CR 16 – Netting, Close-out and Related Aspects

Two banks that trade over-the-counter (OTC) derivatives with each other rarely hold a single contract. A dealer and a hedge fund may build up hundreds or even thousands of open swaps, options and forwards, some of which are worth money to one side and some of which are worth money to the other. Many of these contracts partly hedge one another, and many of them require cash flows or deliveries on the same dates. Netting is the family of techniques that collapses this tangle of offsetting obligations into fewer, smaller amounts.

Netting plays two distinct roles. The first is operational and concerns money that is actually due on a given day: if both parties owe each other payments on the same date, it is simpler and safer to exchange one net amount. This is called payment netting or settlement netting, and it is aimed mainly at settlement risk, the danger that one side pays away its leg and never receives the other. The second role only matters when something goes wrong. If a counterparty defaults, the surviving firm needs a legal right to stop performing, value everything, and offset what it owes against what it is owed. This is called close-out netting, and it is aimed squarely at counterparty risk. A central counterparty (CCP) faces the same problem on a much larger scale when one of its clearing members fails.

Why two trades between the same parties create problems

Picture parties A and B with just two transactions between them, each generating its own schedule of payments. Two difficulties appear even in this tiny portfolio. On any date when both transactions pay, the parties would be exchanging gross amounts in opposite directions, so each is exposed to the risk that its own payment leaves while the incoming one never arrives. Folding the two payments into a single net transfer removes most of that risk. The second difficulty concerns default. If B fails, A might be legally bound to keep paying on the transaction that has moved against it while having no realistic prospect of collecting on the transaction that has moved in its favour. That asymmetry also creates doubt about whether A can replace the two transactions with a new counterparty at sensible terms.

Seen broadly, netting aggregates obligations while leaving the market risk position unchanged or nearly unchanged. What it reduces are the other risks that come from carrying many gross obligations:

The rest of the lesson follows a simple split. Some netting works on cash flows, which must be interchangeable for the offset to make sense. Other netting works on values, which can be offset across any transactions whose mark-to-market (MTM) can be measured. Cash flow netting includes payment netting, currency netting, clearing rings and portfolio compression. Value netting includes close-out netting and set-off. The legal framework that makes value netting work in practice is the International Swaps and Derivatives Association (ISDA) Master Agreement.

Payment netting combines payments that are due between the same two counterparties on the same day, normally in the same currency. The payments can come from a single contract, such as the fixed and floating legs of an interest rate swap that happen to fall on one date, or from separate contracts, such as two swaps in the same currency that both reset that day. Instead of two transfers in opposite directions, one party sends the difference.

The benefits are practical. Fewer payments mean less processing time and cost, which also reduces operational risk. Because each side delivers less, the amount that could be lost if the other side failed mid-settlement is smaller, so counterparty and liquidity risk fall too. None of this depends on how the trades are cleared: payment netting applies in bilateral markets and in centrally cleared markets alike.

Bilateral compression: removing redundant trades between two parties

Two counterparties can go further than netting individual payment dates. If their book contains positions that offset each other, such as a payer swap and a receiver swap on the same terms, they can agree to tear up the redundant trades altogether. This is bilateral compression. It does not reduce the net exposure between the two parties, because the offsetting trades contributed roughly equal and opposite values anyway. What it does reduce is gross exposure and gross notional, and that brings real gains: fewer transactions to administer, fewer documents that could be challenged legally, and lower operational cost.

There is also a regulatory motive. Some regulatory methods are deliberately simple and measure risk partly from gross notional or gross exposure. Under such methods, a book full of offsetting trades can look riskier than it is. Shrinking gross notional without changing the net position is one way to stop being penalised for redundancy. When three or more parties cooperate to do the same thing across a network of trades, the exercise becomes multilateral compression, which is treated later in the lesson.

Payment netting works neatly when both payments are in one currency. Foreign exchange (FX) contracts are harder, because settling a currency trade means paying one currency and receiving another. The two legs settle in different payment systems, often in different time zones and at different hours. A bank can therefore deliver the full notional of the first currency, N_curr1, before it has received the equivalent notional of the second currency, N_curr2. If the counterparty fails in that window, the whole principal is at risk, not just the MTM of the trade. This is known as FX settlement risk or Herstatt risk, a name taken from a historic bank failure in which counterparties paid one currency and never received the other.

