern books, how you modify an order can dominate slippage:

• Queue-preserving amend (when venue allows it) can keep time priority.
• Cancel/replace usually re-enters at the back of queue.

If your router defaults to cancel/replace for every tweak, you silently pay a Priority Reset Premium (PRP): more missed fills, more late chasing, worse q95 implementation shortfall.

2) The key asymmetry

Let an order at time (t) have queue state (s_t) and remaining parent inventory (R_t).

For an action (a \in {\text{hold},\text{amend}_{qp},\text{cancel-replace},\text{cross}}), define expected cost-to-completion:

[ J(a\mid s_t)=\mathbb{E}[C_{fill}+C_{delay}+C_{chase}+C_{impact}\mid s_t,a] ]

In many regimes:

[ J(\text{cancel-replace})-J(\text{amend}_{qp}) = \text{PRP} > 0 ]

because cancel/replace adds queue-age loss and re-entry delay without improving adverse-selection risk enough.

3) Practical slippage decomposition

For parent order cost:

[ IS = C_{spread}+C_{impact}+C_{timing}+C_{opportunity}+C_{priority-reset} ]

with

[ C_{priority-reset} = C_{queue-loss}+C_{reentry-delay}+C_{catchup-convexity} ]

• Queue loss: forfeited near-touch rank and fill option value
• Re-entry delay: dead time between cancel ACK and effective new queue position
• Catch-up convexity: later urgency causes nonlinear impact when remaining time shrinks

This term is often hidden inside "timing" in dashboards, so teams underestimate it.

4) Venue semantics are not uniform

A robust engine must encode a venue-specific modify matrix (examples):

• quantity decrease: often queue-preserving
• quantity increase: often priority-losing
• price change: usually priority reset
• display flag / TIF changes: venue-dependent reset rules

Treating all modifies as equivalent is a modeling bug.

5) Action-value framing (fill hazard × toxicity)

Model two hazards while order rests:

• fill hazard: (\lambda_f(t))
• adverse-selection hazard: (\lambda_a(t))

Expected edge of keeping queue priority over horizon (H):

[ V_{queue} \approx \int_0^H e^{-\int_0^u(\lambda_f+\lambda_a)dv} \left(\lambda_f(u),g_f(u)-\lambda_a(u),g_a(u)\right)du ]

Where:

• (g_f): expected gain if filled (vs fallback)
• (g_a): expected loss if toxic fill occurs

Approximate decision rule:

[ \text{Prefer amend}{qp} \text{ over cancel-replace if } V{queue} > \Delta C_{risk-reduction} ]

In short: only reset priority when toxicity reduction clearly dominates lost queue option value.

6) New metric stack

• PRP (Priority Reset Premium): [ \mathbb{E}[IS\mid cancel-replace]-\mathbb{E}[IS\mid amend_{qp}] ]
• QOR (Queue Option Retention): expected value retained by queue-preserving actions
• RAD (Reset ACK Delay): cancel/replace acknowledgment + re-accept delay
• RCR (Reset Catch-up Ratio): fraction of residual completed aggressively after reset
• PQS (Priority-Qualified Share): share of modifications executed via queue-preserving path when eligible

Healthy production trend: PRP↓, RAD↓, RCR↓, PQS↑ (without completion misses).

7) State machine for live control

• HOLD_EFFICIENT: fill hazard strong, toxicity moderate -> keep
• QP_AMEND: small adjustment needed and venue allows priority preservation -> amend
• RESET_CAUTION: priority reset allowed only with explicit toxicity/delay trigger
• DEADLINE_COMPLETE: near close or residual risk high -> completion-first policy

Use hysteresis on transitions to avoid flip-flop modify storms.

8) Calibration workflow

Offline

1. Build event-level ledger: submit, amend, cancel ACK, replace ACK, fills, markouts.
2. Label whether each modify was queue-preserving or priority-resetting.
3. Estimate fill and adverse hazards by queue age, spread state, imbalance, and venue.
4. Fit PRP by regime (open, midday drought, close, high-vol windows).
5. Backtest policy variants with q95 and completion constraints.

Online

• Continuously track RAD and RCR drift.
• Disable aggressive reset policies when ACK latency or reject ratio spikes.
• Refit hazard priors weekly; refit regime multipliers daily.

9) Guardrails that prevent expensive mistakes

• hard cap on reset frequency per minute
• reset cooldown after consecutive failed re-entries
• mandatory queue-preserving path when eligible and no red toxicity trigger
• kill-switch to fallback schedule if venue modify semantics become uncertain (rule-change / incident)
• dual-key promotion gate: q95 improvement and completion reliability

10) Implementation checklist

• Venue-specific modify-semantics table versioned in code
• Queue-preserving eligibility evaluated before cancel/replace logic
• PRP/QOR/RAD/RCR/PQS metrics on dashboard + alerting
• Reset throttles and hysteresis implemented
• Canary rollout with per-venue rollback knobs
• Post-trade TCA explicitly reports (C_{priority-reset})

11) References

• Harris, L. (2003), Trading and Exchanges (market microstructure fundamentals).
  https://global.oup.com/academic/product/trading-and-exchanges-9780195144703
• Cartea, Á., Jaimungal, S., Penalva, J. (2015), Algorithmic and High-Frequency Trading.
  https://doi.org/10.1017/CBO9781139342674
• Gould, M. D., et al. (2013), Limit Order Books. Quantitative Finance.
  https://doi.org/10.1080/14697688.2013.803148
• Huang, W., Lehalle, C.-A., Rosenbaum, M. (2015), Simulating and Analyzing Order Book Data: The Queue-Reactive Model.
  https://arxiv.org/abs/1312.0563
• FIX Trading Community, Order Cancel/Replace Request (MsgType=G) semantics.
  https://www.fixtrading.org/standards/

One-line takeaway

In queue-driven markets, modification path is alpha: use queue-preserving amend whenever eligible, and treat cancel/replace as a costly risk action—not default plumbing.