Closing Auction Extension-Uncertainty Slippage Playbook

Why this matters

Most close execution logic assumes a fixed endpoint:

• expected uncross time is known,
• imbalance trajectory near the bell is stable enough,
• residuals can be cleaned in a predictable post-close path.

In practice, some venues can delay/extend the close auction window when imbalance or price-formation stress is high. That uncertainty creates a hidden slippage tax:

1. overpaying urgency too early (crossing spread before an extension would have improved pairing),
2. waiting too long and getting trapped with convex residual cost,
3. repeated reprice/cancel churn while the endpoint keeps moving.

The key is to treat auction-end time as a random variable, not a constant.

Core failure mechanism

Define:

• T0: scheduled uncross time,
• DeltaT_ext: random extension duration (>= 0),
• T* = T0 + DeltaT_ext: realized uncross time.

If policy is built around T0 only, urgency is systematically misallocated.

A simple extension-hazard representation:

h_ext(t) = f(|I_t|, dI/dt, sigma_auction_t, collar_distance_t, reject_rate_t)

Where:

• I_t: auction imbalance size/sign,
• dI/dt: imbalance velocity,
• sigma_auction_t: short-horizon auction-price volatility,
• collar_distance_t: distance to auction price guardrails,
• reject_rate_t: order-reject/update-friction proxy near cutoff.

When hazard rises, the value of immediate aggression vs waiting flips quickly.

Slippage branch decomposition

Let expected cost be:

E[Cost] = P(no_ext)*C_no_ext + P(ext)*(C_wait_ext + C_reprice_churn + C_residual_convex)

Where:

• C_no_ext: baseline close execution cost when uncross occurs on schedule,
• C_wait_ext: adverse drift while capital is still exposed during extension,
• C_reprice_churn: queue resets + reject/retry friction while endpoint is uncertain,
• C_residual_convex: late catch-up cost if residual remains after uncertain close.

Operationally: extension risk is both a timing risk and a cost-convexity amplifier.

Metric stack

1) Extension Hazard Score (EHS)

Model-implied probability of extension within the next decision interval.

• High EHS = fixed-endpoint assumptions are unsafe.

2) Time-to-Uncross Uncertainty (TTU)

Dispersion (e.g., p90-p10) of predicted remaining time to uncross.

• Wide TTU = urgency policy should avoid brittle one-shot actions.

3) Residual Convexity Pressure (RCP)

Marginal expected cost increase per residual unit if the close resolves unfavorably.

• Rising RCP = residual inventory is becoming expensive nonlinearly.

4) Auction Churn Tax (ACT)

Cost attributed to cancel/replace/reject loops during uncertain close windows.

• High ACT = micro-updates are paying queue-reset tuition.

5) Extension-Regret Gap (ERG)

Counterfactual gap between:

• policy chosen with fixed T0,

• policy chosen with stochastic T*.

• Persistent positive ERG = model architecture is missing extension uncertainty.

Control policy state machine

STATE 1 — SCHEDULED_STABLE

Conditions:

• low EHS,
• narrow TTU,
• low ACT.

Policy:

• standard close schedule,
• normal imbalance-responsive participation.

STATE 2 — EXTENSION_WATCH

Conditions:

• EHS rising,
• TTU widening,
• imbalance velocity unstable.

Policy:

• reduce over-frequent repricing,
• stagger child placements,
• pre-allocate residual fallback paths (auction-safe alternatives).

STATE 3 — EXTENSION_ACTIVE

Conditions:

• extension confirmed or effectively certain,
• elevated ACT and RCP.

Policy:

• switch from static cutoff logic to hazard-conditioned pacing,
• preserve queue quality over cosmetic micro-updates,
• cap residual-risk growth with explicit tail-budget gates.

STATE 4 — SAFE

Conditions:

• repeated extension/churn with deteriorating fill certainty,
• residual risk near hard limits.

Policy:

• stop aggressive churn tactics,
• execute through pre-approved low-regret completion ladder,
• prioritize completion reliability and exposure containment.

Use asymmetric hysteresis to avoid state flapping.

Modeling pattern (production)

1. Extension survival model

   • estimate P(DeltaT_ext > tau | features_t) continuously,
   • keep calibration by symbol-liquidity bucket.

2. Conditional uncross-cost model

   • train separate no-extension vs extension-conditioned slippage surfaces,
   • blend by real-time extension probability.

3. Policy optimizer with residual convexity term

   • objective should include mean + q95 cost and residual completion penalty,
   • avoid average-cost policies that silently leak tail risk.

4. Replay with synthetic extension shocks

   • inject delayed uncross scenarios into historical close windows,
   • verify controller enters EXTENSION_WATCH/ACTIVE before tail blowups.

Practical rollout checklist

• Add EHS/TTU/RCP/ACT dashboard per symbol tier.
• Implement state machine with manual override and incident logging.
• Canary on small fraction of close flow before wider rollout.
• Track ERG weekly (fixed-time baseline vs uncertainty-aware policy).
• Define SAFE-entry hard limits tied to residual and tail-budget burn.

Bottom line

Close execution risk is not only about imbalance size; it is about endpoint certainty.

If your router behaves as if auction time is fixed when extension risk is regime-dependent, you will pay hidden slippage through urgency mistakes, churn, and late residual convexity. Model the close as stochastic in both price and time, and execution behavior stays robust exactly when auction mechanics become unstable.