Speed-Bump Venue Routing Slippage Playbook

Date: 2026-03-10
Category: research (quant execution / slippage modeling)

Why this matters

Speed-bump venues (intentional micro-delays on order handling) can improve execution quality by filtering toxic, race-driven flow.

But they also introduce a hidden tradeoff:

• Pros: less adverse selection on fills
• Cons: more miss risk, queue-aging, and reroute churn

If you model only average fill price, speed-bump routing can look great while completion reliability quietly degrades.

Core concept: Speed-Bump Net Slippage (SBNS)

Treat delayed-venue routing as a branch decision problem:

[ \text{SBNS} = \text{AdverseSelectionSaved} - (\text{DelayMissCost} + \text{RerouteCost} + \text{DeadlinePenalty}) ]

Equivalent portfolio view:

[ \min_a; \mathbb{E}[\text{Cost}|a] + \lambda,\text{CVaR}_{95}(\text{Cost}|a) + \eta,\text{CompletionRisk}(a) ]

where action (a) includes delayed-venue allocation, fallback trigger, and urgency schedule.

Mechanism map

Path A: Delay protects you

• Toxic fast flow moves first.
• Your delayed interaction avoids the worst prints.
• Markout improves.

Path B: Delay makes you late

• Displayed liquidity vanishes during delay window.
• Fill probability drops.
• Residual order reroutes in a worse local state.
• End-of-parent catch-up becomes expensive.

You need both paths in one model.

Observable metrics

1) Speed-Bump Participation Ratio (SPR)

[ \text{SPR} = \frac{\text{notional routed to delayed venues}}{\text{parent notional}} ]

2) Delay Markout Benefit (DMB)

Difference in post-fill markout (e.g., 1s/5s) between delayed and non-delayed route cohorts after risk matching.

3) Delay Miss Rate (DMR)

[ \text{DMR} = \frac{#\text{delayed-route attempts with zero fill}}{#\text{delayed-route attempts}} ]

4) Reroute Bounce Cost (RBC)

Incremental bps paid after delayed attempt fails and residual is rerouted aggressively.

5) Late Completion Share (LCS)

Fraction of parent slippage realized in final urgency window due to earlier delayed-route misses.

Modeling approach

Use a two-layer model.

Layer 1: Attempt-level branch model

Estimate for each routed child order:

• (P(\text{fill} | \text{delayed route}))
• (P(\text{miss} | \text{delayed route}))
• expected markout conditional on fill

Features:

• quote age / spread / touch depth,
• cancel intensity + refill half-life,
• micro-volatility and microprice drift,
• parent residual ratio + time-to-deadline,
• venue delay class and recent route-specific success.

Layer 2: Parent-level completion model

Condition on route-episode trajectory:

• branch cost from misses,
• branch cost from reroute escalation,
• terminal completion penalty under deadline pressure.

Optimize routing mix for tail-safe completion, not just average first-fill quality.

Execution state machine

• BALANCED: moderate toxicity, stable completion; use mixed routing.
• TOXIC_FAST_TAPE: markout risk high; increase delayed-route share with bounded miss budget.
• MISS_ACCUMULATION: delayed misses and residual pressure rising; reduce delayed share, widen fallback.
• DEADLINE_PROTECT: completion risk dominates; prioritize deterministic completion routes.
• SAFE: protect risk budget when uncertainty spikes.

Use hysteresis + minimum dwell time to avoid mode flapping.

Practical policy knobs

1. Delayed-route cap by urgency
   Cap delayed share when residual/time ratio crosses threshold.

2. Miss-budget stop
   If DMR or RBC exceeds intraday budget, cut delayed allocation.

3. Adaptive fallback latency
   Shorten wait-to-reroute under rising deadline pressure.

4. Venue diversification
   Avoid repeated delayed misses on a single path.

5. Terminal-window guardrail
   Pre-switch to completion mode before final window to avoid panic cross.

Data contract (minimum)

Per child order:

• decision/send/ack/fill/cancel timestamps,
• intended venue class (delayed vs non-delayed),
• route hop history,
• quote/depth snapshot at decision and effective interaction time,
• fill status and size,
• post-fill markout ladder,
• parent residual + urgency features,
• reroute reason code.

Without route-hop and timestamp integrity, delay benefit attribution is mostly noise.

Calibration cadence

Weekly

• Refit fill/miss branch probabilities by liquidity bucket.
• Re-estimate DMB and RBC surfaces.
• Recalibrate state thresholds.

Daily

• Monitor SPR/DMR/RBC/LCS by symbol and time bucket.
• Compare delayed-route cohort vs control on q95 cost and completion.

Intraday guardrails

• If DMR + RBC exceed threshold for N windows -> force MISS_ACCUMULATION.
• If residual/time crosses critical level -> force DEADLINE_PROTECT.

Rollout plan

1. Shadow (2 weeks): compute route-branch metrics only.
2. Canary (5–10%): apply miss-budget stop + adaptive fallback.
3. Scale by cohort: liquid names first, then thinner books.
4. Promotion gates:
   • q95 slippage non-inferior/improved,
   • completion reliability stable,
   • RBC and LCS reduced.

Rollback if completion misses or tail-cost drift rises.

Common failure modes

• Treating delay venues as always “better quality.”
• Over-allocating delayed routes near deadline.
• Ignoring reroute bounce cost after delayed misses.
• Optimizing markout while silently losing completion discipline.

Bottom line

Speed-bump routing is a regime-dependent edge, not a default setting.

Model delay benefit and delay penalty in the same control loop, and govern allocation with miss budgets + deadline-aware fallback. That is how delayed-venue alpha survives real completion constraints.