Non-deliverable forwards

One way to sidestep the problem is to avoid exchanging principal at all. A non-deliverable forward (NDF) is settled in cash: at maturity, the parties compute the difference between the agreed NDF rate and the spot FX rate prevailing at that time, apply it to the notional, and one party pays the resulting amount. Because no gross exchange of notionals ever takes place, payment netting is in effect written into the contract terms.

Continuous linked settlement and payment versus payment

Many FX trades cannot be settled net because one or both parties genuinely need the foreign currency. Those trades must be settled by a gross exchange of notionals. To make that safer, banks launched the continuous linked settlement (CLS) system in 2002, covering a set of eligible currencies. In a CLS settlement, bank A pays its currency into CLS Bank and bank B pays the opposite currency into CLS Bank. CLS Bank releases the outgoing payments to A and B only once both incoming currencies have arrived. This arrangement is called payment versus payment (PVP). Both parties still make their full intended payments, but neither leg can settle without the other, which removes the window in which one bank has paid and the other has not. CLS also allows payments in the same currency to be netted multilaterally across all trades settling that day, which is an operational saving on top of the risk reduction.

The episode attracted heavy criticism; one German newspaper labelled KfW the country’s dumbest bank, and two management board members and the head of risk control were suspended. One of the board members later won a lawsuit against the bank over his dismissal. The lesson for risk managers is that settlement risk is not a technicality: on the day a counterparty fails, a single unmatched payment can cost the entire notional.

Bilateral netting needs only the agreement of the two parties concerned, but its benefit stops at the edge of each pair. Once three or more parties are willing to cooperate, obligations can be offset around a chain of counterparties, which is multilateral netting. The oldest form of this is the clearing ring, used on futures exchanges well before modern central clearing existed.

A clearing ring was an informal arrangement among a group of exchange members, three or more, who agreed to accept one another’s contracts and to let counterparties be swapped around. Joining was voluntary, but once a member joined, the exchange’s rules bound it to the settlements the ring produced. Some members stayed out of rings altogether, and others belonged to several. The essential condition was that each participant had to accept a substitute for its original counterparty.

In the arrangement shown, D has an obligation of 100 to B and is owed 100 by C. Whatever else is happening in the market, those two equal positions allow D to be rung out: D leaves the ring, and the two obligations are replaced by one obligation running directly from C to B. D’s net position was zero before and remains zero, but D now has no open contracts left to manage or to close out. Rings in this way simplify the web of dependencies among members and make it easier to close out positions.

Who gains from a ring

The benefit is collective but not shared evenly. D gains clearly because its offsetting positions disappear. A is indifferent, since none of its positions change. B and C have the same economic positions as before, but each now faces a different counterparty: B is owed by C instead of by D, and C owes B instead of D. Whether B regards that as an improvement depends on whether C is a stronger or weaker credit than D. If C is weaker, B has been made worse off by a ring it did nothing to create.

A member sitting at the end of a ring, holding only a long position or only a short position, has nothing to offset and so no reason to ring out. Historically, some members declined to join rings for exactly this kind of reason; a firm might prefer a larger exposure to a counterparty it trusted over a smaller exposure to one it did not. So a ring that reduces total exposure across the group can still be resisted by individual participants, a tension that reappears in modern compression services.

Consider three parties, A, B and C, whose same-day, same-currency payments have already been reduced bilaterally. B owes A 10, A owes C 30 and C owes B 20. Each party both pays and receives, which signals that further reduction is possible if all three cooperate. A’s position is the key: it receives 10 from B and pays 30 to C. If the 10 that B owes A is instead routed as a reduction in what A pays C, A no longer needs anything from B at all.

The modern version of the clearing ring in bilateral OTC markets is portfolio compression, which delivers multilateral netting benefits through the cooperation of many counterparties. TriOptima’s TriReduce service is the best-known provider, compressing trades in interest rate, credit, FX and commodity derivatives; by 2017 it reported more than $1 quadrillion of notional principal eliminated by participants.

