Maker–Taker Net Slippage and Queue-Aware Routing Playbook

Date: 2026-03-02
Category: research (quant execution / slippage modeling)

Why this playbook exists

Many execution stacks still optimize venue routing with a shallow rule:

• “Take where fees are low” or
• “Post where rebates are high.”

In production, this underperforms because fees/rebates are second-order unless combined with queue risk and adverse selection. A venue with attractive rebate can still be expensive if:

• fill probability is low,
• fills cluster when price is about to move against us,
• repeated cancel/replace burns queue priority.

This playbook defines a practical all-in slippage model for maker/taker routing and shows how to turn it into controllable production decisions.

1) All-in net slippage objective (per child order)

Use expected cost in bps versus arrival mid:

[ \mathbb{E}[C] = C_{fee/rebate} - C_{spread\ capture} + C_{impact} + C_{adverse} + C_{nonfill} + C_{latency} ]

Where:

• (C_{fee/rebate}): taker fees minus maker rebates,
• (C_{spread\ capture}): half/full spread captured by passive fills,
• (C_{impact}): mechanical impact from own aggression/size,
• (C_{adverse}): post-fill markout loss (toxicity),
• (C_{nonfill}): opportunity cost from misses and later chasing,
• (C_{latency}): stale-quote and routing-delay penalties.

This extends the practical “all-in cost” framing often used in TCA (fees/rebates + spread capture + markout/impact) to include explicit queue and timing penalties.

2) Venue-mode expected cost decomposition

For venue (v) and mode (m \in {passive, midpoint, taker}):

[ \mathbb{E}[C_{v,m}] = f_{v,m} - r_{v,m} - s_{v,m},p^{fill}{v,m} + a{v,m},p^{fill}{v,m} + n{v,m}(1-p^{fill}{v,m}) + \ell{v,m} ]

Interpretation:

• (f_{v,m}): explicit fee,
• (r_{v,m}): expected rebate credit,
• (s_{v,m}): spread capture conditional on fill,
• (a_{v,m}): adverse-selection markout penalty conditional on fill,
• (n_{v,m}): non-fill chase penalty,
• (\ell_{v,m}): latency/staleness penalty.

A key diagnostic metric:

[ \text{Rebate Illusion Gap}_{v} = r_v - a_v ]

If this is negative, “high rebate venue” is economically worse than it looks.

3) Queue-aware fill probability model

For passive posting, estimate fill chance before decision horizon (\Delta):

[ p^{fill}{v}(\Delta) = 1 - \exp\left(-\int_0^{\Delta} \lambda^{take}{v}(t)\cdot g(Q^{ahead}_{v}(t), c_v(t), \mu_v(t)),dt\right) ]

Features:

• (Q^{ahead}): queue volume ahead of our order,
• (c_v): cancel/dequeue dynamics at best level,
• (\mu_v): opposing market-order intensity.

Practical implementation:

1. Discretize in 100–500ms buckets,
2. Fit hazard/logit for fill event by horizon (0.5s / 2s / 5s),
3. Keep symbol-time-of-day specific calibration,
4. Add regime tags (calm/news/open/close).

4) Adverse selection (toxicity) model

Model short-horizon markout after fill:

[ \text{Markout}_{\tau} = \alpha + \beta_1,\text{OFI} + \beta_2,\text{queue_imbalance} + \beta_3,\text{trade_sign_burst} + \beta_4,\text{quote_age} + \epsilon ]

Use multi-horizon targets (1s/5s/30s).
For routing, maintain a conservative aggregate:

[ a_v = w_1,\widehat{M}{1s} + w_2,\widehat{M}{5s} + w_3,\widehat{M}_{30s} ]

This captures “toxic fills” that often dominate nominal rebate gains.

5) Routing optimizer (online decision)

At each cycle, allocate child slices (x_{v,m}):

[ \min_{x} \sum_{v,m} x_{v,m},\mathbb{E}[C_{v,m}] + \lambda,\text{CVaR}_{95}(C) + \eta,\text{CompletionRisk} ]

subject to:

• (\sum_{v,m} x_{v,m} = X_{remain}),
• max participation and venue throttle caps,
• urgency/deadline constraints,
• compliance constraints (venue eligibility, order-type rules).

Control intuition:

• Low urgency + strong queue metrics → passive bias,
• Rising completion risk → shift toward midpoint/taker,
• Toxicity spike on a venue → temporary downweight despite rebate.

6) Production data contract (minimum)

Per child-order decision, log:

• timestamp, symbol, side, residual quantity,
• venue + mode candidates with score components,
• fee tier snapshot and expected rebate,
• queue-ahead estimate + fill probability by horizon,
• predicted markouts (1s/5s/30s),
• selected action and fallback ladder,
• realized fill path + realized all-in cost decomposition.

Without this schema, rebate-vs-toxicity attribution will be guesswork.

7) Validation protocol

Offline replay

Compare policies:

1. fee-only router,
2. fee+spread router,
3. full queue-aware all-in router (target).

Required out-of-sample wins:

• lower mean implementation shortfall,
• lower q95 tail cost,
• stable completion SLA under stress regimes.

Online rollout

• 5% shadow → 10% canary → 25% production tranche,
• automatic rollback if q95 shortfall or completion miss rate breaches threshold,
• weekly recalibration of queue/markout models.

8) Failure modes and fixes

1. Overfitting to rebate-heavy periods
   Fix: regime-stratified validation + stress windows.

2. Queue estimate drift after venue microstructure changes
   Fix: drift detector on fill-prob residuals; trigger rapid recalibration.

3. Markout leakage from delayed market data
   Fix: clock sync audit + latency-adjusted feature timestamps.

4. Completion failures from overly passive policy
   Fix: hard urgency floor and dynamic taker fallback ladder.

9) 30-day implementation slice

Week 1

• Build all-in cost decomposition in TCA store,
• Add fee-tier and rebate snapshots per fill.

Week 2

• Deploy queue-aware fill model (0.5s/2s/5s horizons),
• Backfill historical predictions for comparison.

Week 3

• Deploy markout toxicity model and venue scorecard,
• Start shadow routing decisions.

Week 4

• Canary rollout with automated rollback guards,
• Publish dashboard: rebate illusion gap, net cost by venue/mode, completion risk.

References

• Cont, R., & Kukanov, A. (2014). Optimal order placement in limit order markets. arXiv:1210.1625. https://arxiv.org/abs/1210.1625
• Nasdaq (2020). Quantifying the Cost of Maker-Taker Markets. https://www.nasdaq.com/articles/quantifying-the-cost-of-maker-taker-markets-2020-10-08
• Axon Trade (2025). Fees, Rebates, and Maker/Taker Math. https://axon.trade/fees-rebates-and-maker-taker-math
• hftbacktest docs. Probability Queue Position Models. https://hftbacktest.readthedocs.io/en/latest/tutorials/Probability%20Queue%20Models.html
• Almgren, R., & Chriss, N. (2000). Optimal Execution of Portfolio Transactions.

TL;DR

Routing by fee/rebate tables alone is a trap.
Model net slippage as a queue-aware all-in objective (fees, spread capture, toxicity, non-fill, latency), then optimize venue/mode allocation under completion risk constraints.

When done right, the desk stops “earning rebates and losing money.”