Locked/Crossed-Market Transition Slippage Playbook

Date: 2026-03-10
Category: research
Scope: Lit-equity execution in fragmented venues when NBBO repeatedly flips between normal, locked, and crossed states.

Why this matters

Most execution policies assume the inside market is stable enough that a quote-touch decision has predictable outcomes.

During locked/crossed bursts, that assumption breaks:

• displayed touch can be non-actionable for short intervals,
• queue value can evaporate on state transitions,
• retry loops inflate message traffic and silently increase implementation shortfall.

Result: “we only crossed a tick” decisions can compound into a large tail-cost invoice.

Core hypothesis

Slippage is strongly conditioned by market-state transition dynamics, not only static spread/depth.

A controller that models transition hazards (normal ↔ locked ↔ crossed) can reduce avoidable chase/retry behavior while preserving completion reliability.

Transition-state model

Define three observable states:

• N (Normal): best bid < best ask
• L (Locked): best bid = best ask
• X (Crossed): best bid > best ask

Track short-horizon transition probabilities:

[ P_{ij}(\Delta t)=P(s_{t+\Delta t}=j\mid s_t=i),\quad i,j\in{N,L,X} ]

Particularly important:

• (P_{LN}): lock resolution speed
• (P_{XN}): crossed resolution speed
• (P_{LX}, P_{XL}): instability loop risk

Signal stack

1) Transition Instability Score (TIS)

Weighted transition entropy + loop intensity:

[ \text{TIS}=w_1 H(P_{i\cdot}) + w_2(P_{LX}+P_{XL}) + w_3\cdot\text{flipRate} ]

Higher TIS means actionability is decaying.

2) Actionable Touch Ratio (ATR)

Fraction of intended touch interactions that become executable within latency budget (\tau):

[ \text{ATR}=\frac{#\text{actionable touch events within }\tau}{#\text{touch attempts}} ]

3) Resolution Half-Life (RHL)

Estimated median time for L/X to return to N.

4) Retry Amplification Factor (RAF)

[ \text{RAF}=\frac{#\text{child messages (new+cancel+replace)}}{#\text{intended slices}} ]

Captures control-plane churn tax.

5) Transition Markout Differential (TMD)

Difference between post-fill markout during transition states vs normal states:

[ \text{TMD}{h}=\mathbb{E}[\text{markout}{h}\mid L/X]-\mathbb{E}[\text{markout}_{h}\mid N] ]

Regime controller

State A — STABLE_N

• Low TIS, high ATR, short RHL.
• Baseline execution policy.

State B — FRAGILE_LOCK

• Lock frequency rising; L↔N flipping fast.
• Reduce unnecessary reprice churn.

State C — CHAOTIC_TRANSITION

• TIS high, RAF rising, ATR falling.
• Prioritize completion safety over micro-optimizing touch.

State D — SAFE_MODE

• Triggered by guardrail breach (tail budget, ATR floor, latency anomaly).
• Deterministic, conservative fallback schedule.

Use hysteresis + minimum dwell times to avoid state flapping.

Branch-cost objective

For action (a \in {join, improve, cross, pause}):

[ \mathbb{E}[C\mid a,s]=\underbrace{\mathbb{E}[\text{fill cost}\mid a,s]}{\text{direct}}+\lambda_1\underbrace{\mathbb{E}[\text{no-fill regret}\mid a,s]}{\text{deadline}}+\lambda_2\underbrace{\text{TailRisk}{q95}(a,s)}{\text{survival}}+\lambda_3\underbrace{\text{RAF penalty}}_{\text{control-plane tax}} ]

The key addition is explicit RAF penalty during unstable transition regimes.

Practical policy mapping

FRAGILE_LOCK

• Widen no-replace band around intended price.
• Increase minimum dwell before cancel/replace.
• Favor fewer, slightly larger passive attempts over rapid micro-repricing.

CHAOTIC_TRANSITION

• Cap retry rate and message burst size.
• Prefer controlled crossing at bounded cadence when ATR collapses.
• Disable tactics that rely on persistent touch priority.

SAFE_MODE

• Enforce hard participation caps and cooldown timers.
• Tighten per-minute slippage budget burn limits.
• Require recovery signals (ATR rebound + TIS drop) before exiting.

KPI pack

Track by symbol and session bucket:

• IS_bps (mean/p90/p95)
• ATR (actionable touch ratio)
• RAF (message amplification)
• RHL (lock/cross resolution half-life)
• State Occupancy (time in STABLE/FRAGILE/CHAOTIC/SAFE)
• Deadline Completion (residual notional at cut-off)

Promotion gate example:

• p95 IS improvement >= 6–10% in high-TIS buckets,
• RAF reduction >= 15%,
• no completion reliability degradation beyond SLA.

Data/engineering checklist

• Event-time NBBO + venue-level quotes/trades + child-order lifecycle events
• Deterministic replay for transition labeling and branch analysis
• Timestamp quality checks (feed age + local clock skew)
• Per-order state-at-decision logging for post-trade attribution
• Feature freshness guardrails (stale -> SAFE_MODE)

Common failure modes

1. Assuming lock = free spread: fills obtained in unstable locks often carry worse markout.
2. Overfitting calm sessions: thresholds fail when transitions cluster.
3. Ignoring control-plane cost: retry storms look “active” but burn edge.
4. No hysteresis: rapid regime toggling creates policy thrash.
5. Single-venue bias: transitions can be fragmented and asynchronous across venues.

2-week rollout

Week 1 (shadow)

• Label regimes and compute TIS/ATR/RAF/RHL.
• No behavior change.
• Validate state stability by symbol bucket.

Week 2 (canary)

• Enable retry caps + no-replace band in FRAGILE_LOCK/CHAOTIC_TRANSITION for small notional.
• Daily review on p95 IS, RAF, and residual completion.

Rollback if:

• p95 IS breaches tolerance,
• completion SLA degrades,
• SAFE_MODE occupancy remains persistently high.

Bottom line

Locked/crossed bursts are not just quote oddities—they are a separate execution microclimate with unstable actionability.

Modeling transition hazards + retry amplification turns a noisy state into a controllable policy input and helps prevent “tiny decision, big tail-cost” outcomes.