Default waterfalls compared on one axis — who is mutualised, and whether anyone priced it
Every market has a constitution nobody reads: its default waterfall — the pre-agreed sequence in which one participant's failure is absorbed by others. This essay puts the major waterfalls side by side on a single axis: the clearing house's layered fund, the crypto exchange's automatic deleveraging, the bank's statutory creditor hierarchy, the discretionary loss allocation of an operator's rulebook, and the residual waterfall of the ordinary insolvency estate. The differences that matter are not the loss-absorbing amounts. They are who consented, whether the consent was priced, and who holds discretion at the moment the trigger is pulled.
The finding is not that mutualisation is a scandal. Mutualisation is frequently the product — netting is bought with a written put to the survivors, and one of the clearest large-scale tests of the bargain passed. The scandal, where there is one, is narrower: the legal allocation is far more precise than the commercial description suggests, its price is separately observable almost nowhere, and in the newest structures — the ones marketed as eliminating intermediary risk — the waterfall is discovered at default rather than disclosed at entry.
An earlier essay argued that the settlement layer of financial markets has no constitutional document — no founding text that says who controls finality, on what terms, accountable to whom — and that the vacancy is maintained by the parties who profit from it. That argument was about the system level. One layer down, the picture inverts. Every individual market has a constitution. It is precise, binding, and in force. It is the default waterfall: the rulebook provision, statute, or contractual terms that specify, in advance of any dispute, whose money absorbs whose failure, in what order, up to what amount.
Almost nobody reads it. Not because it is hidden — CCP rulebooks are public, resolution law is in the Official Journal, exchange terms are a click away — but because it describes a state of the world that every participant, at the moment of signing, assigns a probability near zero. The waterfall is the contract you agree to most carelessly and depend on most completely.
This essay reads five loss-allocation mechanisms — only some of them literal waterfalls — and compares them on the axis the outcome turns on: who is made to absorb another party's loss, and whether that absorption was priced — carrying a separately observable price for the specific loss-transfer mechanism — or merely signed.
The central counterparty is the only structure on this list whose waterfall is the explicit product. Under EMIR, the sequence is codified: the defaulter's own initial margin is consumed first; then the defaulter's contribution to the default fund; then a dedicated tranche of the CCP's own capital — the "skin in the game", calculated by reference to a quarter of the CCP's minimum capital requirement; and only then the default fund contributions of the surviving members (Regulation (EU) 648/2012, Art. 45; RTS 153/2013, Art. 35). Beyond the funded layers sit the recovery tools: capped cash calls on survivors, and — under the EU's CCP recovery and resolution regime — haircuts on the variation margin gains of members who happen to be in profit (Regulation (EU) 2021/23).
Read as architecture, three features stand out. First, defaulter pays — in theory. The first two layers belong to the failed member, and in the ordinary case the waterfall never reaches anyone else. Second, the CCP's own tranche is deliberately thin, and thinness is the design: a clearing house that absorbed losses from its own equity would be running a proprietary insurance book; a thin tranche keeps the CCP an administrator of mutualisation rather than a principal, and keeps its incentives aligned with margin adequacy rather than loss appetite. Third — and this is the line the marketing does not carry — the survivors' layer is a written put. Within the relevant default fund, every clearing member has effectively written default protection on the losses of the others, in an amount that is funded, capped, and then, through cash calls and gains haircutting, less capped than the funded number suggests.
The put gets exercised. In September 2018 a single Norwegian power trader, clearing his own positions as a member of Nasdaq's commodities arm, blew through his margin on a Nordic–German spread trade; the loss ran to some €114 million and consumed his collateral, the CCP's junior capital tranche, and then roughly two-thirds of the surviving members' commodities default fund, which they were required to replenish within days (Finansinspektionen, default notices 2018; FI stability reporting; sanction decision, 2021). The members had signed the put years earlier. What none of them had was a price for it: default-fund contributions carry real economic costs — collateral, capital, liquidity — but the mutualised tail risk the power trader imported into the pool was never charged to anyone as a separately quoted premium.
Within the fund, every clearing member has written default protection on every other. The premium on that put is not separately quoted.
An exchange running auto-deleveraging executes the same logic at machine speed. When a leveraged position is liquidated into a moving market, the proceeds fail to cover the loss, and the venue's insurance or liquidation fund is exhausted, ADL closes the winning positions on the other side, at a price and a size the winners did not choose. The waterfall is an algorithm, the mutualised parties are — in the common design — selected by profitability and leverage rank, and consent was granted by clickwrap. In the standard design, open positions on the profitable side are ranked by unrealised profit and effective leverage, and the queue is deleveraged from the top — the traders most right and most levered absorb the defaulter's loss first. A companion essay, The Foreign Hands in Your Pocket, dissects the mechanics; for present purposes one comparative point matters. ADL is the most honest waterfall on this list about timing — it executes instantly and visibly, position by position — and the least honest about consent: the counterparties who absorb the loss were not asked at the moment of allocation — their consent, such as it is, sits in the venue's terms — funded no dedicated layer, and price the put, if at all, through folklore about which venues socialise losses.