In a compression exercise, several counterparties partly or wholly terminate existing transactions, sometimes replacing them with new ones, so that the overall notional of the trades in the system falls. Strictly, compression targets notional value rather than the count of contracts, although the two usually move together. The objective is to cut the total notional of everyone involved; an individual participant may not see its own notional fall, although participants can impose constraints, for example to stop their exposures rising.

Compression is needed because OTC books grow over time and accumulate redundant positions, for example when a trade is unwound by entering an offsetting trade rather than terminating the original. Compression strips out that redundancy while keeping each participant’s market risk the same, so settlement risk, counterparty risk and operational cost all fall. It may also lower systemic risk because fewer contracts would need replacing if a large participant defaulted. Two limits apply. Returns diminish as the achievable multilateral netting is used up, and the process assumes counterparties are broadly interchangeable, which in practice means they need similar credit quality.

Steps in a compression cycle

The best overall answer can produce awkward results for a single pair of firms. Positions between two counterparties may grow or even change sign, and MTM values and risk sensitivities may shift. For that reason participants can set constraints, such as a cap on total exposure to a given name that mirrors their internal credit limits.

Credit default swaps (CDS) are the cleanest case for compression because the market standardised them precisely so that offsetting trades would be interchangeable. Large banks and ISDA agreed standard coupons and standard maturity dates. CDS now trade with a fixed premium plus an upfront payment, and scheduled termination dates fall on 20 March, 20 June, 20 September or 20 December. Under the older convention every trade could carry its own coupon and maturity, which made offsets messy. With standard terms, positions can be bucketed simply by reference entity (a single name or an index) and maturity, and everything in a bucket is fungible. The same standardisation also made central clearing of CDS practical.

Problems with compressing CDS and swaps

Compression is not free of side effects. CDS settlement can depend on whether the credit event is triggered by the protection buyer or the protection seller, so a long and a short position that look like perfect offsets on paper may not have exactly the same economic value. New trades created by a compression run may also fall under regulation, such as bilateral margin rules, from which the old trades they replace were exempt.

Interest rate swaps are harder still. They are normally struck at par with a fixed rate set by the market on the trade date, so two swaps can have identical floating legs but different fixed legs. Plain notional netting cannot cancel them. More advanced techniques such as coupon blending combine swaps with different fixed rates into fewer trades with an equivalent blended rate.

Regulatory push for compression

Post-crisis regulation has encouraged compression. Under the European Market Infrastructure Regulation (EMIR), financial and non-financial counterparties with more than 500 non-centrally-cleared contracts outstanding must have procedures to consider, at least twice a year, whether a compression exercise would reduce counterparty risk, and must explain to their supervisor if they decide not to compress. Legitimate reasons include having positions so directional that no offsets exist, or accounting, tax and legal drawbacks. An end user hedging a loan with a swap, for example, may find that compressing it against other swaps disturbs hedge accounting.

A small network shows why compression is an optimisation problem rather than a mechanical calculation. Five counterparties hold positions in a single fungible product. The figures below are called notionals, but they could equally be exposures; only their relative sizes matter.

Counting every position from both sides, as each counterparty would report it, total gross notional is 2 × 625 = 1,250. There is no position at all between counterparties 1 and 3.

What should the algorithm minimise?

The aim is to cut gross notional while leaving every counterparty’s net position untouched, but the objective is a matter of judgement. Minimising total notional is the obvious choice, yet it does not penalise one very large position or a large number of small ones. Minimising the sum of squared notionals would target large exposures, and minimising the count of positions would target interconnectedness. Constraints matter as well: counterparties 1 and 3 might insist on keeping no exposure to each other. Commercial services have tended to use fairly simple methods.

The right-hand solution achieves 110, which is the lowest possible here: net payers owe 10 + 25 + 20 = 55 in total, and at least that much must flow, giving 2 × 55 = 110. It gets there by creating a position between 1 and 3 where none existed before. If 1 and 3 refuse exposure to each other, the best attainable answer is the left-hand one at 130, which shows how counterparty constraints weaken compression. Both end states also flip the direction of the 4 and 5 position: originally 5 owed 4 a notional of 75, and in each solution 4 owes 5 a notional of 20.

