Forex Broker Risk Management: A-Book vs B-Book 2026

An A-book broker is a pass-through model that routes client orders to the market and earns from spread markup or commission; a B-book broker internalises those orders and takes the other side, earning from client losses but carrying the market risk; a hybrid broker does both, routing flow by client profile. In 2026 almost every serious retail forex/CFD broker runs a hybrid book governed by a risk engine, because pure A-booking caps margins and pure B-booking concentrates catastrophic risk. The operator decision is therefore not 'A or B' but 'how do I classify, measure, and hedge net exposure so the B-book is profitable and the A-book is funded by the right flow.' This guide explains the three models operator-first, how a dealing desk actually measures and manages net open exposure, the client-classification logic behind smart B-booking, and why your IB and affiliate channel materially changes the risk math.

A-book, B-book, and hybrid: the three models

The A-book and B-book distinction is the choice of what a broker does with a client's order, not how the order is executed technically. An A-book (or 'agency') broker offsets every client position with an equal and opposite position against a liquidity provider, so the client's profit or loss is matched by the broker's hedge; the broker keeps a spread markup and/or a per-lot commission and carries effectively no directional market risk. A B-book (or 'principal' / market-maker) broker does not hedge the position externally — it warehouses the risk internally and becomes the client's direct counterparty, so when the client loses the broker keeps the loss, and when the client wins the broker pays it. A hybrid broker maintains both books simultaneously and decides, per order or per client, which book the flow goes into.

These models are easy to confuse with execution models (market maker, STP, ECN), but they answer different questions. Execution model describes how price and fills are produced; book model describes where the risk ends up. A broker can run STP-style execution and still B-book the risk by hedging selectively rather than per-trade. We separate those two questions in detail in the [STP vs ECN vs market maker decision framework](stp-vs-ecn-vs-market-maker-broker-operator-decision-framework-2026); this guide stays focused on the risk and book-structure side.

Read the table as a spectrum rather than three boxes. The reason hybrid dominates is arithmetic: profitable retail flow is dominated by clients who lose over time, so internalising that flow captures margin a pure A-book gives away to the LP; but a minority of clients are consistently profitable or trade in correlated, news-driven bursts, and warehousing their risk is how brokers blow up. The whole craft of the dealing desk is sorting the two and routing accordingly.

Why pure A-book or pure B-book rarely survives

A pure A-book is the safe answer that quietly starves the business. Every position is hedged, so the broker earns only markup and commission while paying away the spread that B-book brokers keep, and on the high-volume professional flow that A-booking suits, markups are thin. A pure A-book also still carries operational and credit risk — LP credit lines, slippage between client fill and hedge fill, and the cost of rejected or re-quoted hedges in fast markets — without the upside that compensates a B-book for taking risk.

A pure B-book is the high-margin answer that periodically detonates. Internalising everything works beautifully until a correlated event — a central-bank surprise, a currency de-peg, a sharp equity-index move — pushes a large share of clients the same direction at once and the broker pays out a year of margin in an afternoon. The 2015 Swiss franc de-peg is the canonical lesson: brokers heavily B-booked in CHF pairs took losses that wiped out capital. Regulators reinforced the message: ESMA's CFD product-intervention measures and the parallel rules adopted by the FCA in the UK and ASIC in Australia capped retail leverage and mandated negative-balance protection, which directly shifts tail risk onto the broker and makes naive B-booking far more dangerous than it was a decade ago.

How a dealing desk measures net open exposure

Risk management on the B-book reduces to one number per symbol, refreshed in real time: net open exposure. For each instrument, the desk aggregates all internalised client positions into a single net long or net short figure — if clients are net long 40 lots of EUR/USD and the broker has B-booked all of it, the broker is effectively short 40 lots of EUR/USD and profits if EUR/USD falls. The desk then decides how much of that net position to keep (warehouse) and how much to hedge externally with an LP. The portion hedged is, in effect, moved to the A-book.

• Net open position per symbol: real-time aggregate of all B-booked client exposure, long minus short, in lots and in base-currency notional.
• Exposure limits: hard per-symbol and portfolio caps that trigger automatic or manual hedging when breached.
• Correlation grouping: USD pairs, gold, indices and crypto CFDs are grouped so the desk sees true aggregate directional risk, not 30 isolated symbols.
• VaR / stress scenarios: what the warehoused book loses in a defined shock (e.g. a 3% gap in a major pair, a de-peg, an index limit-down).
• Hedge ratio: the share of net exposure offloaded to LPs versus warehoused, adjusted by symbol volatility and event calendar.

This is where a real risk engine earns its keep. A modern engine ingests live positions from the trading server (MT4 Manager API, MT5 Gateway/Manager API, or a bridge), computes net exposure and VaR continuously, and either auto-hedges above thresholds or alerts the desk to act. Accurate, low-latency volume data from the trade server is non-negotiable: if exposure is calculated on stale data, the hedge is wrong. The same real-time volume feed that powers the risk engine also powers IB commission accruals, which is why brokers care so much about [real-time reporting](/features/real-time-reporting) accuracy across the stack.

