Every latency arbitrage trader eventually hits the same wall. The strategy works on a demo account and works for the first two weeks live, and then things start to go wrong. Fills get slower. Slippage only ever goes against you. A support ticket mentions “abnormal trading activity”. The strategy didn’t change. What changed is which book your trades were sitting in.
“A-book” and “B-book” are broker jargon for where your order goes after you click buy. Most articles on the topic are written for people starting a brokerage. This guide is written from the other side of the screen: what the two models mean for an arbitrage trader, why hybrid brokers move profitable flow between books, which signs tell you where you are, and how to pick an execution venue that won’t turn on you after the first good month.
A-book vs B-book in one minute
A-book means the broker passes your trade to an outside liquidity provider (a bank, a non-bank market maker or an aggregator) and hedges it one-to-one. The broker earns from the spread markup and commission and has no direct interest in whether you win or lose. These are the accounts usually sold as STP, ECN or “no dealing desk”.
B-book means the broker keeps your trade on its own books and takes the other side itself. When you lose, the broker books the profit. When you win, the broker pays you out of its own pocket. For a large, mostly losing retail client base this is a perfectly rational business, and most retail brokers run at least part of their flow this way.
Hybrid is what nearly every mid-size broker actually runs. A risk engine scores each account and routes it: unprofitable or random flow stays in the B-book, while flow that looks consistently profitable (the industry calls it “toxic”) is pushed to the A-book or restricted. That routing decision is the single most important fact about your arbitrage account, and it is almost never shown to you.
Why latency arbitrage and the B-book don’t mix
Latency arbitrage makes money from one thing: a broker’s quote that lags behind a faster reference feed. If EURUSD has already moved up 1.2 pips on the fast feed and your broker still shows the old price, you buy at the stale ask and close a moment later once the quote catches up.
Now ask who sold you that stale price. In the A-book the broker forwarded your order to a liquidity provider, so the LP absorbed the loss, and the LP is also the party that will tune its pricing or flag your flow upstream. In the B-book the broker itself sold you the stale price. Each winning arbitrage trade comes straight out of the dealing desk’s P&L, and B-book profit only works on average if clients lose on average. An account that wins 80% of trades with an average hold time of seconds is the exact profile the risk engine is built to catch.
This is why the same strategy can give completely different results at two brokers with identical spreads. The difference isn’t the quote. It’s who pays when you are right.
The metric that gives you away: markout
Brokers rarely catch arbitrage by looking at your strategy. They look at markout: how far the market moves in your favour in the first few hundred milliseconds or seconds after your fill. A normal retail trade has a markout that hovers around zero, because the price after entry is close to random. A latency arbitrage trade has a markout that jumps positive almost immediately, because you only enter when you already know where the price is going. The curve below shows the idea with illustrative numbers.
What happens to an arbitrage account in each model
Once the risk engine labels your flow, the broker has a menu of responses. Which ones it reaches for depends mostly on the execution model behind your account:
| Response | Pure B-book | Hybrid | True A-book / ECN |
|---|---|---|---|
| Added execution delay (e.g. a “virtual dealer” style plug-in) | Common, often the first step | Common before the account is moved | Rare; delay comes from the LP’s last look |
| Asymmetric slippage (negative fills, positive ones capped) | Common | Possible | Slippage should run both ways |
| Spread widening on your account group | Common | Common | Only market-wide |
| Moved to the A-book | Not an option | The usual outcome; edge then shrinks to what the LP allows | Already there |
| Trade cancellation / profit clawback citing terms of business | The real risk | Possible | Uncommon, but LPs can reject or re-price |
| Account closure | Common end state | Possible | Rare if the terms allow the style |
The pattern is the point. In a pure B-book, a winning arbitrage account is a direct cost, so the broker’s job is to make it stop winning: delays, slippage, and eventually cancelled trades or a closed account. In a true A-book, the broker has no reason to fight you, but your stale-quote edge is smaller because LP pricing is fresher and LPs have their own last-look checks. Most of the practical game happens at hybrid brokers, where an account can be profitable in the B-book for days before the engine flags it. If execution has gotten mysteriously worse, our guide to how virtual dealer plug-ins delay and re-quote trades explains the mechanics behind the first row of the table.
How to tell whether your broker is A-book or B-book
Brokers don’t publish a routing table, and the “ECN” label on an account type proves nothing. A hybrid can sell an “ECN” account and still internalise most of it. You can, however, read the evidence. None of these signals proves anything on its own, but three or four together tell a clear story:
- Read the order execution policy, not the marketing page. Regulated brokers must publish one. Phrases such as “we may act as principal” or “we may be the counterparty to your trades” mean B-book is at least possible. A broker that only ever acts as “agent” or on a “matched principal” basis is the A-book profile.
- Look at the revenue model. Zero commission plus very tight fixed spreads on a small minimum deposit is the classic B-book economics. Raw spreads plus a per-lot commission points toward A-book, though hybrids offer these too.