The traditional techniques covered so far all shrink the gross size of a portfolio without altering its market risk. Their headline benefit is lower settlement risk and, through smaller amounts owed in future, lower counterparty risk. Three further benefits are less obvious.

The regulatory row deserves emphasis. Cutting gross notional can look pointless when the net exposure is the same before and after. But some regulatory measures are deliberately crude and partly built on gross notional, and the leverage ratio in particular has given banks a strong reason to compress. Several thresholds that decide whether a firm must clear centrally or exchange bilateral margin are also set in gross notional terms.

The limit of cash flow netting

Cash flow netting only works when flows are interchangeable: the same asset class, the same product, the same payment date and the same reference rate. Yet a firm’s portfolio with one counterparty usually contains transactions that are similar but not fungible, and often spans several asset classes. A swap in euros and an option on an equity index cannot have their cash flows netted. The more general idea is to net the values of transactions instead.

Value netting matters most in default, and default in derivatives is unlike default on a loan. A lender to a bankrupt firm knows its claim: the outstanding principal and interest. The recovery is uncertain but can be estimated, and bankruptcy simply takes time. A derivatives user faces something worse. Its contracts were probably hedges that need constant rebalancing, and once the counterparty is in default the cash flows stop, so the firm needs to put on replacement trades quickly. A long and unpredictable bankruptcy process compounds the loss because the firm cannot be sure when, or whether, its contracts will be terminated.

Two further questions follow. Because derivatives can be either assets or liabilities, can the ones owed to the defaulter be offset against the ones owed by it? And can that offset reach across different lines of business, such as a loan and a swap with the same client? If the client defaults, the bank wants to know whether those exposures stand alone or can be combined. Close-out netting answers the first question; set-off, discussed later, addresses the second.

Close-out netting is the contractual right, usually set out in an ISDA Master Agreement, that takes effect when a counterparty defaults and lets the surviving party terminate every transaction and settle one net value promptly. It has two parts that work together:

With close-out netting, every transaction, whatever its maturity and whether in or out of the money, collapses into one number. If the survivor owes that number, it pays it; if it is owed, it lodges a bankruptcy claim for it. The effect is that the survivor realises its gains against its losses straight away and only joins the queue of unsecured creditors for the net amount. Because the calculation uses MTM values at the time of default, close-out netting works across any products covered by the agreement and needs no matching of cash flows. The survivor’s valuation may later be disputed in court, but that does not stop it from terminating at once and replacing the contracts elsewhere.

Payment netting addresses settlement risk on the day money changes hands; close-out netting addresses counterparty risk before that day, sometimes called pre-settlement risk. It also simplifies what happens after a default. Surviving firms normally try to replace the trades they lose, and without netting the number and notional of replacement trades would be much larger, which could itself disturb markets at a fragile moment.

Legal enforceability

Close-out netting is only as good as the law behind it. Most countries with significant financial markets have adopted netting legislation for derivatives. ISDA has obtained legal opinions supporting the close-out and netting provisions of its Master Agreements in 54 jurisdictions, and 37 countries have legislation that explicitly makes netting enforceable. Some jurisdictions remain where enforceability in default is unclear. There, a bank may decide netting is enforceable enough for pricing and perhaps accounting, accepting the legal risk, but may be unable to recognise it in regulatory capital.

This also explains why actively removing trades, as in bilateral compression, beats relying on close-out netting. If two parties hold two exactly opposite positions, tearing them up removes all legal risk, whereas leaving them in place means depending on a court to enforce netting if one party fails.

Standard counterparty risk models, including the usual approach to computing credit value adjustment (CVA), assume a simple rule for default. If the survivor owes money to the defaulted party, it still pays in full, so its position after default is the same as before. If the survivor is owed money, it loses part of that amount depending on recovery. This asymmetric rule is convenient for modelling, but real defaults often depart from it.

Walkaway and tear-up features

Some older OTC contracts included a walkaway or tear-up clause, which let a firm cancel its transactions if its counterparty defaulted. A firm would only use this when it owed money to the defaulter, because walking away from a contract that is in its favour would mean giving up a claim. The clause does nothing to reduce credit exposure. What it does is give the survivor a gain on contracts that are against it, by allowing it to stop paying and never settle the amount owed. Such clauses were common before the 1992 ISDA Master Agreement and have been rare since; they are not part of standard ISDA documentation, though they have appeared occasionally in trades after 1992. They do not mitigate counterparty risk as such, but the potential gains offset the potential losses.

