Where the Risk Moved

Before assessing the costs, the achievement should be stated accurately, because it is substantial and it is often understated by people making the argument that follows.

In a bilateral market, exposures form a network. Each participant faces many others, nobody can see the whole graph, and a default propagates along edges that its victims did not know existed. During a crisis this produces a specific failure: participants stop trading with anyone, because they cannot tell who is exposed to whom.

Central clearing collapses that network into a hub. Every participant faces one node whose finances are published, whose margin model is documented, and whose default resources are sized to a stated standard.

The gains are concrete. Exposures net multilaterally rather than bilaterally, which reduces them enormously. Positions are marked and margined daily, so exposure cannot silently accumulate. A default has a rehearsed procedure rather than a litigation. And the uncertainty that freezes bilateral markets is removed, because there is nothing to be uncertain about.

Any critique has to hold all of that fixed. The question is not whether clearing helps, but what it does with the risk it takes on.

The same property that makes a hub safe makes it critical. Every participant now depends on one institution, so the CCP cannot be allowed to fail.

This is not a hypothetical concern about an unlikely event. It is a structural observation: the resilience of a large fraction of the financial system has been made contingent on a small number of entities, each of which is now more systemically important than most of its members.

The waterfall is deep and prefunded and would absorb a great deal. But it is finite, and beyond it lie rights of assessment, recovery tools that allocate losses to surviving members, and finally resolution.

The uncomfortable feature is what those later stages mean. Loss allocation beyond the prefunded resources works by calling cash from surviving members, precisely when they are least able to supply it, since whatever caused the default is affecting them too.

So the tail of a CCP's own distribution has the same procyclical shape the risk cursus identified everywhere else: the demand for resources peaks exactly when the capacity to meet it is lowest.

The two structures fail differently, and the comparison is the substance of the debate.

A bilateral network fails by contagion: unpredictable propagation along paths nobody had mapped, with the damage determined by where the shock landed.

A cleared market fails by correlation: everyone receives a margin call from the same institution, computed by the same model, at the same moment. Nothing propagates unpredictably, because there is nothing to propagate along. Instead, every participant is squeezed simultaneously.

The right-hand branch is the one to sit with. The margin lesson established that requirements rise with volatility. In a cleared market that increase is applied to every member at once, by one model, so the CCP becomes a synchronising mechanism for liquidity demand.

That is a genuinely better failure mode: predictable, quantifiable, and rehearsed, rather than opaque. It is not an absence of one.

One consequence deserves separating out, because it converts a technical choice into a public policy question.

A CCP's margin model determines how much collateral the whole market must post. When volatility rises and the model demands more, that demand lands on every member simultaneously. The model is therefore a determinant of system-wide liquidity demand, not merely of one institution's protection.

This creates a design tension with no clean resolution. A model that responds quickly to volatility protects the CCP well and generates large, sudden, correlated calls. A model that responds slowly produces gentler calls and leaves the CCP under-collateralised exactly when risk is rising.

The usual mitigations are dampening devices: floors so margin never falls too far in calm periods, and buffers that absorb part of an increase rather than passing it through. Both work by deliberately mis-measuring current risk in order to smooth the demand for cash.

That is a strange thing for a risk model to do, and it is defensible only once you accept that the model's output is itself a driver of the risk it is measuring.

A second consequence is quieter and probably larger in aggregate: clearing consumes collateral permanently.

Initial margin is not a payment, so it is often treated as costless. It is not. Assets posted as margin are encumbered: they cannot be lent, sold or used elsewhere while they sit there. Multiply that across every cleared position in every market and the total is a substantial quantity of high-quality assets held immobile.

That creates demand for collateral, and demand creates an industry to satisfy it. Collateral transformation is the practice of borrowing eligible collateral against ineligible assets, so a participant holding securities a CCP will not accept can obtain ones it will.

That service is useful and it reintroduces exactly what clearing was meant to remove. It is a bilateral, leveraged, secured financing transaction, typically short term, and it must be rolled. If it cannot be rolled in a stressed market, the participant cannot meet its margin call.

So a chain that begins with removing bilateral counterparty risk ends with a bilateral funding dependency sitting behind the ability to post margin. The risk did not vanish; it moved one layer down and became harder to see.

The hub has few spokes, and that is a separate concentration from the CCP itself.

Clearing membership requires capital, operational infrastructure and default fund contributions, so the number of direct members is small, and a large share of client clearing runs through a handful of them. The business is capital-intensive and low-margin, which pushes toward consolidation rather than away from it.

Two exposures follow. A member's failure requires its clients to be ported elsewhere, and if the remaining members are few and already at capacity, porting may not be possible at the speed required. And the default auction depends on surviving members having both the appetite and the balance sheet to absorb a large portfolio, which is a thinner assumption when there are fewer of them.

The auction point is the sharper one. The resource that ultimately protects a CCP is not its collateral but the willingness of surviving members to take on the defaulter's positions. That willingness is not prefunded and cannot be, and it is least available in exactly the conditions where it is needed.

Juniorisation makes participation rational. It cannot make capacity exist.

The top half is why clearing was mandated and the bottom half is why it is still argued about. Both halves are true simultaneously, and a view that only holds one of them is not a view about the system.

The defensible summary: clearing traded a large number of small, opaque, unpredictable exposures for a small number of large, visible, correlated ones. That is a good trade if you believe opacity and unpredictability are the harder problems, which the crisis it followed suggests.

It is a good trade rather than a solution, and the distinction is what makes the remaining exposures worth naming rather than assuming away.

This path covered the half of trading that happens after the price is agreed, and its lesson generalises past clearing.

Every piece of market infrastructure is a risk transformation. Delivery versus payment converts principal risk into settlement failure. Netting converts operational load into time, and time into counterparty exposure. A shorter cycle converts counterparty exposure into operational pressure. Central clearing converts network contagion into concentrated correlation. Margin converts credit risk into liquidity risk.

Not one of them destroys risk. Each moves it into a different category, usually one that is better understood, more measurable, or borne by someone better placed to carry it.

So the question to ask of any proposed infrastructure change is not whether it makes things safer. It is: which risk does this convert into which other, and who ends up holding it? That question has an answer for every reform in this cursus, and it is a more useful frame than a verdict.

It also connects back to where the trading path started. A market has no privileged observer and no mechanism that removes uncertainty. It only has arrangements for deciding who bears which part of it, and infrastructure is the most consequential of those arrangements precisely because it is invisible while it works.