Bank resolution is the waterfall written by the sovereign. The BRRD hierarchy runs from shareholders through additional tier one and tier two instruments and senior debt down to eligible deposits — covered deposits are excluded from bail-in altogether and stand behind the guarantee scheme, itself funded ex ante by levies on the industry (Directive 2014/59/EU; Directive 2014/49/EU). The statute also fixes the point at which public money may enter the waterfall: as a general condition, shareholders and creditors must first have absorbed losses of at least eight per cent of the bank's total liabilities including own funds before the resolution financing arrangement can contribute (BRRD Art. 44(5)). On paper this is the most explicitly codified waterfall in the comparison — set by statute, instrument terms and resolution rules rather than an operator's choice, and reflected, at least partly, in the yields of the instruments that stand in it.
Practice has twice shown where the paper ends. In Cyprus in 2013 — before the BRRD existed, in the episode that helped produce it — depositor loss allocation ran in real time: uninsured deposits at the island's largest bank were converted to equity in stages — an initial conversion, then a further tranche from temporarily frozen balances — that ultimately took 47.5 per cent of every balance above the insured ceiling (Central Bank of Cyprus; ECB materials, 2013). The depositors were in the statutory waterfall all along; they discovered their rung at the moment of conversion. And in March 2023, holders of Credit Suisse's additional tier one instruments — roughly sixteen billion francs of them — were written to zero by administrative order while shareholders, nominally junior, received UBS shares worth around three billion (FINMA decision, 19 March 2023). The write-down was contractually available: the instruments' documentation permitted extinction at a viability event, ahead of any formal resolution. The holders had bought an instrument whose contractual loss-absorption mechanics differed from the hierarchy the market commonly associated with seniority — and the yield they had collected priced the assumed hierarchy, not the documented one.
The bank waterfall therefore teaches the general lesson in its sharpest form: seniority is not what the label says, it is what the documents permit under the trigger that actually fires. A statutory hierarchy with a contractual trapdoor is, at the moment of default, a contractual hierarchy.
In March 2022, with nickel prices vertical and a short position of extraordinary size on the other side, the London Metal Exchange suspended trading and cancelled every trade executed that morning. A session's trades — roughly twelve billion dollars in aggregate — were unwound as if they had never existed; Elliott Management alone claimed over four hundred and fifty million dollars in forgone gains and pursued the exchange through judicial review, unsuccessfully — the courts upheld the exercise of the rulebook's emergency powers (R (Elliott Associates) v LME, EWHC 2023, upheld on appeal 2024).
The nickel episode belongs in this comparison because it is the limiting case of the entire subject. The LME's loss allocation that morning did not run through the margin sequence in its clearing rulebook; it ran through a discretion clause. Strictly, cancellation is not a waterfall at all — no funded layer was consumed in order. It is an emergency loss-allocation mechanism that works by editing the ledger: retroactively deciding which gains and losses had occurred at all. Analytically, it functions as tear-up performed before settlement rather than after default. And the courts' answer — the rulebook permits it — confirms the essay's premise from the other direction: consent to a rulebook is consent to everything the rulebook allows, including the clause that lets the operator decide, after the fact, who eats the loss. Consent by rulebook is formal, not informational.
At the bottom of the comparison sits the structure with no bespoke mutualisation waterfall: the ordinary insolvency estate. Insolvency law has a ranking of its own — security, expenses, preferences, unsecured, subordinated — but a holder whose instrument turns out to be a bare contractual claim enters the estate without a proprietary claim to the underlying assets: no segregation, no client-asset protection, no right that survives the intermediary. They enter the residual sequence pari passu with every other unsecured creditor, behind every security interest and statutory preference, for whatever the estate recovers over however many years the administration runs. Two earlier essays established how often tokenised and platform-held structures resolve to exactly this position — the wallet does not tell you, and the offering documents frequently do, in language no purchaser reads with insolvency in mind.
The crypto insolvencies added a refinement worth recording: even inside the residual waterfall, allocation rules get written after the fact. In the FTX proceeding, claims denominated in coins were converted to dollar amounts as of the petition date, so that creditors' recoveries — respectable in dollar terms — were measured against a bottom-of-the-market valuation of what they had actually deposited. Fixing conversion at the petition date is no crypto invention — ordinary insolvencies fix valuation and currency dates the same way; what was distinctive was the scale of the gap between what the wallet showed and what the claim became. That is a waterfall decision — who bears the asset's subsequent appreciation — made by the forum, at default, under rules no depositor evaluated at entry. The residual waterfall is not the absence of allocation; its statutory ranks are fixed. What it leaves open — the characterisation of the claim, the valuation date, the currency of account — is settled at the worst possible moment to negotiate any of it.