The Enron and Lehman cases show that even without an explicit walkaway clause, a survivor can try to profit from a default. Instead of closing out contracts that are out of the money (OTM) to it, the firm leaves them in place and simply stops making the underlying payments, a stance that some readings of ISDA documentation appeared to support. When the swaps are deeply OTM for the survivor, and so strongly in the money (ITM) for the defaulter, the gain from delay can be large. In each case the defaulter’s administrator went to court, with mixed results.

Why walkaway is undesirable

Any walkaway arrangement imposes an additional cost on the defaulting party’s estate and its other creditors. It also creates moral hazard: a firm that stands to profit when its counterparty fails has an incentive to help bring that failure about, for example by withdrawing funding or cutting lines. For these reasons, walkaway features are widely regarded as something to avoid.

Close-out exists to put a number on the portfolio at the moment of default. If the survivor is a creditor, that number is its claim in the bankruptcy; if it is a debtor, that number is what it must pay into the estate. The valuation generally refers to replacement cost, because the survivor will usually replace the lost trades, either directly or indirectly through replacement trades, macro-hedges or the unwinding of existing hedges. Replacement costs tend to be added to the value to compensate the survivor.

Base value or actual value?

The first difficulty is choosing which value to use. The base value is the standard valuation without valuation adjustments. The actual value includes the survivor’s xVA terms, such as credit, funding and capital adjustments, and so better reflects what the trades are really worth to it. The actual value is more relevant economically, but it requires xVA to be defined, and xVA is complex and hard to pin down objectively in a legal setting.

When the survivor is a creditor, the overall xVA adjustment is usually negative because costs outweigh benefits, although a strongly negative future profile could change that. So the actual value is below the base value, and the survivor would do better to claim on a base valuation, which gives it a bigger claim. When the survivor is a debtor, the position can reverse. Funding benefits can make the actual value higher than the base value, and since a higher (less negative) value means less to pay, the survivor would then prefer to reference the actual value. Either way, the survivor has an incentive to pick whichever definition favours it.

Replacement costs and the legal value

On top of xVA there is the plain cost of replacing the trades, such as bid-offer spreads. The survivor may reasonably replace its trades, so these costs can form part of the value at default, and they are hard to separate from xVA because both appear as charges on the replacement trades. The result is a commercially reasonable value that includes such costs.

Because default valuation is so contestable, it helps to name one more quantity, the legal value. This is what a survivor might claim on the strength of the contract wording, possibly inflated relative to the true value plus genuine costs. A gap between the legal value and the commercially reasonable value is likely to end in litigation, and good documentation should try to close it, since any excess the survivor obtains comes out of the pockets of the defaulter’s other creditors.

The ISDA Master Agreement is the market-standard contract under which OTC derivatives are documented between two parties. Its purpose is to put every transaction between them under one legal umbrella, so that on a default all of them can be terminated together and reduced to a single net amount. It is this framework that makes close-out netting enforceable, defines how the amount owed is calculated, and, in the 2002 version, provides a set-off right. Because the close-out definition fixes the economics of default, it also feeds directly into how counterparty risk and xVA are measured.

The close-out amount is what one party owes the other after a default. If it is positive from the non-defaulting party’s side, that party has a claim on the defaulter’s estate; if negative, it must pay the estate. Even though the claim will not be paid in full, its size matters. The parties will disagree about it: the non-defaulting party sees its cost of replacing the trades, its replacement cost, as the right answer, while the defaulter may object that this includes charges such as bid-offer spreads that it does not itself face.

The 1992 methods

Under market quotation, the determining (surviving) party asks market makers for at least three quotes on the portfolio and combines them, using the middle value when there are exactly three. If three quotes cannot be found, market quotation is treated as having failed. Under loss, the determining party calculates in good faith its total losses or gains, meant to be the amount that would put it in the position it would have been in had the contracts been performed. That can include loss of bargain, funding costs and trading costs such as terminating or re-establishing hedges.

