Queue Priority Reset Tax in Dynamic Repricing

Date: 2026-03-06
Category: research (execution / slippage modeling)

Why this playbook exists

Many desks "improve" passive execution by repricing child limits every tiny quote move. In live books, that can backfire:

• each replace often loses queue priority,
• you spend more time at the back than near the front,
• fills arrive later (or never),
• urgency later forces aggressive catch-up.

That hidden drag is the Queue Priority Reset Tax (QPRT).

Core mechanism

For a buy child order:

1. You post at best bid and start aging in queue.
2. Microprice tilts up by 1 tick; controller sends replace.
3. Venue applies replace as cancel+new (priority reset).
4. You move from near-front to back-of-queue.
5. Either:
   • you miss prints that would have filled old queue position, or
   • you fill later at worse levels after urgency escalation.

Symmetric for sells.

The error is not "too passive" or "too aggressive" alone. It is over-churn without valuing queue age.

Data contract (minimum)

Child-order event tape:

• parent_id, child_id, symbol, side, venue
• submit_ts, ack_ts, replace_send_ts, replace_ack_ts, cancel_ts
• limit_px_before, limit_px_after, reason_code
• visible_size, remaining_qty
• L2/L3 snapshots (or best-effort queue estimates)
• trade prints with aggressor side
• benchmark refs: arrival mid, decision mid, microprice
• fees/rebates and odd-lot flags

Without replace and queue-state timestamps, QPRT is mostly invisible.

Metrics that expose QPRT

1) Replace Churn Intensity (RCI)

[ RCI = \frac{#(replaces)}{\text{child active seconds}} ]

Track p50/p90 by symbol and regime.

2) Queue Age at Fill (QAF)

[ QAF_i = t_{fill,i} - t_{last_priority_reset,i} ]

Falling QAF with rising RCI is a red flag.

3) Missed Queue Opportunity (MQO)

Estimated shares that would likely have filled without the last reset:

[ MQO = \sum \max(0, \hat q^{no-reset}{fill} - q^{actual}{fill}) ]

Use conservative queue-depletion simulator.

4) Queue Priority Reset Tax (QPRT, bps)

[ QPRT = 10^4 \cdot \frac{C_{actual} - C_{counterfactual,no-reset}}{\text{notional}} ]

Counterfactual must replay real prints/quotes with reset removed, not a static spread model.

5) Churn-to-Catchup Ratio (CCR)

[ CCR = \frac{\text{cost from urgency catch-up}}{\text{cost from passive fills}} ]

When CCR rises with RCI, controller is borrowing pain from the future.

Modeling blueprint

Model expected incremental cost of sending a replace at time t:

[ \Delta C_{replace}(t) = \underbrace{\Delta C_{adverse,selection,avoided}}_{benefit}

• \underbrace{\Delta C_{queue,age,loss}}_{QPRT}

• \underbrace{\Delta C_{deadline,risk}}_{urgency,interaction} ]

Only replace when \Delta C_replace(t) < -\epsilon.

Component models

1. Adverse-selection component (benefit of moving quote)

   • Inputs: microprice drift, imbalance, cancel/trade intensity, short-horizon volatility.
   • Output: expected bps saved if price moved now.

2. Queue-age loss component (QPRT term)

   • Survival/hazard model for fill conditional on queue age and depth ahead.
   • Output: expected additional delay + expected missed-fill notional from reset.

3. Deadline-risk component

   • Remaining time, residual size, participation caps.
   • Output: expected future catch-up cost if current passive path under-fills.

State machine for live control

1. STABLE_QUEUE

   • RCI low, QAF healthy
   • normal repricing allowed

2. CHURNING

   • RCI rising, QAF dropping
   • widen no-replace band, enforce min dwell

3. QUEUE_DAMAGE

   • QPRT burn-rate breach or MQO spike
   • freeze non-essential reprices, split child into layered levels

4. DEADLINE_RISK

   • underfill risk high near cutoff
   • controlled aggression with cap; avoid panic sweep

Use hysteresis (harder exit than entry) to avoid oscillation.

Practical controls

Control 1: No-replace micro band

Do not replace for tiny reference moves unless move persists:

• e.g., require |microprice shift| >= θ for τ ms.

Control 2: Minimum dwell timer

After any replace, require dwell before next replace unless risk override triggers.

Control 3: Queue-age-aware sizing

If queue age is high (good position), prefer size reduction over full-price reset.

Control 4: Layered child structure

Use two layers:

• anchor layer: protects queue age (low churn)
• reactive layer: handles short-term micro alpha

Control 5: Deadline guardrail

If projected completion probability drops below threshold, pre-emptively shift a bounded slice aggressive (instead of late full chase).

Backtest and promotion protocol

1. Build event-level replay with queue reset semantics per venue.
2. Run A/B policies:
   • baseline (naive frequent replace)
   • QPRT-aware gating policy
3. Evaluate by regime: open, lunch, close, news windows.
4. Track tails (q90/q95/q99), not only mean slippage.

Promotion gates (example)

• q95 slippage improvement >= 5 bps
• MQO reduction >= 15%
• completion ratio not worse by > 1.0 pp
• aggression cost at deadline not worse by > 2 bps

Rollback if two consecutive windows breach q95 or completion floor.

Common mistakes

1. "More responsive quoting is always better"
   False when priority resets dominate micro-alpha gains.

2. "Replace cost is just exchange fee"
   Bigger cost is queue age destruction, often uncaptured in fee analytics.

3. "Average slippage improved, so done"
   Tail and deadline blow-ups can worsen while mean improves.

4. "One policy fits all venues"
   Priority/replace semantics differ materially by venue.

Minimal pseudo-policy

score = benefit_adverse_selection - cost_queue_reset + cost_deadline

if state == STABLE_QUEUE and score < -eps:
  replace()
elif state == CHURNING:
  replace_only_if(score < -eps_strict and dwell_passed)
elif state == QUEUE_DAMAGE:
  freeze_nonessential_replaces()
  maintain_anchor_layer()
elif state == DEADLINE_RISK:
  bounded_aggressive_slice()

if QPRT_burn_rate > limit:
  widen_no_replace_band()

Desk-level takeaway

Repricing logic must optimize price + priority + time, not price alone.
If your controller ignores queue age, frequent replaces can turn into a steady hidden tax that only appears when urgency arrives.