What Toxic Flow Actually Costs a B-Book Broker
18 August 2026 · 5 min read
Every B-book broker knows toxic flow exists. Fewer can say what it costs them in a given month. The reason is simple: toxic flow does not announce itself. It looks like ordinary retail volume until you aggregate it, and by then the damage is already sitting in your hedging costs and LP spreads.
On MT5 and cTrader servers we review, the pattern is consistent. A small cluster of accounts, often under five percent of the active base, is responsible for a disproportionate share of negative markouts. The broker internalises the flow, hedges late, and pays the spread twice.
Where the cost shows up
- Hedge slippage: positions hedged after the move has started, at worse prices than the client fill
- LP spread widening: LPs detect the flow profile and widen quotes or reduce fill ratios for the whole book
- Rejection costs: legitimate client flow caught in tighter LP risk controls meant for the toxic cluster
- Opportunity cost: exposure limits consumed by flow you did not want to carry
How to measure it
The measurement toolkit is not exotic. What matters is running it continuously rather than after a bad month.
- Markout analysis: where price trades 1, 5 and 30 seconds after each client fill, per account
- Hold-time distribution: clustering of sub-minute holds around price updates or news
- Timing correlation: entries from different accounts landing within the same sub-second window
- Win rate by account tenure: new accounts that win immediately and consistently are a different risk than experienced traders
What to do about it
The answer is rarely to A-book everything. Most brokers run hybrid books for good commercial reasons. The practical steps are routing rules that move flagged clusters to the A-book automatically, leverage and group settings that reduce the economics of the exploit, and a documented review process so decisions are consistent between dealers and shifts.
If you cannot put a monthly number on toxic flow today, that is the first gap to close. It is measurable, and once measured it is manageable.
