Index Futures Roll-Window Slippage Playbook

Date: 2026-03-10
Category: research (execution microstructure / futures roll operations)

Why this matters in production

A lot of desks model execution slippage as if liquidity is stationary. During index futures roll windows, that assumption breaks.

Liquidity migrates from front-month to next-month, calendar spread books become the true execution venue, and basis dynamics can move faster than single-leg impact models expect.

If you run size during the roll and still treat fills as “normal day + slightly wider spread,” you usually pay hidden costs in:

• legging drift
• queue reset churn
• forced late-day participation
• basis overshoot around crowded windows

Core objective

For a roll parent order, optimize net roll cost under deadline and risk constraints:

[ \min ; \mathbb{E}[C] + \lambda_{95},Q_{0.95}(C) + \lambda_T,\text{UnderfillPenalty} ]

where total cost is decomposed as:

[ C = C_{spread/fees} + C_{impact} + C_{basis,drift} + C_{legging} + C_{timing/deadline} ]

Cost decomposition you should track explicitly

1) Calendar spread execution cost (C_spread/fees)

• Fill in exchange-listed calendar spread book when possible
• Include spread level, fees, rebates, and queue position effects

2) Impact and resiliency (C_impact)

• Not just top-of-book spread; include depth slope and refill half-life
• Roll weeks often show brittle refill after aggressive clips

3) Basis drift cost (C_basis_drift)

• If execution is legged or delayed, fair-value basis movement becomes a real PnL driver
• Treat basis drift as stochastic, state-dependent risk

4) Legging risk (C_legging)

• Time gap between near-leg and far-leg completion
• Exposure to directional move and spread deformation during the gap

5) Deadline penalty (C_timing/deadline)

• Cost of residual inventory when you approach the hard roll cutoff
• This branch can dominate tail outcomes if you defer too much volume

State machine for roll execution

Use a simple operational regime controller:

1. PRE-ROLL (Green)
   Liquidity migration started but books are orderly.
2. CROWDED ROLL (Amber)
   Participation spikes, queue decay accelerates.
3. BASIS STRESS (Red)
   Basis z-score and spread volatility jump; legging risk dominates.
4. DEADLINE MODE (Safe)
   Completion priority over micro-optimization.

A practical trigger set works better than one “magic score.”

Monitoring metrics (minimum viable stack)

Roll Participation Congestion (RPC)

[ RPC_t = \frac{\text{Market roll volume in window }t}{\text{Expected roll volume baseline}} ]

High RPC implies crowded queue dynamics and faster adverse move after misses.

Calendar Depth Elasticity (CDE)

[ CDE_t = \frac{\Delta \text{visible spread depth}}{\Delta \text{aggressive traded size}} ]

Low/negative elasticity means depth does not heal after impact.

Basis Dislocation Z-score (BDZ)

[ BDZ_t = \frac{basis_t - \mu_{basis,t}}{\sigma_{basis,t}} ]

Track intraday rolling reference, not stale daily averages.

Legging Exposure Time (LET)

[ LET = t_{second;leg;fill} - t_{first;leg;fill} ]

Model E[cost | LET]; tail LET drives ugly episodes.

Residual Deadline Pressure (RDP)

[ RDP_t = \frac{\text{remaining qty}/\text{remaining minutes}}{\text{current executable pace estimate}} ]

RDP > 1 means you are behind schedule and likely entering forced behavior.

Modeling approach

A) Branch model for execution path

At each decision point, estimate branch probabilities:

• P(spread fill)
• P(legged fill)
• P(partial + retry)
• P(deadline forced cross)

Then compute expected and tail cost per branch:

[ \hat C = \sum_b P(b|x_t),\mathbb{E}[C_b|x_t] ]

Use quantile models per branch (q50/q90/q95), not only means.

B) Deadline-aware policy layer

Map regime + RDP to action set:

• clip size
• passive/urgent mode
• spread-book priority vs leg fallback
• max tolerable LET

C) Anti-churn constraint

Avoid queue-reset spam by adding hysteresis:

• minimum dwell time before repricing
• max replace rate per minute
• venue retry cooldown

Practical control policy (example)

PRE-ROLL (Green)

• Prefer passive spread-book posting
• Small clips, low replace frequency
• Minimize information leakage

CROWDED ROLL (Amber)

• Increase urgency gradually with RDP
• Tighten max LET threshold
• Shift from pure passive to mixed passive/aggressive schedule

BASIS STRESS (Red)

• Prioritize atomic spread fills where possible
• Reduce legging tolerance sharply
• Enforce stricter per-slice tail budget

DEADLINE MODE (Safe)

• Completion first
• Hard caps on residual carry into close
• Accept higher temporary impact to avoid catastrophic underfill branch

Backtest and replay design

Use historical roll days as first-class episodes, not random sample days.

Minimum protocol:

1. Reconstruct spread-book + leg-book timestamps at decision granularity
2. Replay policy with realistic queue/latency assumptions
3. Include reject/retry and partial-fill branches
4. Score both median and q95 cost
5. Add completion reliability KPI at deadline

Do not promote policies that improve average cost by sacrificing completion in stress windows.

Roll-day dashboard checklist

• Current regime (Green/Amber/Red/Safe)
• RPC, CDE, BDZ, LET, RDP live values
• Remaining quantity vs expected executable pace
• Branch-cost forecast (median + q95)
• Replace/retry rate and queue-reset diagnostics
• Hard cutoff countdown and residual risk projection

Common failure modes

1. Treating roll as normal session liquidity
2. Using only single-leg impact models
3. Ignoring LET tails in policy optimization
4. Late deadline panic from under-trading early windows
5. Over-replacing orders and burning queue priority

References (selected)

• CME Group education notes on index futures roll mechanics and calendar spreads
• Hasbrouck, J. (2007). Empirical Market Microstructure.
• Cartea, Á., Jaimungal, S., Penalva, J. (2015). Algorithmic and High-Frequency Trading.
• Gatheral, J., Schied, A. (2013). Dynamic execution and market impact frameworks.
• Almgren, R., Chriss, N. (2000). Optimal execution under impact/risk tradeoff.

Bottom line

Roll windows are a distinct microstructure regime.

A production-grade slippage model should explicitly price calendar spread liquidity migration, basis drift, legging exposure, and deadline pressure—otherwise “average bps” can look fine right until roll day tails dominate realized PnL.