Market quotation is designed to be objective, but it needs liquid markets in the products involved. Liquidity tends to vanish just after a large default, as with Lehman Brothers, and is thin for exotic products or non-standard terms that the survivor wants to replicate. Since 1992 structured trades and unusual terms, such as one-way margin agreements, have multiplied, which has made it hard to find dealers willing to quote realistically and has led to many disputes. Loss has the opposite flaw: it hands the determining party too much discretion. A survivor might even have an incentive to let market quotation fail so it can fall back on loss, and any gain it takes would come at the expense of the defaulter’s other creditors.

The 2002 close-out amount

The 2002 Master Agreement replaced both methods with a single close-out amount. It reflects the determining party’s losses or costs, or gains, in replacing the transactions or obtaining their economic equivalent, determined in good faith and in a commercially reasonable way. In reaching it, the determining party may use firm or indicative quotes from third parties (which may reflect its own creditworthiness and the terms of collateral agreements with those parties), market data from third parties, and internal information it uses routinely to value similar trades. Compared with market quotation it is diluted, since it does not demand tradable quotes; compared with loss it is more disciplined, since reasonableness is judged objectively. So the three methods sit on a line: market quotation is the most objective, loss the most flexible, and close-out amount in between.

Some firms kept the 1992 agreement because market quotation seemed more objective, but the problems it caused in the crisis, especially around Lehman, pushed the market towards the 2002 definition. In 2009 ISDA published a close-out amount protocol that let parties amend their 1992 agreements to use close-out amount instead of market quotation and loss by signing a single document, rather than renegotiating with each counterparty. Close-out valuations under these definitions appear to allow xVA to be included to the extent it is part of the cost of new trades, unlike valuations for collateral or ordinary terminations, which usually use base values. That is more realistic but makes xVA calculation recursive, since the xVA at the default time itself enters the close-out value.

Close-out netting under an ISDA agreement is accepted as enforceable in almost every major jurisdiction, so banks routinely count on it when calculating regulatory capital and pricing new trades. The harder question is how far netting can reach. Large institutions trade loans, repos and derivatives across interest rates, FX, commodities, equities and credit, and they would like to net as much of that as possible against a given client. Enforceability gets murky because these trades are booked in different legal entities across different regions.

Netting across products within one category, which is cross-product netting such as rates against FX, is normally possible because the same Master Agreement covers both. Netting across categories is a different matter. The case of derivatives and loans is common: a bank may lend to a corporate at a floating rate and pair the loan with an interest rate swap linked to it, offering the client a packaged fixed-rate loan. Inside the bank the two are handled separately. If the client defaults when the bank is owed money on the loan but owes money on the swap, it is not obvious that the two will be offset, and it is equally unclear whether loan collateral can be used to cover derivative exposure. Banks may build such offsets into pricing and accounting, but they get no capital relief for them.

What set-off is

Set-off is a broad right that lets a party reduce an amount it owes to another by an amount that party owes it.

Set-off and payment netting have the same economic result in most cases but differ in form. Set-off acknowledges that cross-claims exist between the parties and allows equal claims in opposite directions to cancel. Netting produces a single contractual claim at any moment, so the separate claims never exist side by side. Set-off also generally concerns actual obligations, whereas close-out netting concerns a calculated amount. That is why set-off can, in principle, apply amounts due under other agreements against an ISDA close-out amount.

The 2002 ISDA Master Agreement includes a standard set-off clause allowing any termination payment to be offset against amounts owed to that party under other agreements. Legally, then, derivatives might be set off against loans. Whether that works depends on the wording of the different documents, the legal entities involved and the law of the relevant jurisdiction. Strong legal opinions are essential, yet because defaults are rare there are often no test cases to show whether set-off would actually be enforced.

Close-out netting is the most powerful single tool for reducing counterparty risk, and the OTC derivatives market would almost certainly not have reached its size and liquidity without it. Because offsetting trades cancel in value, the credit exposure in the market grows much more slowly than the notional outstanding. Regulators have recognised netting, at least in part, in capital rules, which allowed banks to expand their derivatives businesses without capital rising in step with notional.

