Auction-Imbalance Publication Cadence Aliasing Slippage Playbook

Date: 2026-03-15
Category: research
Focus: Modeling and controlling slippage when child-order dispatch cadence phase-locks to auction-imbalance publication intervals, causing stale-signal chasing and avoidable auction transition cost.

1) Why this failure mode matters

Auction execution logic often consumes imbalance snapshots (size, direction, indicative price, matchable volume) on a periodic publication schedule.

A common hidden assumption is: “fresher snapshot = better decision.”

In practice, the failure mode is subtler:

• imbalance feed updates at interval (\Delta_f),
• strategy dispatches child updates at interval (\Delta_d),
• if (\Delta_d) becomes harmonically aligned with (\Delta_f), dispatch decisions repeatedly sample the same phase of the auction information cycle.

That creates cadence aliasing:

• stale or phase-biased interpretation of imbalance trajectory,
• overreaction to periodic prints,
• repeated cancel/replace bursts right before the uncross,
• higher queue reset and transition slippage at the auction boundary.

This is not a pure “bad model” issue. It is a control-loop timing issue that turns good signals into bad actions.

2) Mechanism map

2.1 Core timing geometry

Let:

• (t_k): k-th dispatch decision timestamp,
• (u_j): j-th imbalance publication timestamp,
• (a_k = t_k - \max{u_j: u_j \le t_k}): imbalance age at decision.

If dispatch cadence synchronizes with publication cadence, (a_k) distribution collapses into a narrow band instead of spreading across ([0, \Delta_f)).

Narrow-band sampling means the strategy repeatedly "looks" at auction state from one timing angle.

2.2 Slippage branches

For each decision near auction cutover, two dominant branches:

1. Stable branch: imbalance trajectory remains consistent until next decision.
2. Flip branch: imbalance sign/magnitude regime changes before next action, making current action stale.

Expected incremental cost:

[ \mathbb{E}[C_{inc}] = p_{stable} C_{stable} + p_{flip} C_{flip} ]

Cadence aliasing increases (p_{flip}) and widens ((C_{flip} - C_{stable})), especially in final auction minutes.

2.3 Why this amplifies at the boundary

As uncross approaches:

• imbalance variance rises,
• publication events become more information-dense,
• decision latency budget shrinks,
• cancel/replace queue-loss cost becomes nonlinear.

So fixed-cadence logic that is "fine" earlier can become expensive exactly when auction sensitivity is highest.

3) Cost decomposition

Decompose auction-window execution cost as:

[ C_{auction} = C_{signal} + C_{queue} + C_{transition} + C_{catchup} ]

Where:

• (C_{signal}): wrong-side or stale-side placement cost,
• (C_{queue}): cancel/replace priority reset cost,
• (C_{transition}): uncross-to-continuous handoff cost,
• (C_{catchup}): urgency tax from residuals created by stale reactions.

Under cadence aliasing:

[ C_{auction} = C_{baseline} + C_{alias}\quad,\quad C_{alias} = C_{phase_bias} + C_{flip_chase} ]

Primary objective is minimizing (C_{alias}) p95/p99 without sacrificing completion reliability.

4) Feature set for modeling

4.1 Timing-alignment features

• imbalance_pub_interval_ms (rolling estimate)
• dispatch_interval_ms (effective, not configured)
• phase_offset_ms = (dispatch_ts - last_pub_ts)
• phase_lock_score (circular concentration of phase offsets)
• imbalance_age_ms

4.2 Trajectory instability features

• imbalance_delta_1pub, imbalance_delta_3pub
• sign_flip_rate_30s
• indicative_price_jump_ticks
• uncross_distance_seconds

4.3 Action-footprint features

• cancel_replace_rate_auction_window
• queue_age_loss_estimate
• residual_notional_ratio
• post_uncross_catchup_participation

4.4 Market context features

• spread/depth around indicative reference,
• volatility burst indicators,
• cross-venue auction stress proxies,
• symbol/event tags (index rebalance, expiration, macro release windows).

