Half-Day Session Compression Slippage Playbook

Date: 2026-03-14
Category: research (execution microstructure)
Scope: Slippage modeling and execution-control design for early-close / shortened sessions (e.g., 13:00 ET close)

TL;DR

Half-day sessions are not just “normal days with fewer bars.”

They compress the intraday liquidity clock, accelerate close-imbalance dynamics, and can trigger late-day crowding into thinner pre-close windows. If you run full-day priors on half-days, you usually under-forecast urgency and overestimate passive completion reliability.

Core fixes:

• introduce session-length-aware time normalization,
• estimate separate half-day volume curves and auction-share priors,
• tighten deadline-aware execution controls before the compressed close,
• monitor a dedicated Compression Tax metric.

1) Problem Statement

Many production execution stacks assume stable intraday structure:

• fixed session length,
• broadly reusable U-shaped volume priors,
• close-phase behavior similar to normal days.

On half-days (e.g., US equities early close at 13:00 ET on specific holidays), these assumptions fail:

1. Volume-time mapping shifts (same wall-clock minute has different urgency meaning).
2. Close auction/imbalance uncertainty arrives earlier and steeper.
3. Liquidity replenishment cadence changes (fewer cycles before hard cutoff).
4. Fallback options shrink (less time for passive retries before forced completion).

Net effect: latent slippage tax in q90/q95 tails, often hidden by aggregate daily averages.

2) Mechanism: Why Compression Raises Costs

Let:

• T_full = standard session duration,
• T_half = shortened session duration,
• c = T_full / T_half = compression factor (>1 on half-days),
• tau = elapsed_time / session_length = normalized session time in [0,1].

If your model uses wall-clock features trained on full days, then at real time t:

• urgency is under-estimated because tau_half(t) > tau_full(t),
• “time remaining” is overstated,
• passive queue optionality is overstated,
• close participation target is started too late.

This creates a deadline convexity effect: small delays near tau -> 1 produce disproportionately large market-impact jumps.

3) Data Contract Additions (Minimal)

3.1 Session Regime Metadata

Per venue/date:

• isShortSession (bool)
• sessionOpenTs, sessionCloseTs
• scheduledAuctionCutoffTs (where applicable)
• holidayTag / reason code

3.2 Intraday Shape Layer

Per symbol bucket (or symbol-venue):

• historical intraday volume curves conditional on isShortSession
• close-auction share prior distributions (mean + dispersion)
• imbalance-message time evolution templates near close

3.3 Execution Layer

Per parent/child:

• normalized start time tau_start
• normalized completion time tau_done
• fallback activation timestamps
• pre-close residual inventory path

Without explicit short-session tags, calibration silently blends incompatible regimes.

4) Feature Engineering

4.1 Replace Wall-Clock with Session-Clock

Use normalized time as first-class feature:

• tau_now = (now - open) / (close - open)
• tau_remaining = 1 - tau_now
• phase = {open, mid, preclose, auction} in normalized time bands

4.2 Compression-Aware Urgency Features

• compressionFactor = T_full / T_session
• urgencyEffective = urgencyBase * f(compressionFactor, tau_remaining)
• residualToExpectedFlow = residualQty / expectedFlowRemaining(tau_now, regime)

4.3 Auction Risk Features

• auctionSharePrior_mean, auctionSharePrior_std (short-session specific)
• imbalanceSlope over recent auction updates
• pairedQtyAcceleration

4.4 Recovery-Capacity Features

• retryBudgetRemaining (max realistic retry cycles until cutoff)
• passiveDwellCapacity (estimated queue dwell opportunities left)
• forcedCrossProbability

5) Model Architecture

5.1 Baseline

Standard slippage model (spread/volatility/imbalance/urgency/venue).

5.2 Regime Overlay

Add interactions:

• isShortSession × tau_now
• isShortSession × residualToExpectedFlow
• isShortSession × imbalanceSlope
• isShortSession × childAggressiveness

5.3 Objective

Optimize tail-aware targets (q90/q95), not just mean bps.

