Collateral vs Netting: Two Ways to Cut the Same Exposure

No. Netting collapses two claims running in opposite directions into one. The amount at risk shrinks before anybody measures it. Collateral does not touch the debt at all and improves only what a lender recovers once the failure has happened. Different objects, different moments. On counterparty C1 the charge is worth twenty six times the netting agreement, in rupees, on the same day.

Two sentences placed side by side are where the confusion starts. Netting reduced counterparty exposure by 27.3 per cent. The charge reduced counterparty exposure by 42.6 per cent. Both sentences are true, both were computed correctly from the same counterparty on the same day, and read together they will teach a committee something false. One of those techniques moved Rs 6,48,000 of expected loss. The other moved Rs 1,68,48,000. The ratio is twenty six to one, hidden entirely inside two percentages that look like near neighbours.

The reason is not sloppy work. The two techniques act on different things at different moments, so the percentages attached to them are shares of different amounts. The confusion is not academic, so the separation is worth getting clean. The separation decides which negotiation a credit officer spends three months on.

The everyday version is already familiar, so start there. A shopkeeper runs a small provisions store and supplies a caterer who also delivers milk to the shop every morning. At the end of the month the caterer has billed the shopkeeper Rs 18,000/- and the shopkeeper has billed him Rs 74,000/-. The shopkeeper and the caterer settle the difference: Rs 56,000/- moves, and that difference, not the gross Rs 74,000/-, is what the shopkeeper would lose if the caterer shut the shutters tonight. Netting is the settlement of that difference. Suppose instead that the caterer leaves his delivery van keys at the shop until the account is cleared. He still owes the whole Rs 74,000/-. Nothing about the debt has moved. The keys change what the shopkeeper would be left holding on the morning the caterer disappears. Collateral is the keys in the drawer. Same relationship, same month, two completely different mechanisms, and no amount of similarity in purpose makes them the same tool.

The worked case is Vindhya Commercial Bank Limited, an invented mid-sized Indian commercial bank, and its largest single-name exposure, counterparty C1, Nirjhar Industries Limited, an invented steel and alloys maker. Every figure here belongs to those two invented entities and the case is in Rs crore throughout. C1 is the only one of the bank's ten largest single names with a netting agreement recorded against it. The scarcity is itself part of the lesson: C1 is the one counterparty in the book where the comparison can even be run.

What is netting, and what exactly is being set against what?

Netting is an agreement between two parties that says: whatever the two end up owing each other under the contracts covered by this document, only the difference is payable. Not each amount separately. The difference. A netting agreement is a document, signed while everybody is calm, and it does its work by changing the arithmetic of what is owed rather than by moving any money or any asset at all.

The single idea underneath it is that two claims running in opposite directions between the same two parties can be collapsed into one claim running in one direction. The collapse is the entire mechanism. Nothing is pledged, nothing is valued, nothing is registered anywhere. Two sums become one sum.

The set of contracts the document covers is called a netting setThe group of trades a single netting agreement covers. Amounts inside the set can be offset against each other; anything outside it cannot., and the boundary of that set matters more than anything else about it. A trade inside the set can be offset. A trade outside it, with the same counterparty, on the same desk, on the same day, cannot. Vindhya Commercial Bank Limited has two derivative trades with C1 and both of them sit inside one netting set. One is worth plus Rs 132 crore to the bank. The other is worth minus Rs 36 crore to the bank, meaning the bank would have to pay that amount if the contract were closed out today.

Now ask why netting is worth having at all. If C1 stopped paying tomorrow, what is the bank's claim? Without the agreement, the answer is Rs 132 crore. In a default the bank's obligation to pay Rs 36 crore does not disappear, so the trade worth minus Rs 36 crore does not reduce the claim. The failed counterparty, or whoever is administering what is left of it, is perfectly entitled to demand that amount while paying the bank only whatever the general pool of claimants eventually receives. The two amounts run in opposite directions and, without a document joining them, they live in two different queues. The asymmetry between the two queues is exactly what the netting agreement removes.

With the agreement in place, the failure of C1 triggers close-out nettingNetting that operates on default: every trade in the set is terminated, valued, and collapsed into one amount payable one way.: every trade in the set is terminated at once, valued, and reduced to a single amount owed one way. Rs 132 crore less Rs 36 crore is Rs 96 crore, and Rs 96 crore is the bank's claim. The reduction is 27.3 per cent of what the claim would otherwise have been. Almost everything that goes wrong in this comparison goes wrong through that 27.3 per cent, so the figure recurs through everything that follows.

