Skip to content
FXOpen Ledger
Directory

Abook Bbook Stp Bridge Routing

Shadow A-Book Paper-Hedging Analytics for Dealing-Desk Calibration

Changing C-Book routing thresholds on live dealer instinct wastes millions; Shadow A-Book simulation engines replay every historical trade against recorded L2 wholesale depth to prove counterfactual net PnL.

By Alistair Kerr, CFAReview not recorded
Shadow A-Book Paper-Hedging Analytics for Dealing-Desk Calibration — FXOpen Ledger 2026 quantitative & supervisory data card
FXOpen Ledger (A-Book STP, B-Book Internalization, C-Book Hedging & LP Toxicity): Counterfactual L2 tick replay, Last-Look cascade simulation, and walk-forward C-Book tuning.Source: FXOpen Ledger Institutional Research DeskLicense: CC-BY-4.0

Every Monday morning, FX risk committees debate the same counterfactual question: "Did our C-Book hedge rules externalize too much benign flow last month—or keep too many sharp accounts inside the B-Book?" Answering that question by comparing realized B-Book PnL against naive mid-price charts overstates A-Book profitability by 30% to 55% because mid-price charts ignore wholesale half-spreads, L2 VWAP sweep impact, Prime-of-Prime ticket fees, and asymmetric Last-Look rejections! Institutional dealing desks resolve routing calibration scientifically using a Shadow A-Book Paper-Hedging Simulator—a parallel telemetry engine that records real-time L2 wholesale liquidity ticks alongside every live client trade to compute exact Counterfactual STP Net PnL down to the basis point.

Connected to our A-Book, B-Book & C-Book FX Dealer Architecture guide and indexed inside the A-Book, B-Book & Bridge Routing hub, Shadow A-Book backtesting prevents over-hedging false positives and under-hedging stealth alpha.

Why Naive "Client PnL Reversal" Fails as an A-Book Counterfactual#

Novice dealing desks assume that if Client X made +$42,000 in the B-Book over 90 days, routing Client X to A-Book would have saved the broker $42,000 plus commission! In reality, three real-world execution frictions separate theoretical mid-price hedging from live STP execution:

  1. Spread + Clearing Drag Can Exceed Modest Client Profit: Suppose Client X traded $2,500 Million of high-turnover scalping volume to earn that +$42,000 (+$16.80/M), while the broker already earned $85/M gross spread—leaving the B-Book net profitable by +$68.20/M (+$170,500)! Had the desk routed Client X to A-Book paying $29/M wholesale STP cost, broker net yield would have dropped from +$68.20/M to +$56.00/M—costing the desk -$30,500 in lost margin!
  2. Simulated Last-Look Reject Replay: During NFP spikes, post-execution auto-hedgers suffer 35%–50% Last-Look rejects; Shadow A-Book engines apply empirical LP reject curves (drift > tolerance -> re-price at t + 20ms) so simulated hedges include realistic cascade slippage.
  3. IB Rebate Interaction: Shadow engines deduct account-specific IB rebate schedules automatically on every replayed ticket.
Parallel Shadow A-Book counterfactual simulation pipeline replaying live client trades against recorded Equinix LD4 L2 ticks
Parallel Shadow A-Book counterfactual simulation pipeline replaying live client trades against recorded Equinix LD4 L2 ticksVerified Exhibit

The Counterfactual Regret Equation & Confusion Matrix#

For every historical account $i$ over a rolling 30-day evaluation window, the Shadow A-Book engine compares Realized B-Book Desk PnL against Simulated Shadow-STP Desk PnL to classify every routing decision into a 2x2 Routing Confusion Matrix (True Sharp, False Positive Lucky, True Benign, False Negative Stealth Toxic):

Formula or model from the articleNot independently verified
BBook_Desk_Net_i = Gross_Spread_Swap_Rev_i - Client_Gross_Trading_PnL_i - IB_Rebate_i
  • ShadowSTPNeti = GrossSpreadSwapRevi - (SimulatedLPVWAPSlipi + PoPClearingi + RejectCascadeEsti) - IBRebatei
  • RoutingRegretUSDi = ShadowSTPNeti - BBookDeskNet_i
  • = ClientGrossTradingPnLi - (SimulatedLPVWAPSlipi + PoPClearingi + RejectCascadeEst_i)
  • Optimal Rule: Externalize Account i to A-Book IFF E[ClientGrossTradingPnLi / Volumei] > ShadowAllInSTPCostper_M!
  • Worked 30-Day Shadow Calibration Audit across Two "Profitable" Client Accounts ($100M Monthly Volume Each):
  • Account 1 ("Lucky High-Volume Day Trader"):
  • Client Gross Trading Profit = +$1,900 (+$19/M); Shadow All-In STP Cost (LP spread + PoP + Cascade) = $31/M ($3,100).
  • Routing_Regret = $1,900 - $3,100 = -$1,200! Keeping Account 1 in B-Book beats A-Book STP by +$1,200 (+$12/M)!
  • Account 2 ("Stealth API Scalper Currently Down -$400 Due to One Fat-Finger Trade"):
  • Headline Net PnL = -$400 (Looks harmless to naive PnL filter!), YET +5s Markout Alpha = +$54/M ($5,400 expected)!
  • Shadow Walk-Forward Replay predicts +$5,400 - $31/M ($3,100) = +$2,300/100M saved by immediate A-Book promotion!

