Two books, two owners, two schedules
Client margin is a commercial decision. It lives in a policy, it is tiered, it has per-client overrides negotiated by people who are paid to win business, and it changes when somebody signs something.
Venue margin is not your decision at all. Each venue publishes its own schedule, changes it when it chooses, and does not ask. The two books therefore drift apart continuously and by construction — and the drift is invisible unless something is comparing them line by line.
How a spread inverts without anyone doing anything wrong
The common sequence has no villain in it. A venue raises initial margin on a symbol, usually because volatility rose — exactly when you least want a surprise. Every client tier priced beneath the new requirement is now loss-making on that symbol. Nothing in an order-management system flags this, because from its point of view nothing happened: no order, no fill, no position change.
The loss shows up later, in funding and financing P&L, attributed to a month rather than to the schedule change that caused it. By then the conversation is about margin compression in general, not about the specific tier that went underwater on a specific Tuesday.
| Client tier | You charge (IM) | Venue charges you | Spread | State |
|---|---|---|---|---|
| Tier 1 | 12.00% | 8.50% | +3.50 | Comfortable |
| Tier 2 | 10.00% | 9.25% | +0.75 | Thin |
| Tier 3 | 15.00% | 18.00% | −3.00 | Inverted |
Four reasons this is harder than it sounds
- Initial and maintenance margin invert independently. A book can be comfortable on IM and underwater on MM, which is the one that matters during a liquidation.
- Margin modes are not comparable. Isolated, cross, portfolio and unified answer different questions, and a client on cross margin cannot be validated against a venue requirement computed in isolation.
- Portfolio margin moves non-linearly. Under a portfolio or unified regime the venue requirement depends on the whole book, so a spread that is comfortable at today's prices can be gone after a 15% move — and it moves against you precisely when everything else does.
- Collateral changes the answer. You accept collateral at one haircut and post it at another, so the true economics include a second spread that the margin schedule itself does not show.
The monitoring that actually works
Reconcile per client, per product and per level, not in aggregate. An aggregate margin spread is almost always positive and tells you nothing; the money is lost on individual rows. Rank by proximity to inversion rather than by size, because the row about to go negative is more urgent than the one already deeply negative and known about.
Then make the venue side event-driven. The moment a venue republishes a schedule, revalue every affected client row against it. The whole failure mode of this problem is latency between "the venue changed something" and "we found out" — everything else is arithmetic.