Execution-Schedule Randomization for Anti-Gaming: A Slippage Playbook

Date: 2026-03-09
Category: research (execution microstructure)

Why this matters

Many execution stacks still leak intent through over-regular child-order patterns:

• fixed time interval (e.g., every 5s)
• fixed slice size
• repetitive venue path
• deterministic urgency transitions

That predictability can be harvested by anticipatory flow, turning “clean schedule discipline” into hidden slippage tax.

The practical goal is not random behavior for its own sake. It is:

reduce exploitability of your footprint while preserving completion reliability.

Threat model: how predictable execution gets taxed

1) Timing predictability

If child orders arrive on a near-clock schedule, counterparties can pre-position around expected replenishment windows.

2) Size predictability

Constant child sizes make residual parent size inference easier.

3) Venue predictability

Repeated venue routing paths leak where your next child likely appears.

4) Urgency-state predictability

If your policy always “panics” at the same remaining-time threshold, counterparties can wait for your forced crosses.

Evidence stack (selected)

1. Order anticipation around predictable trades reports higher execution costs when child-order patterns are more predictable.
   Source: https://onlinelibrary.wiley.com/doi/10.1111/fima.12255

2. Predatory trading theory shows strategic traders can profit by anticipating forced/liquidity-demanding flow.
   Source: https://eprints.lse.ac.uk/24829/1/dp441.pdf

3. BIS FX execution algorithms report notes practical randomization in TWAP-style execution to reduce signaling/predictability.
   Source: https://www.bis.org/publ/mktc13.pdf

4. Baseline optimal-execution frameworks (e.g., Almgren–Chriss) are useful as cost-risk backbones, but production policy often needs anti-gaming overlays beyond deterministic trajectories.
   Source: https://cims.nyu.edu/~almgren/papers/optliq.pdf

Core design principle

Treat anti-gaming as a controlled entropy problem:

• too little entropy → easy to anticipate
• too much entropy → completion/impact efficiency degrades

So we optimize for a middle zone: low predictability, bounded execution risk.

Practical leakage metrics

Track these per parent order and intraday bucket:

1. Inter-Arrival Regularity (IAR)
   CV of child-order inter-arrival times (lower CV = more clock-like = more predictable)

2. Child Size Entropy (CSE)
   Entropy of slice-size distribution (very low entropy = easier inference)

3. Venue Concentration (VHHI)
   HHI of child fills/notionals across venues (high concentration = path leak)

4. Footprint Autocorrelation (FAC)
   Lag-1/lag-k autocorrelation of signed child participation rate

5. Predictability Score (PS)
   Out-of-sample score from a simple adversary model that predicts next child timing/size/venue from recent sequence

If the adversary model predicts you too well, you are leaking.

Objective function (desk-friendly form)

For policy action (a):

[ \min_a; \mathbb{E}[\text{IS} \mid a]

• \lambda_1 \cdot \text{LeakagePenalty}(a)
• \lambda_2 \cdot \text{CompletionRisk}(a) ]

Where:

• (\mathbb{E}[\text{IS}]): expected implementation shortfall
• LeakagePenalty: function of IAR/CSE/VHHI/FAC/PS
• CompletionRisk: underfill probability, deadline miss, end-of-window forced-cross risk

This keeps anti-gaming tied to actual PnL constraints, not a “randomize everything” ideology.

Three-state controller

STEALTH

Use when toxicity/anticipation risk is elevated and deadline slack exists.

• widen time jitter band
• increase size randomization (within cap)
• diversify venue path
• prefer passive/pegged variants where safe

BALANCED

Default mode.

• moderate jitter
• bounded size randomness
• venue rotation with liquidity scoring

URGENT

When residual risk dominates anti-gaming concerns.

• reduce randomness
• tighten schedule to hit completion SLA
• controlled aggression with explicit tail-cost guardrails

Use hysteresis for transitions to avoid mode flapping.

Implementation knobs (safe ranges to start)

1. Time jitter: random offset around schedule points (e.g., ±5–20% of base interval)
2. Size jitter: randomization around target slice (e.g., ±10–30%) with min/max caps
3. Venue randomization: probabilistic routing over top-k venues, weighted by live liquidity/toxicity
4. Order-type mix jitter: controlled blend of passive/marketable/post-only depending on regime
5. Urgency-threshold jitter: avoid deterministic “panic switch” timestamp

All knobs must be bounded by hard risk rules (max participation, max spread-cross, max residual at end).

Validation plan (before full rollout)

1. Shadow adversary test
   Train a predictor on child sequence features. Target: reduced next-child predictability.

2. Canary cohorts
   Compare deterministic baseline vs randomized policy on matched orders.

3. Joint success criteria

   • IS mean non-inferior or improved
   • q95/q99 IS not worse
   • Predictability Score reduced
   • completion SLA preserved

4. Failure gates

   • forced-cross rate spike
   • venue reject/retry surge
   • end-window residual accumulation

Common failure modes

1. Entropy overdose
   Too much randomization increases delay/opportunity cost.

2. Ignoring market state
   Randomization should be regime-aware, not static.

3. No adversary benchmark
   If you never test predictability explicitly, leakage remains invisible.

4. Single KPI obsession
   Reducing predictability alone can still hurt net execution quality.

Minimal rollout checklist

• Add leakage telemetry (IAR/CSE/VHHI/FAC/PS)
• Ship STEALTH/BALANCED/URGENT controller with hysteresis
• Define hard completion and tail-cost guardrails
• Run canary with adversary backtest + live monitoring
• Promote only if IS + tail + SLA + predictability all pass

Bottom line

Deterministic schedules are easy to operate, but often easy to game.

A good production execution stack treats predictability itself as risk exposure and applies bounded, state-aware randomization to reduce signaling tax without sacrificing completion discipline.