Decision-to-Arrival Drift & Opportunity-Cost Implementation-Shortfall Playbook

Date: 2026-03-15
Category: research
Focus: Separating execution slippage into delay, trading, and unfilled opportunity costs so desks can control the decision-to-arrival leak explicitly.

1) Why this failure mode matters

Many desks optimize child-order execution quality after order arrival, but ignore what happens before the order reaches the execution engine.

That blind spot creates a recurring leak:

signal fires -> PM decides -> ticket waits (approval/risk/routing) -> order arrives -> execution

If price drifts during the waiting interval, the desk often blames “market impact,” even though a meaningful part was actually decision-to-arrival delay cost.

Without explicit decomposition, teams over-tune execution algorithms while under-fixing upstream latency, risk checks, and ticket batching behavior.

2) Cost decomposition (decision vs arrival vs execution)

Let:

• (S): target shares,
• (\sum s_j): filled shares,
• (P_d): decision price,
• (P_0): arrival price (order entry benchmark),
• (P_{avg}): average execution price,
• (P_n): end-of-horizon price,
• fees: explicit costs.

A practical expanded decomposition (Perold/Wagner style) is:

[ IS = S(P_0-P_d) + (\sum s_j)(P_{avg}-P_0) + (S-\sum s_j)(P_n-P_0) + fees ]

Interpretation:

1. Delay Cost (S(P_0-P_d)): cost before trading starts,
2. Trade Cost ((\sum s_j)(P_{avg}-P_0)): cost from execution path,
3. Opportunity Cost ((S-\sum s_j)(P_n-P_0)): cost of unfilled quantity.

This is the minimum decomposition needed for operational accountability.

3) Modeling framework for production

3.1 Stage A — Delay-drift model (decision -> arrival)

Model signed drift during delay window:

[ \Delta P_{delay} = P_0 - P_d ]

Condition on:

• signal half-life / alpha decay,
• queueing and approval latency,
• symbol volatility burst state,
• market regime (open/close/news windows).

Use a probabilistic model (quantile GBM or Bayesian state-space) to estimate not just mean drift but tail delay risk.

3.2 Stage B — Execution model (arrival -> fills)

Condition trade cost on:

• participation rate,
• spread/depth,
• venue mix,
• urgency and child-order cadence,
• queue-position proxies.

3.3 Stage C — Fill-completion model (unfilled tail)

Model fill probability and residual risk to estimate expected opportunity cost under horizon constraints.

3.4 Unified objective

Optimize expected total shortfall, not isolated execution slippage:

[ \min ; \mathbb{E}[Delay + Trade + Opportunity + Fees] ]

If delay dominates, the right fix is upstream process redesign—not faster sweeping.

4) Feature set that usually moves the needle

4.1 Decision-to-arrival latency features

• ticket_to_riskcheck_ms
• riskcheck_to_route_ms
• route_to_ack_ms
• manual_touch_count
• approval_queue_depth
• parent_order_age_ms

4.2 Alpha-decay and urgency features

• signal_age_ms
• forecast_half_life_ms
• alpha_confidence
• event_proximity_score (earnings/macro/open-close)

4.3 Execution-path features

• spread z-score at arrival,
• depth-at-touch and refill half-life,
• participation trajectory vs plan,
• venue-level reject/retry patterns.

4.4 Completion/opportunity features

• residual shares at checkpoints,
• remaining time to deadline,
• passive-fill hazard,
• adverse markout probability if crossing now.

5) Desk KPIs (simple, actionable)

5.1 DSR — Delay Share Ratio

[ DSR = \frac{Delay\ Cost}{IS - fees + \epsilon} ]

Shows how much shortfall came before trading began.

5.2 OSR — Opportunity Share Ratio

[ OSR = \frac{Opportunity\ Cost}{IS - fees + \epsilon} ]

Flags under-completion and too-patient scheduling.

5.3 D95 — p95 Decision-to-Arrival Latency

Track by strategy, symbol-liquidity bucket, and trader workflow.

5.4 CDA — Cost of Delay per second

[ CDA = \frac{Delay\ Cost}{decision_to_arrival_seconds + \epsilon} ]

Useful for SLA negotiation with risk/OMS/EMS owners.

6) Control policy (regime-based)

NORMAL

• standard risk checks,
• default urgency model,
• normal participation caps.

DELAY_CRITICAL

Trigger when expected delay cost exceeds threshold:

• fast-lane routing for decaying-alpha orders,
• pre-approved risk envelopes for bounded notional,
• bypass nonessential manual touches,
• shorten child-order intervals early in schedule.

OPPORTUNITY_CRITICAL

Trigger when residual shares + time-left imply high unfilled risk:

• increase urgency ladder,
• widen passive/aggressive mix,
• dynamic cap uplift with hard slippage guardrails.

MARKET_STRESS

When spread/depth regime deteriorates:

• avoid panic crossing if delay source is internal and fixable,
• escalate to trader with explicit decomposition: “delay vs trade vs opportunity”.

7) Data contract requirements (often missing)

You need immutable timestamps for each parent order:

1. decision_ts
2. ticket_created_ts
3. risk_check_pass_ts
4. route_ts
5. arrival_ts
6. child-order/fill timeline
7. horizon_end_ts with end price snapshot

No reliable decision_ts -> no reliable delay attribution.

8) 30-day rollout plan

Week 1 — Measurement first

• instrument decision->arrival timestamps,
• produce daily decomposition dashboard (Delay/Trade/Opportunity),
• identify top-10 workflows by delay cost.

Week 2 — SLA + queue surgery

• set per-strategy latency SLOs,
• remove redundant approvals for low-risk flow,
• add delay alerts when CDA exceeds budget.

Week 3 — Policy activation

• enable DELAY_CRITICAL fast lane for selected books,
• A/B test against baseline schedules,
• compare IS decomposition, not headline IS only.

Week 4 — Scale and harden

• extend to all high-turnover strategies,
• codify exception handling and rollback criteria,
• review capacity with compliance/risk to keep controls auditable.

9) Common anti-patterns

• Benchmarking only from arrival and claiming “execution did fine.”
• Treating all slippage as market impact.
• Ignoring unfilled opportunity cost in end-of-day reporting.
• Optimizing VWAP slippage while alpha decays pre-arrival.
• Letting workflow latency drift because it is “not a trading algo problem.”

10) Bottom line

If you do not separate delay, trade, and opportunity costs, you cannot fix the true bottleneck.

For many real desks, the easiest slippage win is not smarter slicing—it is shrinking decision-to-arrival latency and governing completion risk with explicit decomposition-aware controls.