The Central Counterparty and Its Default Waterfall

Novation gave every participant a single counterparty. That solved their problem by creating one central problem, and the CCP's design is entirely about handling it.

When a clearing member defaults, the CCP still owes everything the defaulter owed. Its obligations to surviving members are unchanged; only its ability to collect from one side has gone.

Its response follows a sequence. It declares the default, taking control of the member's positions. It hedges the resulting exposure, since it is now unmatched on one side and exposed to the market. It closes out the positions, by auction to other members or by trading them in the market. And it allocates whatever loss remains.

The closeout is where the money is lost. The defaulter's positions must be liquidated in whatever conditions caused the default, which are rarely calm, and the difference between their value at declaration and what they fetch is the loss to be covered.

Everything else in this lesson is about having enough resources ready, in a defined order, to cover that number.

The first defence is collateral collected in advance, and it comes in the two forms the leverage lesson distinguished.

Variation margin settles gains and losses daily, or more often. Its function is to stop exposure accumulating: because losses are paid as they occur, the CCP is never owed a large accrued amount by anyone. It is a transfer, not a buffer.

Initial margin is the buffer. It is sized to cover the plausible loss on a member's portfolio over the period it would take to close it out, which is why it depends on both volatility and expected liquidation time.

That second dependency is the one people miss. Initial margin is not a percentage of notional; it is an estimate of a closeout loss, so it rises with illiquidity as well as with volatility. A concentrated position in a thin instrument attracts far more margin than its size suggests, because the CCP is pricing how long it would take to exit.

Crucially, initial margin is calibrated per member, so the defaulter's own margin is the first resource used. The design intent is that a default costs no one but the defaulter.

The default waterfall is the ordered stack of resources consumed in a default, and the order is the design.

The first two layers are the defaulter's own money: their initial margin, then their contribution to the shared default fund. If these cover the loss, nobody else is affected and the event is invisible outside the CCP.

The third layer is the CCP's own capital, committed ahead of surviving members' funds. This is called skin in the game, and it exists for incentive reasons rather than because the amount is large. A CCP that sets margin too low to attract volume would burn its own equity before touching members' money, which aligns its risk management with its members' interests.

Only beyond that does mutualisation begin: the shared default fund contributed by non-defaulting members, and then rights of assessment, under which the CCP can demand further cash from survivors.

Read the order carefully. Losses are individualised first and mutualised only after both the defaulter's resources and the CCP's own capital are exhausted.

Making the sequence concrete shows how much has to go wrong before mutualisation starts.

def waterfall(loss, def_margin, def_fund_contrib, ccp_capital, mutual_fund):
    layers = [("defaulter initial margin", def_margin),
              ("defaulter fund contribution", def_fund_contrib),
              ("CCP skin in the game", ccp_capital),
              ("mutualised default fund", mutual_fund)]
    remaining, used = loss, []
    for name, size in layers:
        take = min(remaining, size)
        if take > 0:
            used.append((name, take))
        remaining -= take
        if remaining <= 0:
            break
    return used, max(remaining, 0)

waterfall(180, 250, 40, 60, 3000)
# ([('defaulter initial margin', 180)], 0)        contained entirely

waterfall(400, 250, 40, 60, 3000)
# ([('defaulter initial margin', 250), ('defaulter fund contribution', 40),
#   ('CCP skin in the game', 60), ('mutualised default fund', 50)], 0)

The first case is the normal one: margin was sized correctly and absorbed the loss. In the second, margin was insufficient, and only after the defaulter's fund contribution and the CCP's own capital are gone does 50 reach the mutualised fund.

That structure is why members care intensely about a CCP's margin methodology. They are the fourth layer, and their exposure is to someone else's positions being under-margined.

The waterfall answers the order of resources. Its total size is a separate question, and the standard answer is a coverage rule rather than a probability.

The widely applied benchmark requires a CCP's prefunded resources to withstand the default of the two largest members under extreme but plausible conditions, commonly called cover 2. Less systemically important CCPs may be held to cover 1.

Notice the form of that standard. It is not a confidence level on a loss distribution, and after the risk cursus the reason should be clear: the distribution is estimated, and its tail is exactly where the estimate is least reliable. A rule stated in terms of specific defaults rather than percentiles does not depend on getting the tail right.

It does depend on something else being right: the scenarios used to determine what those two defaults would cost. Those come from stress testing, with all the strengths and weaknesses that lesson covered.

So the calibration has moved rather than disappeared. It sits in the stress scenarios instead of in a quantile, which is a better place for it, because a scenario is a statement someone can argue with.

The step that determines the size of the loss is the closeout, and it is more interesting than it sounds.

A CCP holding a defaulter's portfolio cannot simply sell it into the market: doing so in size, during stress, is the impact problem from the microstructure cursus at its worst. Instead it usually auctions the portfolio to surviving clearing members.

The auction has an obvious problem. A member bidding a low price makes the closeout loss larger, and that loss falls on the mutualised fund which that same member contributes to. Bidding aggressively low to buy a portfolio cheaply is partly self-defeating.

CCPs make this explicit through juniorisation: members who bid poorly, or not at all, have their default fund contributions consumed earlier in the waterfall than those who bid well. Participation in the auction is effectively compulsory, enforced by exposure rather than by rule.

That is an elegant piece of design. It converts a collective action problem into an individual incentive, and it means the resource that actually protects the system is not a pile of collateral but the surviving members' willingness to take on the positions.

Not everyone clears directly. The structure is a hierarchy, and its shape matters for who bears what.

Clearing members face the CCP directly. Membership requires capital, operational capability and default fund contributions, so the number of firms is small.

Clients access clearing through a member, in a relationship called client clearing. A client's positions are held at the CCP through the member, and the client faces the member rather than the CCP.

That indirection creates a distinct exposure. If a clearing member fails while its clients are solvent, the clients need their positions and collateral moved to another member, which is called porting. Whether that works depends on how the collateral was held.

Segregation models differ. Individually segregated accounts keep a client's collateral identifiable and are the most protective and expensive. Omnibus accounts pool clients, which is cheaper and means a client can be exposed to losses caused by another client of the same member.

The practical instruction is that clearing does not make a client's exposure to their broker disappear. It changes its shape, and the segregation choice determines how much.

The third column is the useful reading. Each layer reached is a statement about how far the event exceeded what was planned for, and the first two layers being sufficient is the definition of the system working as intended.

What this structure genuinely delivers is that a member's failure does not propagate directly to its counterparties, because they were never facing it. That is a real achievement, and it is why central clearing was extended after the financial crisis.

What it does not deliver is the elimination of the risk. It has been pooled, ordered, and concentrated in one institution, and the consequences of that concentration are the last lesson.