Two properties of netting come back later as limits on what it can do, and both are worth naming now. The first is that netting only reaches amounts owed in both directions. A loan the bank has made is owed in one direction only, so there is nothing to set it against and no agreement can create something. The second is that netting is a promise about the future conduct of a legal process. Nothing has been handed over. The bank holds a piece of paper saying that, on the day it matters, two claims will be treated as one, and whether that promise holds is a question of enforceabilityWhether an agreement would actually be honoured on the day it is needed, in the place where the dispute would be heard. A legal question and not a measurement one. in the place the dispute would be heard.

What is collateral doing, if it is not making the exposure smaller?

The tempting summary, that both techniques reduce the exposure, is exactly what makes the two look interchangeable. Collateral has to be defined just as carefully as netting. Collateral does not reduce the exposure. Collateral never has.

A collateral arrangement gives the lender a claim on named assets of the borrower, created in advance, to be relied on if the borrower fails. The definition of a collateral agreement, the discount applied to the value of the assets, and the computation of the resulting cover are covered under collateral agreements and are used here as facts about C1. The three that matter to this comparison are these: the charge Vindhya Commercial Bank Limited holds over C1's inventory and receivables is valued at Rs 1,440 crore; after the bank's own 35.0 per cent discount, Rs 936 crore of it counts; and Rs 936 crore is 43.3 per cent of the Rs 2,160 crore exposure at default that the bank actually reports for C1.

Now the separation that governs everything that follows: after all of that, C1 still owes exactly what it owed before. Not one rupee of the facility has been repaid, not one rupee of the undrawn line has been cancelled, and the two derivative trades are the same two trades. The exposure is untouched. The change is in the answer to a different question: how much of that exposure would actually be lost on the day C1 fails. The answer falls because the bank has a claim on stock and book debts that it can turn into money.

The household version is the cleanest way to hold this. The caterer's van keys in the shopkeeper's drawer do not reduce his bill. The keys decide what is left if he never returns. The month-end settlement of the two accounts against each other does reduce his bill, right now, before anything goes wrong at all. One is a claim on value. The other is arithmetic on the debt.

Do the two techniques act at the same moment?

The two techniques do not act at the same moment, and every other difference follows from that one. Stated as a criterion and held, it classifies any credit risk mitigation technique in about four seconds.

Netting acts while the exposure is still being counted; collateral acts on what is left after the failure has already happened. Netting is arithmetic performed on the way to a number. Collateral is a claim exercised after that number has become a loss. A bank that has netted has a smaller exposure. A bank that has taken security has the same exposure and a better outcome inside it.

Two criteria, then, and they are the first two of six laid against both techniques. The first is what the technique acts on: netting acts on the exposure, collateral acts on the loss. The second is the moment it acts: netting before the exposure is measured, collateral after default is assumed. Everything else, the reach, the requirements, the failure modes and the arithmetic, drops out of those two.

How much of this counterparty can each technique actually reach?

Here the comparison stops being philosophical. The third criterion is reach: which parts of an exposure can a technique physically get hold of. And reach is where the answer on this particular counterparty becomes lopsided in a way no percentage on its own will show.

Take C1 apart before either technique is applied. The bank has lent Rs 1,680 crore and it is drawn. There is an undrawn committed line of Rs 720 crore, of which the bank counts half, Rs 360 crore, on its own assumption about how much of a line a failing borrower pulls before it goes. There is the derivative current exposureWhat would be lost on a contract if the other side failed today, being its positive value and nothing else. of Rs 132 crore before netting. And there is potential future exposureAn allowance for how far a contract's value could still move in the lender's favour before it ends. It is an estimate about the future, not a present amount. of Rs 24 crore. The bank sets it as an add-onA percentage of the size of a contract used to estimate how much its value could move, applied to that size rather than to what the contract is worth today. of 1.5 per cent, its own figure, on a notional of Rs 1,600 crore. The four parts add to Rs 2,196 crore.