5) Operational metrics

5.1 PLI — Phase Lock Index

Measure concentration of phase offsets (circular statistic):

[ PLI = \left|\frac{1}{N}\sum_{k=1}^{N} e^{i\theta_k}\right|\quad,\quad \theta_k = 2\pi\frac{phase_offset_k}{\Delta_f} ]

• near 0: well-dispersed sampling,
• near 1: dangerous phase lock.

5.2 IAS — Imbalance Age Skew

[ IAS = \frac{Q90(a_k)-Q10(a_k)}{\Delta_f+\epsilon} ]

Low IAS + high PLI indicates narrow-phase sampling (aliasing risk).

5.3 FCR — Flip-Chase Rate

Fraction of actions where imbalance regime flips before next actionable decision:

[ FCR = \frac{#(flip\ before\ next\ decision)}{#(decisions)} ]

5.4 AAT — Auction Aliasing Tax

[ AAT = \frac{C_{alias}}{executed_notional} ]

Track by symbol bucket, auction window segment, and tactic version (control vs treatment).

6) Control state machine

DESYNC_HEALTHY

• low PLI,
• acceptable FCR,
• normal auction tactics.

PHASE_LOCK_WATCH

Triggered by PLI or IAS warning thresholds.

Controls:

• inject bounded dispatch jitter,
• reduce cancel/replace aggressiveness,
• increase trajectory smoothing weight.

ALIASING_ACTIVE

Triggered by sustained high PLI + elevated FCR/AAT.

Controls:

• enforce phase-randomized dispatch windows,
• cap replace frequency near uncross,
• switch to robust residual-protection policy,
• increase uncertainty penalty on imbalance-driven directional actions.

SAFE_TRANSITION

When boundary risk remains high near uncross:

• prioritize completion safety over micro-optimization,
• freeze nonessential repricing loops,
• preserve queue priority unless hard constraints are hit.

Return to normal only after hysteresis windows confirm PLI/FCR normalization.

7) Practical modeling workflow

1. Reconstruct event timeline at ms granularity

   • imbalance publications,
   • dispatch decisions,
   • order acks/fills/cancels,
   • uncross marker.

2. Compute phase features per decision

   • phase offset, age, rolling PLI/IAS.

3. Label stale-trajectory episodes

   • flip-before-next-action events,
   • high queue-reset clusters,
   • post-uncross catch-up bursts.

4. Estimate incremental aliasing cost

   • matched control windows with similar volatility/liquidity but low PLI.

5. Build branch-aware predictor

   • estimate (p_{flip}) and conditional branch cost.

6. Optimize policy under tail constraints

   • target p95 AAT and completion-rate guardrails,
   • require non-regression in transition residual risk.

8) 30-day rollout plan

Week 1 — Instrumentation

• add publication/dispatch phase telemetry,
• log effective (actual) dispatch intervals,
• add per-decision aliasing feature snapshot.

Week 2 — Shadow analysis

• compute PLI/IAS/FCR/AAT by symbol/session,
• identify high-risk cadence pairs,
• validate auction-window branch attribution quality.

Week 3 — Canary controls

• deploy bounded jitter + replace caps to 5–10% flow,
• compare p95 AAT, residual ratio, completion reliability,
• tune hysteresis to avoid control flapping.

Week 4 — Scale + runbook

• expand to full auction-sensitive universe,
• add alerting on sustained PLI spikes,
• codify incident workflow: detect → dampen → transition-safe mode → review.

9) Common anti-patterns

• Using fixed dispatch cadence because it is "operationally clean."
• Measuring only average auction slippage, ignoring p95 transition tails.
• Treating imbalance age as enough, while ignoring phase concentration.
• Allowing unlimited cancel/replace loops close to uncross.
• Tuning on calm days only, then overpaying during rebalance/expiry stress.

10) Bottom line

Auction imbalance signals are not just values; they are time-structured observations.

When dispatch cadence phase-locks to publication cadence, the strategy can repeatedly act on a biased slice of reality. That timing aliasing becomes real slippage through stale chasing, queue resets, and transition catch-up.

Modeling phase risk (PLI/IAS/FCR/AAT) and controlling cadence adaptively turns a subtle timing artifact into an explicit, manageable execution risk.