Data from the Bank for International Settlements (BIS) measure the effect through a netting benefit. Gross market value adds up the positive and negative replacement values of all contracts; gross credit exposure is what remains after legally enforceable netting. As markets have grown and concentrated among fewer dealers, the netting benefit has risen steadily, and netting now reduces exposure by close to 90%.

Side effects of netting on market behaviour

Netting changes how the market behaves in subtler ways. Net positions, though smaller, are more volatile than the gross positions behind them, since a net figure is a small difference between two large ones, and that volatility can itself add to systemic risk.

Netting also makes it costly to exit a position. Suppose a firm wants to unwind a trade. If it hedges with a new dealer, it removes the market risk but now carries counterparty risk to both the original dealer and the new one, unless a later compression run removes the offsetting pair. Getting rid of the counterparty risk properly means trading back with the original dealer, who knows the firm is under pressure to do so and may quote poor terms. The firm must either accept those terms or live with the extra counterparty exposure. The same logic applies when a firm needs several different hedges. If it needs both interest rate and FX hedges, which are imperfectly correlated, placing them with the same dealer lowers the combined counterparty risk and may earn better pricing. But it also encourages firms to keep trading with the same dealers, which builds concentration risk.

Finally, netting changes reactions to a counterparty in trouble. If exposures were driven by gross positions, every firm trading with a weakening counterparty would rush to cut positions and stop new trades, which would deepen its distress. With netting, a firm whose net MTM to the troubled counterparty is negative has no current exposure and is much less alarmed. It will still worry about potential future exposure and may ask for collateral, but the reduced panic can lower systemic risk.

Regulatory mandates for central clearing and bilateral margin complicate netting. With central clearing, trades that would have sat in a bilateral netting set with one counterparty are moved to the CCP, where each clearing member faces the CCP on all its cleared trades. That allows multilateral offset across what used to be many bilateral counterparties, and it is often cited as the main advantage of central clearing; an IMF report described the ability to cut systemic risk through multilateral netting as a CCP’s primary advantage.

This looks similar to compression but differs in an important way. Compression needs standardised, fungible contracts so that trades can be torn up and replaced. CCPs also need standard contracts, but for operational reasons, margin calculation and the close-out of a defaulting member. Because a CCP offsets the value of trades rather than notionals or cash flows, it captures multilateral netting that compression cannot. Two weakly correlated trades with different counterparties, such as interest rate swaps in different currencies, gain from netting at a CCP but are not fungible enough to compress. Put simply, CCPs can compress risk, while bilateral compression can work only on objective quantities such as notionals and cash flows.

When clearing only part of the market raises exposure

Multilateral netting is clearly best when every trade is covered. In practice the netting is split, or bifurcated, because some trades cannot be cleared and stay bilateral, and because there are several CCPs. This bifurcation is acute for firms with offsetting trades in different products, such as a swaption hedged with a swap or an index CDS hedged with single-name CDS, where one leg may be clearable and the other not.

The same pattern would arise if the uncleared trades were cleared at a second CCP. Losing bilateral netting can outweigh the gain from multilateral netting, so that clearing increases total exposure. Duffie and Zhu showed with a simple model that achieving an overall netting gain from a CCP is not guaranteed: it needs a relatively small number of CCPs clearing a relatively large share of trades. Bilateral margin rules avoid this clearing split in principle, because trades stay bilateral, but many firms signed new collateral agreements for new trades while leaving old trades under existing ones. Collateral is then bifurcated across two agreements, and there may even be legal risk if the trades were judged to sit in two separate netting sets.

Close-out netting looks like pure gain from the derivatives market’s point of view: exposures are smaller and close-outs are simpler. For the financial system as a whole the picture changes, because netting does not create value; it shifts value towards derivatives creditors and away from everyone else with a claim on the defaulter.

The broader point, made by Pirrong among others, is that netting, margining and central clearing can all be good for the OTC derivatives market without being good for financial markets overall, because they redistribute risk rather than remove it. A bank’s derivatives counterparty risk, and its capital, may fall through netting, while bondholders and other unsecured creditors bear more of the loss in a default. That can change behaviour elsewhere on the bank’s balance sheet. It is fair to ask whether lowering systemic risk in derivatives markets at the cost of raising it in other markets is a good trade.