Frequent-Batch Auction (FBA) Hybrid Routing Slippage Playbook

Modeling Hidden Cost When Continuous Routers Interact with Periodic-Auction Venues

Why this note: Many execution stacks are tuned for continuous limit-order books (CLOBs), but liquidity increasingly appears in short, repeated periodic auctions. If your router treats those venues like continuous books, you get systematic miss-and-chase slippage.

1) Failure Mode in One Sentence

A router optimized for continuous fills can underprice batch-timing risk in periodic auction venues, causing repeated missed uncrosses, toxic fallback aggression, and tail slippage inflation.

2) Cost Decomposition (Add the Missing Terms)

For a parent order, model expected implementation shortfall as:

[ C = C_{spread} + C_{impact} + C_{timing} + C_{batch_miss} + C_{fallback} + C_{deadline} ]

Where:

• (C_{batch_miss}): opportunity cost of missing favorable batch uncrosses.
• (C_{fallback}): cost of rerouting residual to lit/continuous venues after miss.
• (C_{deadline}): convex urgency cost if repeated misses compress remaining schedule.

Most systems have first three terms. FBA-aware systems must model all six.

3) Key Telemetry (Per Symbol × Venue × Parent)

1. Batch Participation Eligibility Rate (BPER) [ BPER = \frac{#(orders eligible before uncross cutoff)}{#(attempted batch submissions)} ]

2. Batch Miss Rate (BMR) [ BMR = 1 - \frac{#(batch fills)}{#(eligible batch submissions)} ]

3. Cutoff Slack (CSL)

   • uncross_cutoff_time - exchange_ack_time distribution (ms).
   • Watch p10/p50/p90; p10 near 0 = fragile path.

4. Uncross Price Edge (UPE)

   • Signed difference between uncross price and nearest continuous fallback benchmark.

5. Fallback Aggression Premium (FAP) [ FAP = IS_{fallback_child} - IS_{batch_counterfactual} ]

6. Batch Retry Cascade Index (BRCI)

   • Number of sequential batch misses before aggressive catch-up.

If BMR↑ + CSL↓ + BRCI↑ + FAP↑, your batch-routing logic is leaking bps.

4) Data Contract (Point-in-Time Safe)

At each child decision:

• decision_time, send_time, exchange_ack_time, batch_cutoff_time, uncross_time
• eligible_flag (true/false + reason code)
• predicted_uncross_px, realized_uncross_px, continuous_ref_px
• remaining_qty, remaining_time_budget, deadline_class
• fallback_path (venue, order type, aggressiveness)
• market_state (spread, imbalance, trade intensity, volatility)

Critical: Separate ineligible from eligible-but-unfilled. They imply different fixes.

5) Modeling Approach (Deployable)

Layer A — Eligibility + Fill Branch Model

Use a two-step branch model:

1. Eligibility model [ P(E=1|x) ] (timing slack, message delay, venue status, queue/cutoff state)

2. Conditional batch-fill model [ P(F=1|E=1,x) ]

Then estimate expected batch execution cost:

[ \mathbb{E}[C_{batch}|x] = P(E=1|x) \cdot P(F=1|E=1,x) \cdot C_{fill} + (1-P(EF))\cdot C_{fallback} ]

Layer B — Parent Completion Convexity Model

Condition expected parent shortfall on:

• cumulative misses (BRCI),
• residual notional,
• remaining schedule time,
• volatility regime.

This captures nonlinearity: early misses are tolerable, late misses are expensive.

6) Execution Controller (Practical State Machine)

• GREEN: Healthy CSL, low BMR. Normal hybrid participation.
• AMBER_BATCH: CSL thinning or BMR rising. Increase submission lead-time, reduce over-tight passive assumptions.
• RED_BATCH: repeated misses + high FAP. Cap batch retries, prioritize completion-safe continuous flow.
• SAFE_COMPLETE: deadline protection mode; minimize residual convexity.

Use metric thresholds + hysteresis, not trader discretion in the loop.

7) Policy Levers That Usually Help

1. Adaptive Lead-Time Buffer

   • Increase pre-cutoff send buffer when CSL quantiles compress.

2. Retry Budget

   • Hard cap on sequential batch retries per parent (max_brci).

3. Fallback Style by Regime

   • AMBER: softer fallback (passive/peg where safe).
   • RED: controlled aggression with burst caps.

4. Deadline-Aware Branching

   • As remaining_time shrinks, penalize uncertain batch branches more heavily.

5. Venue-Specific Latency Priors

   • Keep per-venue ACK/cutoff delay priors; don’t share one global timing prior.

8) Rollout Plan

1. Shadow (2–3 weeks): compute BPER/BMR/CSL/UPE/FAP/BRCI without policy change.
2. Canary by liquidity bucket: start where uncross events are frequent and observable.
3. Guardrails: cap residual notional and end-window aggression.
4. Promotion criteria: improve p95/p99 IS and completion reliability, not mean IS only.

9) Fast Checklist

[ ] Log cutoff and uncross timestamps with exchange-ack precision
[ ] Distinguish ineligibility vs eligible-no-fill branches
[ ] Add batch-miss and fallback-premium terms to slippage objective
[ ] Introduce adaptive lead-time + retry budget
[ ] Deploy GREEN/AMBER_BATCH/RED_BATCH/SAFE_COMPLETE controller
[ ] Gate rollout on tail IS + completion KPIs

10) References

• Budish, E., Cramton, P., Shim, J. (2015), The High-Frequency Trading Arms Race: Frequent Batch Auctions as a Market Design Response, QJE.
• Pagano, M., Schwartz, R. (2003), A Closing Call’s Impact on Market Quality (call auction microstructure context).
• ESMA materials on periodic auctions under MiFID II (market-structure and transparency context).

TL;DR

Periodic auctions are not just another venue—they are a different execution clock. If you model them like continuous books, you underprice miss-and-fallback convexity. Add eligibility/fill branching, cutoff-slack telemetry, and retry-budget controls to protect tail slippage while keeping completion stable.