Half-day failures concentrate in tails due to compressed fallback window.

6) Real-Time Control State Machine

State A: NORMAL

• No short-session flag.

State B: SHORT_SESSION_ACTIVE

Trigger: isShortSession=true.

Actions:

• switch to short-session volume/auction priors,
• increase minimum progress trajectory in normalized time,
• reduce passive dwell caps earlier than usual.

State C: COMPRESSION_STRESS

Triggers:

• residual-vs-flow gap breach,
• imbalance acceleration breach,
• projected completion miss probability above threshold.

Actions:

• step up aggressiveness in bounded increments,
• widen eligible liquidity destinations (within policy),
• pre-arm auction participation earlier.

State D: SAFE_COMPLETION

Trigger: hard cutoff risk high.

Actions:

• prioritize completion certainty,
• cap additional queue-risk exposure,
• enforce tail-budget protection logic.

Use hysteresis to avoid oscillation between B/C in noisy pre-close windows.

7) Control Levers That Usually Work

1. Time-Normalized POV/TWAP
   Drive schedule by tau, not wall-clock.

2. Pre-Close Inventory Guardrail
   Hard cap residual inventory at key normalized checkpoints (e.g., tau=0.85, 0.92).

3. Adaptive Passive Dwell
   Shrink maximum passive dwell as compression and imbalance risk rise.

4. Earlier Auction Decision Gate
   Move auction-join decision earlier on short sessions.

5. Residual Convexity Penalty
   Penalize plans that defer too much quantity into final compressed window.

8) Monitoring Metrics (Dedicated Dashboard)

• q50/q90/q95 slippage: short vs regular sessions
• completion miss probability calibration by tau
• residual inventory profile at normalized checkpoints
• forced-cross ratio in final phase
• auction participation error (planned vs realized)
• Compression Tax (CTAX):
  • CTAX = realized_slippage - counterfactual_slippage_if_full_session_model_corrected
  • track by symbol bucket and venue

If CTAX drifts positive for multiple short sessions, regime handling is stale.

9) Validation Protocol

Offline

• split backtests by isShortSession and by symbol-liquidity buckets,
• compare baseline vs compression-aware policy on q95 cost + completion,
• include stress windows (index rebalance proximity, high volatility).

Shadow Online

• run old/new policies in parallel decisions,
• monitor decision divergence and projected deadline misses.

Canary

• small notional rollout on short sessions only,
• rollback gates on q95 degradation, completion miss, and control churn.

10) Common Failure Modes

1. “Same curve, fewer minutes” shortcut
   Misses structural shift in close-phase behavior.

2. Wall-clock features only
   Creates hidden urgency miscalibration.

3. No short-session auction priors
   Auction participation becomes reactive, not planned.

4. Late fallback activation
   Leaves insufficient retry budget before cutoff.

5. Average-only KPI governance
   Tail damage stays invisible.

11) Deployment Checklist

• session calendar + early-close metadata integrated at runtime
• normalized session-clock features in production
• short-session-specific volume and auction priors calibrated
• compression-state controller enabled with hysteresis
• residual checkpoint guardrails configured
• q95 + completion canary rollback policy active
• CTAX dashboard live and reviewed after every short session

12) Key Takeaway

Half-day sessions are a distinct execution regime with compressed optionality.

Treating them as ordinary days causes systematic under-urgency and late forced impact. A session-clock-aware model plus compression-specific controls is usually enough to recover most of the hidden tax.

References (starting points)

• NYSE, Holidays & Trading Hours (early close schedule examples)
  https://www.nyse.com/trade/hours-calendars
• Traders Magazine, Closing Volume Discovery (closing-auction share context)
  https://www.tradersmagazine.com/departments/ecns_and_exchanges/closing-volume-discovery/