Settlement, Custody, and What Happens When It Fails

Settlement is simpler than clearing and has one rule that carries almost all of its safety.

The danger is principal risk: delivering your side and receiving nothing. Unlike replacement cost, which is proportional to a price move, this is proportional to the full value, and no amount of collateral makes it acceptable.

The fix is mechanical rather than contractual. Delivery versus payment links the two legs so that the securities transfer and the cash transfer either both occur or neither does. Neither party can be left having performed alone.

That is an atomicity guarantee. It is not a promise that both sides will behave, and it does not depend on either side's creditworthiness. It is a property of the settlement system, enforced by whoever controls both records.

The result is that principal risk in domestic securities settlement is essentially engineered out. What remains is the risk of the trade not settling, which is a liquidity and replacement problem, not an existential one.

The word ownership hides a chain of intermediaries, and knowing its shape explains most operational risk.

Almost all securities are immobilised or dematerialised: they exist as book entries at a central securities depository rather than as certificates anyone holds. Settlement is the depository amending its records.

But the depository's records rarely name the investor. They name participants, typically large banks. Those banks are custodians, and they hold securities on behalf of clients, who may themselves be intermediaries holding for others.

So a typical holding runs: depository, to global custodian, to local sub-custodian, to broker, to investor. The investor's claim is against their immediate intermediary, and each link is a record that must agree with the one above it.

Two consequences follow. Most settlement problems are reconciliation failures between links rather than defaults. And an investor's protection in an intermediary's insolvency depends on whether their assets were properly segregated at every level, which is the same segregation question the previous lesson raised, applied to custody rather than clearing.

A settlement fail occurs when a trade does not settle on the intended date, usually because the seller does not have the securities available to deliver.

The important framing is that this is a normal operational occurrence rather than a crisis. Causes are mundane: securities out on loan and not recalled in time, a mismatch in instructions between two custodians, an allocation that arrived late, a chain of dependent deliveries where one link failed and the failure propagated.

That last cause explains why fails cluster. If A is delivering to B who is delivering to C, one absent delivery at the front fails the whole chain. Fails are correlated, and they spike when something upstream has gone wrong.

The consequences are graded rather than dramatic. The buyer does not receive securities but also does not pay, so principal risk never arises: delivery versus payment saw to that. What they lose is the use of the position, and any onward delivery they had planned.

So a fail is a liquidity event. It is expensive, it propagates, and it is not a default.

Markets discipline fails through an escalating set of mechanisms, each more forceful than the last.

Cash penalties accrue daily on the failing party, sized to make failing more expensive than borrowing the securities. This is the ordinary case and it resolves most fails without further action.

Securities borrowing is the practical fix. A seller short of stock borrows it to make delivery. This is a large part of why securities lending markets exist at all, and it means the borrowing rate is effectively the price of avoiding a fail.

Buy-in is the backstop. After a defined period, the buyer, or the market infrastructure on their behalf, purchases the securities in the market and charges the failing seller the difference. This resolves the fail by force.

The pricing signal in the middle row is worth noting. When many participants are short the same security, borrowing it becomes expensive, and the cost of avoiding a fail rises. A security that is hard and expensive to borrow is described as special, and that rate is a directly observable measure of how crowded the short side is.

The cycle has shortened repeatedly over decades, and each move trades the left branch against the right one.

The benefits are real and quantifiable. Exposure between trade and settlement is proportional to the time it lasts, so halving the cycle roughly halves the replacement cost risk in the system, and the margin a CCP must collect falls correspondingly. That is capital released across the whole market.

The costs are operational. Every step in the earlier lesson, allocation, funding, currency conversion, recalling loaned securities, must now complete in less time. Processes that ran comfortably overnight become same-day, and anything requiring a human becomes a bottleneck.

The cross-border case is the sharpest. An investor in a different time zone may have almost no working hours between trade date and the deadline to fund, and the currency transaction they need may itself settle on a longer cycle than the securities trade it funds.

So the cycle is not simply a dial to turn down. Each reduction forces automation of whatever was still manual, and the transition is where the cost lands.

The direction of travel is one way, and the specifics are worth knowing because they create a live mismatch.

The United States, Canada and Mexico moved to T+1 in May 2024, so a trade executed on Monday settles on Tuesday. India had already moved. Most of Europe remains on T+2, and the European Union, the United Kingdom and Switzerland are scheduled to transition together on 11 October 2027.

The interval between those dates is the interesting part. A firm holding both US and European securities settles the same strategy on two different cycles, which breaks anything that assumed they matched.

The clearest example is a funding chain. Selling a European holding to buy a US one leaves a day where the US purchase must settle before the European sale has, so the cash is needed before it arrives. That gap has to be bridged with borrowing or pre-funding, and the cost is a direct consequence of the mismatch rather than of either cycle.

This is the general shape of settlement problems. They are rarely about a single market's rules and usually about two systems meeting at different speeds.

If shorter is safer, the endpoint appears to be settling every trade immediately. It is a serious proposal and it has a serious objection.

Instant settlement means no netting. Every trade settles on its own, gross, so a participant that bought and sold the same security a hundred times must fund and deliver all two hundred legs rather than the single net position at the end of the day.

The first lesson's example showed six trades collapsing to no securities movement at all. Removing netting reverses that, and the funding requirement rises by a large multiple, because cash and securities must be in the right place for every leg at the moment it occurs.

So instant settlement trades counterparty risk for liquidity risk, and the exchange is not obviously favourable. Counterparty risk over one day, backed by margin and a default waterfall, is well understood and cheap to cover. A permanent requirement to pre-fund gross positions is a continuous cost borne by everyone.

That is why the industry has moved in steps rather than jumping to zero. Each step is a judgement about where the two curves cross, and the answer depends on how automated the plumbing has become.

The first row is the only unambiguous win in the table, which is why it is universal. Every other row is a genuine trade with a defensible position on both sides.

The pattern is that settlement design moves risk between categories rather than removing it. Counterparty risk becomes liquidity risk, liquidity risk becomes operational risk, and operational risk becomes cost.

That framing is the useful takeaway, and it is the same one the risk cursus reached from a different direction. The question is never whether a design eliminates risk, but which risk it converts into which other, and whether the recipient is better placed to bear it.