Protected-Quote Fade & Intermarket-Sweep Slippage Playbook

Date: 2026-03-05
Category: research
Domain: finance / execution / market microstructure / slippage modeling

Why this matters

When your router decides using a snapshot of protected quotes, those quotes may disappear before the child order reaches the venue.
That quote-fade gap turns expected spread capture into reject loops, misses, and worse sweep prices.

The practical failure mode:

You route for displayed liquidity that no longer exists, then pay escalation cost to finish.

1) Cost decomposition with fade-aware terms

Let:

• (P_0): decision-time benchmark price (arrival or decision mid)
• (P_{fill}): realized fill price
• (side\in{+1,-1}): buy/sell sign
• (C_{obs}): observed slippage (bps)

[ C_{obs}= side\cdot\frac{P_{fill}-P_0}{P_0}\times 10^4 ]

Define additional decomposition:

[ C_{obs}= C_{spread}+C_{impact}+C_{timing}+C_{fade}+C_{retry} ]

Where:

• (C_{fade}): cost from protected quote vanishing before arrival
• (C_{retry}): additional cost from reroute/retry escalation (IOC/FOK rejects, queue reset, urgency bump)

A useful operational metric:

[ \text{Fade Tax} = C_{fade}+C_{retry} ]

This is often hidden if TCA only reports top-level implementation shortfall.

2) Core fade-risk features

For each child-order decision, log:

• Quote Age (Q_{age}): ms since quote update at decision point
• Path Latency (L_{path}): decision → venue ingress latency estimate
• Book Churn (\chi): touch update rate (events/sec)
• Protected Depth Fragility (F_d): probability that top depth is canceled within (L_{path})
• Sweep Pressure (S_p): expected same-side demand arriving in the next 10–100ms

Define Fade Probability:

[ \pi_{fade}=\Pr(\text{quote unavailable at arrival}\mid Q_{age},L_{path},\chi,F_d,S_p) ]

Estimate with survival/hazard models or calibrated gradient-boosted probabilities.

3) Fade-aware expected cost model

At decision time for candidate route (r):

[ \mathbb{E}[C_r] = (1-\pi_{fade,r}),C^{hit}r + \pi{fade,r},\big(C^{miss}_r + C^{retry}_r\big) ]

Choose route minimizing tail-aware objective:

[ \min_r; \mathbb{E}[C_r] + \lambda,\mathrm{CVaR}_{95}(C_r) ]

This avoids overfitting to average outcomes while ignoring bursty fade episodes.

4) Execution policy state machine

STATE A — STABLE_TOUCH

Trigger: low (\pi_{fade}), low churn.

• normal passive/IOC mix
• standard venue ranking

STATE B — FRAGILE_PROTECTED

Trigger: medium/high (\pi_{fade}) on top-ranked venue.

• penalize stale quotes by route score
• reduce child size per attempt
• require fresher quote timestamp threshold

STATE C — RETRY_SPIRAL_RISK

Trigger: rising reject loop count + widening markout tails.

• cap retries per slice
• switch to deterministic fallback schedule
• tighten urgency only within bounded participation limits

STATE D — DISLOCATED_SWEEP

Trigger: multi-venue fade spike + depth collapse.

• safe mode: lower participation and wider completion horizon
• optional pause gate for non-urgent residuals
• manual oversight for large parent orders

Use hysteresis and minimum dwell times to prevent mode flapping.

5) Data contract (minimum viable)

• decision-time full quote snapshot (venue-level, with timestamps)
• per-venue clock sync metadata (NTP/PTP health)
• child-order lifecycle (send/ack/reject/fill/cancel)
• reject reason taxonomy (throttle, protection, no-liquidity, venue busy)
• per-attempt route path latency samples
• short-horizon markout ladder (100ms/1s/5s/30s)

Without aligned timestamps across market data + order events, fade attribution becomes unreliable.

6) Validation protocol

Offline counterfactual replay

• replay historical decisions with fade-aware routing scores
• compare baseline vs candidate on:
  • mean bps
  • p95 / CVaR95 slippage
  • reject-loop rate
  • completion ratio and duration

Shadow mode

• generate candidate decisions without sending live orders
• monitor disagreement rate and projected cost delta

Canary rollout

• start with low-notional symbols
• auto-rollback if any guardrail breaks:
  • markout tail deterioration
  • reject-loop spike
  • completion shortfall

7) Common failure modes

1. Quote timestamp blindness
   Router treats fresh and stale quotes equally.

2. Single-point latency assumptions
   Using average latency instead of percentile distribution understates tail fade risk.

3. Retry escalation without budget
   Repeated misses silently convert into marketable urgency.

4. Venue-agnostic fade model
   Fade dynamics are venue- and symbol-specific.

5. No reject taxonomy
   All misses look the same, so policy cannot improve the right failure mode.

8) Minimal implementation checklist

• Add fade-probability model (\pi_{fade}) to route scoring
• Log Fade Tax ((C_{fade}+C_{retry})) in daily TCA
• Add retry-budget guardrails per parent order
• Penalize stale quotes via quote-age thresholds by regime
• Deploy canary with p95/CVaR rollback gates
• Build symbol×venue fade heatmap dashboard

References to review

• Budish, E., Cramton, P., & Shim, J. (2015), The High-Frequency Trading Arms Race.
• Hasbrouck, J. (2018), High-Frequency Quoting: Short-Term Volatility in Bids and Offers.
• O’Hara, M. (2015), High Frequency Market Microstructure.
• Gatheral, J. (2010), No-Dynamic-Arbitrage and Market Impact.
• Cartea, Á., Jaimungal, S., & Penalva, J. (2015), Algorithmic and High-Frequency Trading.

If your router is fast but not fade-aware, it is still trading against a ghost book.