Netting can reach exactly one of those four parts, and it is the smallest but one. The funded loan is owed one way and cannot be set against anything. The undrawn line is a commitment the bank has given, not an amount C1 owes the bank in the other direction. The add-on is not a claim at all. The add-on is an estimate about a future movement, computed on the size of the contract rather than on its value, so it does not shrink when the value is netted. The remainder is the Rs 132 crore of derivative current exposure. Rs 132 crore out of Rs 2,196 crore is 6.0 per cent of the counterparty.

Collateral has no such boundary. The charge over inventory and receivables is not attached to a trade or to a facility; it is a claim on assets that stands behind whatever C1 fails to pay, so it reaches into the loan, the drawn part of the line and the derivative position without distinction. One technique can touch 6.0 per cent of this counterparty and the other can touch all of it, and that single fact settles the comparison on C1 before any percentage is computed.

Which one does more here, measured on the same figure?

Two techniques cannot be compared on two different figures. Netting's headline is a share of derivative current exposure. Collateral's headline is a share of exposure at default. The comparison needs one number that both of them move, and there is exactly one: expected loss on the counterparty, being the probability that C1 fails multiplied by the share of the exposure that is lost when it does multiplied by the amount outstanding at that moment. C1 sits at internal grade 4, whose one-year probability is 0.45 per cent, and that is the grade's figure rather than a name-specific estimate. The bank assumes 40.0 per cent loss given default before any security is counted. Both figures are the invented bank's own.

Run C1 four ways and the argument ends. With neither technique the exposure at default is Rs 2,196 crore and expected loss is Rs 3.95 crore. With netting alone the exposure at default falls to Rs 2,160 crore and expected loss to Rs 3.89 crore. With the charge alone the exposure at default stays at Rs 2,196 crore, the Rs 936 crore of eligible collateral covers 42.6 per cent of it, loss given default falls to 22.95 per cent and expected loss to Rs 2.27 crore. With both, the position Vindhya Commercial Bank Limited actually reports, the exposure at default is Rs 2,160 crore, cover is 43.3 per cent, loss given default is 22.67 per cent and expected loss is Rs 2.20 crore.

So netting takes Rs 3.95 crore to Rs 3.89 crore, a saving of Rs 6,48,000. The charge takes Rs 3.95 crore to Rs 2.27 crore, a saving of Rs 1,68,48,000. Measured on the one figure both of them move, the charge is worth twenty six times the netting agreement on this counterparty.

And there is a reason the ratio is exactly twenty six rather than approximately twenty six. The reason is the cleanest way to hold the whole comparison. Once cover is expressed as eligible collateral divided by exposure at default, the expected loss computation collapses to a single line: 0.45 per cent times 40.0 per cent, being 0.0018, multiplied by the exposure at default less the eligible collateral. Every rupee removed from that bracket, by whichever technique, is worth the same 0.0018 of expected loss. Netting removes Rs 36 crore from the bracket. The charge removes Rs 936 crore from it. Rs 936 crore divided by Rs 36 crore is twenty six, to the paisa, and no rounding was involved anywhere.

Is there a setting at which netting overtakes the charge?

No. A comparison of this kind trains a reader to expect a crossing point, and on this counterparty there is no crossing point to find at any value of anything.

Push the control all the way. Suppose Vindhya Commercial Bank Limited negotiated so well that the second trade in the netting set were worth minus Rs 132 crore instead of minus Rs 36 crore, exactly cancelling the first. A perfect cancel is not a better agreement, it is the best conceivable outcome of the agreement the bank already has: netted current exposure of zero. Exposure at default falls from Rs 2,196 crore to Rs 2,064 crore and expected loss from Rs 3.95 crore to Rs 3.72 crore. The whole possible range of netting on this counterparty is a fall of 6.0 per cent in expected loss, and 6.0 per cent is also the share of the exposure the derivative position was to begin with. The two figures are the same statement written twice.

The charge alone, with no netting agreement in the picture at all, puts expected loss at Rs 2.27 crore. Perfect netting cannot get below Rs 3.72 crore. There is a gap of Rs 1.45 crore between the best netting can ever do and what the security the bank already holds does on its own, and nothing in the range of the control closes it. The two lines on that chart do not converge, do not touch and do not cross.

Do not draw the wrong conclusion from that. Netting is not a weak technique. Netting is a technique whose reach is decided by the shape of the relationship, and the relationship with C1 is a lending relationship with a small derivative position attached. On a counterparty where the two sides face each other under dozens of contracts running in both directions, the same agreement would reach almost the entire exposure and the ranking would reverse completely. The ceiling is a fact about C1, not a fact about netting.