Client classification: the logic behind smart B-booking

Smart B-booking is the practice of deciding which clients to internalise and which to hedge based on behaviour, not on covering the whole book blindly. The desk scores each client and each strategy on profitability, holding time, trade frequency, and toxicity — 'toxic' flow being latency-arbitrage, news-scalping, or copy-trading clusters whose orders are systematically right at the broker's expense. Profitable and toxic flow is A-booked (hedged) so the broker takes only the markup; small, retail, losing-skewed flow is B-booked to capture the negative expectancy.

Classification is dynamic, not a one-time tag. A client who starts as casual retail can drift profitable or pick up a scalping EA, and the engine must re-route them. Most brokers run this as a scoring model with a manual override for the dealing desk, recalculated daily or intraday. The accuracy of classification is the single biggest driver of B-book P&L stability — get it wrong systematically and the warehoused book becomes a slow leak instead of a margin centre.

The brokers who survive the next franc-style event are not the ones who avoided the B-book. They are the ones who knew, to the lot, what they were warehousing and what it would cost them in a gap.

How IB and affiliate flow changes the risk math

Brokers cannot price IB- and affiliate-sourced flow as if it were direct flow, because partner channels deliver distinct risk and cost profiles. IBs and affiliates are paid on volume, a spread share, lot-based rebates, CPA, or revenue share, which means the broker carries an acquisition cost per client that direct flow does not. On B-booked partner clients the economics are usually strong — the broker keeps client losses and pays the partner a slice — but on A-booked profitable clients sourced by an IB, the broker can find itself paying both a hedge cost and a partner commission against a thin markup, turning a 'won' client into a loss leader.

There is also a behavioural dimension. Certain IBs and affiliate channels systematically deliver a particular flow profile: a signals-group IB may bring correlated, copy-trading-style flow that nets into concentrated directional exposure; a cashback-affiliate channel may bring high-frequency, rebate-driven scalpers whose flow is toxic to the B-book. Because of that, the risk desk needs partner-level reporting alongside client-level reporting — net exposure and profitability sliced by IB and by sub-IB tier, and tracked across the trader lifetime rather than at a single point, not just by symbol. When you can see that 'IB #214's clients are net long gold and net losing,' you can price that partner's commission and route their clients deliberately. We unpack the full revenue picture in the [forex broker business model and revenue economics guide](forex-broker-business-model-revenue-economics-2026).

This is why the commission engine and the risk engine should read from the same source of truth. If your partner platform calculates commissions on different volume data than your risk engine sees, partner P&L and book P&L will never reconcile, and you will discover the gap only when a partner disputes a payout or a B-book segment underperforms inexplicably. Running partner economics on infrastructure wired into the same trade-server feed — multi-tier overrides, per-instrument commission rules, and partner-segmented reporting — turns the IB channel from a blind spot into a managed input to the book. For the broader LP and hedging side, see the [forex liquidity providers selection guide](forex-liquidity-providers-how-to-choose-operator-guide-2026).

Building the risk function: people, tech, and limits

1. Define the risk policy first: per-symbol and portfolio exposure limits, VaR budget, hedge ratios by volatility band, and the event-calendar overrides — written down and signed off by management.
2. Wire the risk engine to the trade server with a low-latency feed (MT4/MT5 Manager API or bridge) so net exposure and VaR are live, not end-of-day.
3. Stand up the classification model: score clients on profitability, holding time, frequency, and toxicity; recalc at least daily; allow desk override.
4. Automate hedging at the limit boundary: above the threshold, auto-route the residual to LPs; below it, warehouse and monitor.
5. Add partner-level risk reporting: net exposure and P&L sliced by IB and tier, reconciled against the same volume feed that drives commissions.
6. Stress-test continuously: model franc-style gaps, de-pegs, and index limit-down on the current warehoused book; size capital to the worst plausible loss, not the average.

A broker can start lean — many launch with a rules-based engine inside their bridge or back office and a single risk manager — but the function has to exist on day one. The cost of getting exposure measurement wrong scales with volume, and the partner channel adds a dimension most new brokers underestimate. Explore how the partner side fits the broader stack on the [Track360 forex industry page](/industries/forex) and the [product overview](/product).

Frequently asked questions

A-book, B-book, and hybrid are not ideologies; they are tools for placing risk where you can afford it and capturing margin where you can manage it. The brokers who scale safely in 2026 measure net open exposure to the lot in real time, classify clients dynamically, hedge the residual deliberately, and — the part most operators underweight — treat their IB and affiliate flow as a distinct, priced input to the book rather than an undifferentiated stream. Get the risk engine and the commission engine reading the same data, and the book becomes a controllable margin centre instead of a recurring source of nasty surprises.