- Measure slippage symmetry. Export your fills and compare the requested price with the fill price. Real market execution produces positive and negative slippage in roughly equal measure. If 90% of slippage goes against you, someone is choosing it.
- Check the regulator’s disclosures. Several regulators require brokers to publish execution-quality reports or the share of client accounts that lose money. A very high “accounts losing money” figure paired with aggressive bonuses is a hint about the dealing model.
- Watch execution over time. A stable fill time that suddenly gets longer after a profitable week is the classic sign of an account being flagged inside a hybrid model.
- Read the terms on “latency arbitrage”, “abusive trading” and “price feed errors”. Clauses that let the broker cancel trades executed on “off-market” prices are written for exactly this strategy. Know they exist before you deposit.
- Ask support directly, in writing. “Do you act as principal on my account type? Who are your liquidity providers?” A clear, specific answer is a good sign. A vague one is also an answer.
The hybrid lifecycle: how an arbitrage account gets reclassified
Seeing the whole cycle helps you plan around it instead of being surprised by it. A typical hybrid treats a new account like this:
Three practical lessons follow. First, early results at a hybrid are not representative. Judge a venue on the fills you get after the profiling stage, not before. Second, withdraw profits regularly, so a retrospective clawback can only touch a small balance. Third, the shape of your flow matters as much as its profitability. Hold time, trade frequency and how your entries line up with the fast feed all feed the engine, and that is exactly where the choice of execution model comes in.
Choosing an execution model for arbitrage
Classic one-leg latency arbitrage (buying a lagging quote and closing when it catches up) is the strategy that suffers most in a B-book, because every trade is a short-hold, sharp-markout win against the dealer. It works best at venues that are either true A-book or openly tolerate the style, with low commissions and fast execution.
Two-leg (hedge) arbitrage opens offsetting positions at two brokers, so each account shows a mix of winning and losing trades. The flow on any single account looks much less “toxic”, and that is the main reason traders use it at brokers that would quickly flag a one-leg account. The trade-offs are double the costs and more moving parts. We compare both approaches in detail in one-leg vs two-leg latency arbitrage, and the two-broker setup itself is covered on our hedge arbitrage strategy page.
Whatever the model, a short checklist covers most of the risk before you fund an account:
- The execution policy says agent / A-book for your account type, or the broker states plainly that it permits arbitrage and HFT.
- Terms of business contain no clause that lets the broker cancel trades on “off-market prices” at its sole discretion, or you have accepted that risk with eyes open.
- A real regulator, segregated client funds, and withdrawals that work. Test one small withdrawal early.
- Commission and spread costs are low enough that your average edge per trade survives them.
- Server location close to the broker’s trade server, so your own latency isn’t the bottleneck.
To shortcut the research, start from our curated list of forex brokers that allow arbitrage, which is updated as broker policies change.
The bottom line
“A-book vs B-book” sounds like back-office trivia, but for an arbitrage trader it decides whether your edge is something the broker can live with or something it has to remove. B-book brokers pay your profits out of their own pocket, so they will eventually remove it. A-book brokers pass the cost to the market, so your edge is smaller but sustainable. Hybrids, which means most brokers, will test you in the B-book first and then decide. Learn to read the signals, choose venues whose model fits your strategy, and you spend your time trading instead of arguing with a dealing desk.
FAQ
What is the difference between an A-book and a B-book broker?
An A-book broker passes your trades to outside liquidity providers and earns from spreads and commissions, so it doesn’t profit from your losses. A B-book broker keeps your trades in-house and takes the opposite side, so your profit is its loss. Most brokers are hybrids that use both and route accounts based on how profitable they look.
Is a B-book broker a scam?
No. Market making is a legal, regulated business model, and B-book execution can be perfectly fair for ordinary retail trading. The conflict of interest only becomes acute for consistently profitable, short-hold strategies such as latency arbitrage, where every winning trade costs the broker money.
How can I tell if my broker is A-book or B-book?
Read the order execution policy for “principal” or “counterparty” wording, compare the revenue model (commission-based vs zero-commission fixed spreads), measure whether your slippage is symmetric, and watch whether execution worsens after a profitable stretch. No single sign is proof, but several together are a reliable indication.
Do ECN brokers allow latency arbitrage?
Some do, many don’t. “ECN” is a marketing label, not a guarantee of A-book routing, and even genuine A-book brokers may prohibit latency arbitrage in their terms because their liquidity providers object. Always check the terms of business and the broker’s stated policy on arbitrage and HFT.
Why did my arbitrage results get worse after a few weeks?
At a hybrid broker that usually means your account finished its profiling stage and was flagged as toxic flow. Typical responses are added execution delay, asymmetric slippage, a wider spread group or a move to the A-book, where stale-quote opportunities are rarer.
Is hedge (two-leg) arbitrage safer at B-book brokers?
It is less conspicuous, because each account shows both winning and losing trades rather than a near-perfect win rate. It is not risk-free: costs double, and brokers can still look at your combined pattern. It is a legitimate way to reduce how toxic each individual account looks.
About The Author
