Post-Trade Fee-Code Drift & Rebate-Reversal Slippage Playbook

Pricing Fee Realization Uncertainty as a First-Class Execution Risk

Why this note: Many routers optimize on expected maker/taker economics, but final billing often resolves later with different fee codes (tier flips, conditional flags, venue-specific reclassification, odd-lot handling, midpoint/hidden nuances). That gap quietly leaks bps even when raw spread/impact models look healthy.

1) Failure Mode in One Sentence

When expected fee treatment at decision time differs from final post-trade billing, the router overestimates edge and underprices true slippage.

2) Extend the Execution Objective with Fee-Realization Risk

For action (a) in context (x):

[ J(a|x)=\mathbb{E}[IS_{gross}|x,a] + \mathbb{E}[Fee_{realized}|x,a] + \lambda,\mathrm{CVaR}{q}(IS{net}|x,a) + \rho,\mathrm{FeeDriftRisk}(x,a) ]

Where:

• (IS_{gross}): spread + impact + delay/opportunity before fees,
• (Fee_{realized}): final billed fee/rebate (signed),
• (IS_{net}=IS_{gross}+Fee_{realized}),
• (\mathrm{FeeDriftRisk}): expected incremental loss from decision-time fee assumptions being wrong.

Without (\mathrm{FeeDriftRisk}), maker-favoring routes can look optimal in backtests but underperform in realized net cost.

3) Minimal Drift Model

Let:

• (f^{exp}_t): expected fee/rebate (bps) used at decision time,
• (f^{real}_t): final realized fee/rebate (bps),
• (\Delta f_t=f^{real}_t-f^{exp}_t): fee drift,
• (I_t): notional or shares (normalizer).

Define fee-drift cost:

[ C^{drift}_t = I_t\cdot \Delta f_t ]

And expected branch risk:

[ \mathrm{FeeDriftRisk}t = \sum{b\in\mathcal{B}} P(B_t=b\mid x_t,a_t)\cdot \mathbb{E}[C^{drift}_t\mid B_t=b,x_t,a_t] ]

Typical branches (\mathcal{B}):

1. Aligned: expected and realized fee code match.
2. Maker→Lower Rebate: passive fill but worse rebate tier/flag than expected.
3. Maker→Taker Reclass: intended passive behavior billed as taker-equivalent outcome.
4. Tier Boundary Slip: aggregate volume path misses expected tier.
5. Special-Case Override: venue-specific exceptions (odd-lot/hidden/midpoint/auction/conditional flags).

4) Branch Taxonomy (What to Label Explicitly)

For each fill cluster:

1. Expected Maker, Realized Maker (Aligned)
2. Expected Maker, Realized Worse-Maker
3. Expected Maker, Realized Taker/Removal
4. Expected Tier-N, Realized Tier-(N-1)
5. Expected Special Program, Program-Ineligible Realization

Net slippage tail often comes from Branch 3/4 concentration during specific microstructure regimes, not average fee error.

5) Telemetry Contract (Required)

A) Decision-Time Fee Context

• expected_fee_code, expected_fee_bps, expected_rebate_bps
• tier_snapshot_id, tier_progress_at_send
• eligibility_flags_at_send (post-only, hidden, midpoint, odd-lot, auction, conditional)
• router_reason_code (why venue/tactic selected)

B) Execution & Fill Facts

• venue_order_id, child_id, parent_id
• send_ts, ack_ts, fill_ts
• passive/aggressive classification at fill time
• quote/depth context near fill (for branch conditioning)

C) Post-Trade Billing Truth

• realized_fee_code, realized_fee_bps, realized_rebate_bps
• billing_timestamp, billing_source (drop copy / clearing / venue report)
• tier_realized, program_realized
• adjustment_reason (if correction/rebill occurred)

D) Reconciliation Fields

• fee_code_match (bool)
• fee_drift_bps, fee_drift_cash
• drift_branch_label
• net_slippage_bps_after_fee

If you cannot reconstruct expected-vs-realized fee path per child order, you cannot control this risk.

6) Label Design

Create labels for model training and monitoring:

1. FeeCodeDivergenceEvent
   • expected code != realized code
2. RebateReversalEvent
   • expected rebate turns into weaker rebate or fee
3. TierSlipEvent
   • realized tier lower than decision assumption
4. AdjustmentLagEvent
   • meaningful delay between fill and final billing truth

These labels explain why “good routing by gross slippage” can still lose on net implementation shortfall.

