Horizon has two parts
Execution is the part everybody models — how long it takes to work an order without moving the market against yourself. Mobilisation is everything that has to happen before a single order can be placed: the custodian's withdrawal SLA, the transfer to a venue account, the multi-signature ceremony, the operations person who has to be awake.
A 24-hour horizon against a custodian with a T+1 withdrawal SLA has nothing left for execution. On screen it reads as a comfortable 24 hours.
The same assets, in three places
Consider identical holdings, identical prices, identical haircuts. Only the location differs:
| Where it sits | Mobilisation | Reachable in a 24h horizon? |
|---|---|---|
| Venue account | ~1 hour | Yes — 23 hours left to execute |
| Qualified custodian | ~24 hours | No — arrives as the horizon expires |
| Unallocated / house wallet | ~48 hours | No |
The second and third rows are not haircut problems. The asset is good, the price is good, the haircut is right. It simply cannot be turned into cash inside the window the risk model assumed, which makes it unusable for the purpose the model was using it for — and every credit cap resting on it is overstated.
Segregation is a separate question, and a bigger one
Where collateral sits determines whether you can reach it. How it is held determines whether it is yours at all. Omnibus holdings are recoverable but attributable only through your own books. Commingled holdings, in an insolvency, are frequently an unsecured claim against the estate rather than property you own.
That is not a haircut adjustment. A haircut says an asset is worth less than face value; this says the asset may not be an asset. The two belong in different columns and are routinely conflated.
Modelling it without breaking everything
- Record a mobilisation time per location, not per asset. The asset does not decide how long the withdrawal takes; the arrangement does.
- Treat a placement that arrives exactly as the horizon expires as stranded. Arriving at the deadline is not a liquidation.
- An unresolvable location is worse than an unrecorded one — recorded-but-wrong should fail loudly.
- Count collateral with no recorded location as reachable, and report that as a named assumption. Excluding it zeroes every cap on the first day, which is defensible and useless. Naming the unknown creates pressure to record it.
Why it is worth the trouble
Because the numbers move materially. In one worked case, the same client with the same assets showed roughly double the usable collateral when those assets sat on a one-hour venue account rather than behind a 24-hour custodian — and the binding constraint on their limit changed from collateral to liquidity. Same book, same market, different answer to "how much can this client trade".
The uncomfortable version of this is that a horizon that ignores mobilisation does not produce a slightly optimistic number. It produces a number that is wrong in a way that only reveals itself during a liquidation, which is the one moment nobody has time to recalculate.