Routing Confusion Matrix & Calibration Benchmarks Across Cohorts#

Quantitative risk teams at institutional multi-asset venues—benchmarked across firms in our regulated dealers directory such as City Index, AMB Prime, 4XC, and Blue Suisse—evaluate C-Book rules quarterly using out-of-sample walk-forward validation rather than in-sample curve fitting:

Quantitative Distribution · Account Empirical Reality
Account Empirical RealityNaive PnL > 0 Rule Mistake
01True Sharp (Alpha > C_stp)Genuine +5s mar…02False Positive (0 < Alpha <…Modest winner (…03False Negative (PnL < 0, Sh…Sharp entry mas…
Confusion Matrix QuadrantAccount Empirical RealityNaive PnL > 0 Rule MistakeShadow A-Book Engine ActionDesk Margin Impact ($ / $1M)
True Sharp (Alpha > C_stp)Genuine +5s markout > +$31/MSlow (Waits for big drawdown)Promotes in ~18 trades via MarkoutSaves +$23 to +$65 / $1M
False Positive (0 < Alpha < C_stp)Modest winner (+$15/M < $31/M STP)Over-hedges 100% to A-Book!Retains in VaR-capped C/B-BookRecaptures +$12 to +$18 / $1M
False Negative (PnL < 0, Sharp Entry)Sharp entry masked by 1 wide lossLeaves 100% unhedged in B-Book!Flags +0.45 bps 5s markoutPrevents future -$45/M shock
True Benign (Alpha < 0)Uninformed retail noise (-$42/M)Correctly internalizesKeeps 100% B-Book + 20s nettingPreserves full $85-$120 / $1M

Walk-Forward Desk Yield & Sharpe Uplift From Shadow L2 Replay#

Out-of-sample 90-day evaluations across $4.2 billion of multi-asset flow quantify the exact revenue and Sharpe gains of full Level-2 counterfactual calibration:

Out-of-sample dealing-desk revenue yield and Sharpe ratio comparing naive PnL routing against Shadow A-Book calibrated C-Book
Out-of-sample dealing-desk revenue yield and Sharpe ratio comparing naive PnL routing against Shadow A-Book calibrated C-BookVerified Exhibit

Deploying Shadow A-Book on Read-Only Drop-Copy Streams#

Because a Shadow A-Book simulator subscribes strictly to read-only FIX Drop-Copy (35=8 / 35=AE) and passive market-data streams stored in ClickHouse, TimescaleDB, or kdb+, it introduces 0.0 microseconds of latency to the live production execution path and zero compliance friction under FCA, ASIC, and CySEC conduct rules. Furthermore, running parallel shadow profiles across candidate upstream Prime-of-Prime feeds lets liquidity managers test a prospective new LP's real L2 depth and hold-time behavior for 30 days on paper before committing a single dollar of live margin collateral.

Frequently asked questions

What is a Shadow A-Book simulator on an institutional FX dealing desk?

A Shadow A-Book is a read-only analytical engine that takes every live B-Book and C-Book client order and simulates what exact fill price, LP fee, and Last-Look rejection rate would have occurred had that trade been hedged live on wholesale ECN feeds.

Why shouldn’t a dealing desk route every profitable client account to A-Book?

If a high-turnover trader earns +$18 per million traded while paying $95/M in spread, the broker still nets +$77/M in B-Book; routing that trader to A-Book at $31/M wholesale clearing cost actually reduces broker net profit to +$64/M!

Does running Shadow A-Book analytics affect client order execution or latency?

No. Shadow A-Book operates asynchronously out-of-band using passive FIX Drop-Copy execution reports and archived market-data ticks in a time-series database (such as kdb+ or ClickHouse), adding zero latency to live orders.

How do liquidity managers use Shadow A-Book to evaluate new Liquidity Providers?

Before funding a new LP margin account, desks pipe the prospective LP live FIX market-data stream into the Shadow simulator for 30 days to measure realistic L2 VWAP depth across actual historical client clip sizes.

Publication record

Published
14 November 2026
Updated
14 November 2026
Author
Alistair Kerr, CFA
Review
Review not recorded
Automated fact check
Passed in the stored pipeline record
Source URLs
3 recorded

AI assistance recorded: anthropic/claude-opus-4.6. This field does not establish human review or source verification.

Academic & Institutional Attribution

Cite This Research & Embed Live Data Card

Open Attribution License (CC BY 4.0)

Standard academic & institutional research paper citation format.

Alistair Kerr, CFA. (2026, November 14). Shadow A-Book Paper-Hedging Analytics for Dealing-Desk Calibration. FXOpen Ledger. https://fxopenledger.com/orderbook/shadow-abook-paper-hedging-analytics-for-dealing-desk-calibration

Regulatory Directory Cross-Reference

Referenced Institutions & Regulatory Registries

Referenced Supervisory Authorities

Explore Abook Bbook Stp Bridge Routing Hub (20) →

FXOpen Ledger Data, Benchmarks & Contributor Records

Shadow A-Book Paper-Hedging for FX Desk Calibration | FXOpen Ledger