7) Modeling Stack (Practical)

Layer A — Divergence Probability Model

Estimate:

[ P(\Delta f_t \neq 0 \mid x_t,a_t) ]

Features: instrument, venue, tactic, order type, odd-lot ratio, hidden midpoint usage, tier-distance-to-threshold, time-of-day, volatility, participation pressure.

Layer B — Branch Classifier

Estimate branch probabilities (P(B_t=b\mid x_t,a_t)).

Layer C — Branch-Conditional Drift Magnitude

Model (\Delta f_t) (mean + upper quantiles) per branch.

Layer D — Net-Cost Policy Scorer

Use net objective (gross IS + expected fee drift + tail penalty) for venue/tactic ranking.

8) KPIs That Reveal Hidden Fee Drift

1. Fee Code Divergence Rate (FCDR) [ FCDR=\frac{N_{expected\neq realized}}{N_{fills}+\epsilon} ]

2. Rebate Reversal Cost (RRC) [ RRC=\mathbb{E}[\max(0, f^{exp}-f^{real})]\times \text{notional} ]

3. Tier Slip Burden (TSB) Incremental net bps from realized-below-expected tier outcomes.

4. Net-vs-Gross Slippage Gap (NGG) [ NGG = IS_{net}-IS_{gross} ]

5. Billing Finality Lag p95 (BFL95) p95(fill→final billed truth). Large lag reduces ability to adapt intraday.

6. Fee Drift Concentration Index (FDCI) Share of drift cost explained by top-k venue×tactic buckets.

9) Control Policy (ALIGNED → SAFE_FEE_CONSERVATIVE)

• ALIGNED

  • standard fee assumptions, normal ranking.

• DRIFT_WATCH

  • increase uncertainty penalty for risky venue/tactic buckets,
  • tighten eligibility checks before routing.

• MISPRICED

  • down-rank buckets with high FCDR/RRC,
  • cap exposure to tier-boundary-sensitive paths,
  • switch to fee-robust alternatives when gross edge is marginal.

• SAFE_FEE_CONSERVATIVE

  • assume pessimistic fee realization,
  • avoid tactics with unstable post-trade billing semantics,
  • prioritize completion/reliability while fee semantics are reconciled.

Use hysteresis and dwell time to avoid rapid oscillation.

10) Rollout Blueprint

1. Backfill reconciliation: build expected-vs-realized fee mapping for recent months.
2. Shadow metrics: run FCDR/RRC/TSB/NGG dashboards without policy changes.
3. Counterfactual replay: re-rank historical decisions using drift-aware net objective.
4. Canary deployment: low-notional subset, with strict rollback thresholds.
5. Promote only if: net slippage improves and completion/SLA remains within guardrails.

11) Common Mistakes

• Optimizing only gross slippage while treating fees as deterministic constants.
• Assuming post-only intent guarantees maker economics.
• Ignoring tier-threshold path dependence (month/day progress effects).
• Treating fee adjustments/rebills as accounting noise instead of model signal.
• Failing to version fee schedules and eligibility rules used by the router.

12) Fast Implementation Checklist

[ ] Persist decision-time expected fee code and tier snapshot per child order
[ ] Join fills to final billing truth with deterministic reconciliation keys
[ ] Label divergence/reversal/tier-slip branches
[ ] Add fee-drift term to net-cost objective and venue ranking
[ ] Deploy ALIGNED→SAFE_FEE_CONSERVATIVE controller with hysteresis
[ ] Gate promotion on NGG and RRC improvement, not gross IS alone

References

• SEC Rule 605 and Rule 606 disclosures (execution quality / routing transparency context).
• Exchange fee schedules and member notices (venue-specific maker/taker and program eligibility rules).
• Cartea, Á., Jaimungal, S., Penalva, J. (2015), Algorithmic and High-Frequency Trading.
• Kissell, R. (2014), The Science of Algorithmic Trading and Portfolio Management.

TL;DR

If fee realization is uncertain, routing on gross slippage alone is a biased optimizer. Model expected-vs-realized fee drift explicitly, track reversal/tier-slip branches, and switch to fee-conservative controls when divergence rises—before hidden billing drift leaks into net execution bps.