Put the five structures on the axis, keeping the kinds of consent distinct. The clearing member contracted and funded. The bank creditor accepted a statutory regime by purchasing into it — the ranking is imposed by law, not agreed. The ADL counterparty assented by clickwrap. The LME participant accepted a discretion clause. The estate creditor generally negotiated nothing bespoke at all. Pricing: every mechanism has an economic cost somewhere — fund contributions tie up collateral, estate risk sits inside credit spreads — but only the bank waterfall comes close to a separately observable price: AT1 spreads exist because the write-down risk is public — though the spread bundles other risks with it — and Credit Suisse showed that even this visible pricing tracked the assumed hierarchy, not the documented one. The operative distinction is not priced against unpriced; it is whether the price of the specific loss-allocation mechanism is observable, separately attributable, and aligned with the documented waterfall. Discretion at the trigger comes in kinds as well as degrees: mechanical in the algorithm, managed in the CCP's default process, legal in resolution — an authority choosing among tools — and highest, though still bounded by the rulebook and the courts, in the operator's emergency powers.
Now the counterargument, at full weight, because it is mostly right. Mutualisation is not a defect of the clearing model; it is the clearing model. Multilateral netting compresses gross exposures by orders of magnitude — a compression this publication has defended elsewhere against the atomic-settlement enthusiasts — and the compression is purchased with the survivors' put. One of the clearest large-scale tests of the bargain did exactly what the model promised: when Lehman defaulted into SwapClear with an interest-rate book of around nine trillion dollars notional, the portfolio was hedged and auctioned off using only a fraction of the initial margin held, and no surviving member's fund contribution was touched (LCH default-management disclosures, 2008). That validates one service in one asset class, not the architecture everywhere — but whoever regards the CCP waterfall as a hidden scandal must still explain why its severest realised test in the flagship asset class ended with: the defaulter paid.
The defensible claim is therefore narrower than the polemic version, and worth stating precisely. The problem is not that losses are shared. It is that the sharing is signed everywhere while separately priced almost nowhere; that the pricing that does exist tracks labels rather than documents; and that in the newer structures examined here — rulebook, terms of service, code — the movement runs toward less consent, less observable pricing, and more discretion, marketed as the opposite.
The scandal is never mutualisation. It is mutualisation discovered at default rather than priced at entry.
The unpriced put has beneficiaries, and they are not hidden. Operators can benefit twice: they sell certainty at the front of the rulebook and retain discretion at the back, and the gap between the two is option-like discretion — exercised, on the record above, at the moments it was worth the most. Sophisticated members benefit relative to unsophisticated ones: a member that reads the default-management provisions can tier its clearing relationships, cap its fund exposure across venues, and route its riskiest flow toward the venues where the mutualisation terms are least analysed. And instruments whose marketed seniority diverges from their documented mechanics stay cheap to fund for exactly as long as the divergence goes unpriced: every basis point between the yield of the assumed waterfall and the yield of the documented one is margin captured from the holder's inattention.
For the holder, the essay compresses to three questions, askable before any account is opened or any instrument bought. First: which waterfall am I in — statute, fund, algorithm, discretion, or estate — and which document says so. Second: is my contribution capped, and in what currency of pain — funded amounts, cash calls, gains haircuts, tear-up, or the open-ended arithmetic of the estate. Third: who holds discretion at the trigger, and what has that party done the last time the discretion was worth money. Nothing in the answers requires prediction. All of it is in documents that exist today, in force, unread.
Loss, in a levered financial system, is conserved in the one sense that matters: someone must ultimately bear it. It is not eliminated by clearing, resolution, algorithms, or tokens; it is routed. The routing table is written down, in every case, before the loss exists — in a rulebook provision, an annex to a directive, a paragraph of contractual terms, a prospectus trapdoor. What differs across markets is only whether the parties on the receiving end knew, funded, and priced their position in the table, or discovered it.
Which yields the essay's one usable sentence. You do not choose your losses; nobody does. You choose your waterfall — and the choice is made at account opening, in the moment of accepting terms unread, not at default, when the reading is done for you.
Figures for the Nasdaq 2018, Cyprus 2013, Credit Suisse 2023, and LME 2022 episodes follow the primary regulatory and court records listed above. The ordering of CCP recovery tools beyond the funded waterfall is described for the EU regime and is not uniform across jurisdictions.
Written in a personal capacity. Analytical views only — not legal or investment advice.
Julian Gretzinger — Investor and writer on monetary history, real wealth mechanics, and financial markets. substack.com/@juliangretzinger