Intraday Liquidity-Drought Slippage Playbook

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

Why this matters

Most slippage models learn a smooth intraday volume curve and assume “midday is slower but predictable.”

In live trading, midday often behaves like a liquidity drought regime:

• displayed depth looks normal until touched,
• refill speed collapses,
• quote lifetimes shrink,
• completion pressure accumulates into the final hour.

If this regime is not explicitly modeled, desks under-price passive miss risk at noon and then overpay catch-up cost later.

Core concept: Liquidity Drought Cost (LDC)

Model midday execution as a branch problem:

[ \text{LDC} = \text{PassiveMissTax} + \text{RerouteTax} + \text{DeadlineCatchupTax} - \text{SpreadSavedByPatience} ]

Control objective:

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

where action (a) includes participation, passive/aggressive mix, and urgency schedule by time bucket.

Regime signals (drought detection)

Build a Liquidity Drought Score (LDS) from standardized features:

1. Refill Half-Life (RHL): time for top-of-book depth to recover after a touch.
2. Quote Survival Median (QSM): median lifetime of best quotes.
3. Executable Depth Ratio (EDR): realized fillable size / displayed size near touch.
4. Micro-Spread Fragility (MSF): probability spread widens within (\Delta t) after interaction.
5. Residual-Time Pressure (RTP): remaining parent size relative to time-to-deadline.

Suggested state map:

• HEALTHY: low LDS, normal refill
• DRYING: LDS rising, refill slowing
• DROUGHT: high LDS, passive miss/cancel risk elevated
• SAFE: uncertainty spike or budget breach; preserve completion

Use hysteresis to avoid flip-flopping.

Modeling stack

Layer 1 — Child-order branch model

Estimate:

• (P(\text{fill})), (P(\text{partial})), (P(\text{miss}))
• post-fill markout (multi-horizon)
• expected retry latency and queue-reset loss

Key features:

• LDS components,
• time-of-day bucket,
• local volatility and microprice drift,
• cancel intensity and depth convexity,
• parent residual and urgency.

Layer 2 — Parent completion model

Estimate cumulative parent cost under state trajectory:

• early patience savings,
• miss accumulation,
• terminal-window catch-up penalty.

This prevents a false “good average fill” conclusion when the endgame becomes expensive.

Policy design

1. State-aware passive cap
   In DRYING/DROUGHT, cap passive exposure and shorten passive dwell time.

2. Retry budget with cooldown
   Limit repeated passive retry bursts when QSM is low and MSF is high.

3. Midday anti-cliff pre-allocation
   Pull a fraction of parent completion earlier if drought probability rises.

4. Deadline-protect trigger
   Force deterministic completion mode when RTP crosses threshold.

5. Venue robustness tilt
   Prefer venues/routes with better historical refill resilience in drought states.

Data contract (minimum)

Per child order:

• decision/send/ack/fill/cancel timestamps,
• venue and route-hop history,
• top-N book snapshot at decision and effective interaction,
• fill status/size/price,
• markout ladder,
• parent residual + deadline metadata,
• reason codes (reject, timeout, cancel-ack lag).

Missing timestamp integrity makes drought attribution unreliable.

Calibration loop

Weekly

• Refit branch probabilities by symbol-liquidity cluster and time bucket.
• Re-estimate LDS thresholds and state transition probabilities.
• Stress-test on low-liquidity days (event + non-event).

Daily

Track by time bucket:

• q50/q95 slippage,
• completion ratio,
• passive miss rate,
• catch-up share (cost paid in terminal window).

Intraday guards

• If LDS stays above drought threshold for N windows: reduce passive cap.
• If catch-up share exceeds budget: move to DEADLINE_PROTECT/SAFE.

Rollout plan

1. Shadow (2 weeks): calculate LDS + simulated decisions only.
2. Canary (5–10% notional): enable state-aware passive cap.
3. Scale: widen symbol universe after q95/completion gates pass.
4. Rollback conditions: completion failures, elevated terminal catch-up tax, or state flapping.

Common failure modes

• Treating midday as a static low-volume bucket rather than a regime.
• Optimizing spread capture while ignoring miss-and-chase convexity.
• Letting retry loops consume queue priority and deadline budget.
• Calibrating only average bps and missing q95 deterioration.

Bottom line

Midday slippage is often a liquidity drought control problem, not a simple slower-volume problem.

Model refill fragility + miss accumulation explicitly, then drive execution with state-aware caps and deadline protection. This is how you avoid noon patience turning into end-of-day panic cost.