What happens when both are used, and how is the saving split?

The reported position at Vindhya Commercial Bank Limited uses both, so the honest presentation is not two percentages but four states of one counterparty. Neither technique: Rs 3.95 crore. Netting only: Rs 3.89 crore. The charge only: Rs 2.27 crore. Both: Rs 2.20 crore. Every percentage anybody could want can be read off those four cells, and none of them can hide what it is a share of.

The total distance from the top left cell to the dark one is Rs 3.95 crore less Rs 2.20 crore, being Rs 1.75 crore, and a committee will eventually ask how much of that each technique earned. The attribution has an answer most people do not expect.

Attribute it in rupees and the answer does not depend on the order at all. Credit netting first and it saves Rs 6,48,000, after which the charge saves Rs 1,68,48,000. Credit the charge first and it saves Rs 1,68,48,000, after which netting saves Rs 6,48,000. Identical, to the rupee, both ways. The reason is the single line the arithmetic collapses to: expected loss is 0.0018 multiplied by exposure at default less eligible collateral. Netting subtracts Rs 36 crore inside that bracket, the charge subtracts Rs 936 crore inside it, and two subtractions inside one bracket do not care which of them is written first. In rupees, the split is fixed and the order is irrelevant.

Attribute it in percentages and the order changes the answer immediately. Applied first, netting is credited with 1.64 per cent of the Rs 3.95 crore it acted on. Applied second, the same Rs 6,48,000 is credited as 2.86 per cent. By then the charge has already brought expected loss down to Rs 2.27 crore, and the same rupees are a bigger share of a smaller amount. The charge moves the other way, from 42.62 per cent going first to 43.33 per cent going second. The rupees are a fact about the counterparty and the percentages are a fact about the order they were written in.

What does each one need before it works at all?

The fourth criterion is the entry price, and the two techniques could hardly ask for more different things. Netting needs two conditions and neither of them involves money. Netting needs positions that actually offset, meaning trades running in both directions between the same two parties, and an agreement that would be honoured on the day it is relied on. Collateral needs an asset that exists, a valuation somebody will stand behind, and a claim registered so that it ranks where the lender thinks it ranks. One technique is bought with paperwork and offsetting positions. The other is bought with somebody else's balance sheet.

The difference in entry price decides who can even attempt each one. A bank cannot net with a borrower who has no trades running the other way, and no amount of goodwill creates a position neither side has taken. But almost any borrower has something to charge. Security is therefore the technique available in nearly every lending relationship, and netting the technique available in a handful of them. Across the ten largest single names at Vindhya Commercial Bank Limited, a netting agreement is recorded against exactly one, C1. Security is described for four of them, and the record simply says nothing about the other six. Silence there is not the same thing as an absence of security, and no reader should convert it into one.

What makes each one fail?

A technique that fails quietly is worse than one never taken, so the fifth criterion is the one a control function cares about most. Netting fails on enforceability and collateral fails on valuation, and those are two entirely different control problems handled by two entirely different sets of people.

Netting fails as a legal event. The agreement is on the file, the arithmetic in the report has already been done using it, and then the moment arrives and a court somewhere declines to collapse the two claims into one. Nothing about the bank's measurement was careless. The Rs 96 crore was reported in good faith, and it becomes Rs 132 crore the moment the assumed offset turns out never to have been available. The everyday version is a promise between two neighbours that neither of them ever wrote down: perfectly sincere, and worth exactly nothing on the morning one of them denies it. Enforceability is therefore a question about the jurisdiction and the counterparty type rather than a question about the bank's own systems, and the answer for an Indian bank comes from the Reserve Bank of India at rbi.org.in rather than from anybody's internal policy.

Collateral fails as a measurement event. The charge is properly registered, nobody disputes it, and the assets are simply not worth what the file says. Steel inventory in a market where a steel maker is failing is not steel inventory in a normal market, and the discount that looked prudent when it was set was set against a valuation taken in a different world. The everyday version is the gold chain in the jeweller's safe when the price of gold has fallen by a fifth since the day it was weighed.

The sixth and last criterion is who provides the protection, and it explains the negotiation that surrounds each. Netting is a document between the two sides, so it costs a legal negotiation and nobody's assets move. Collateral is value handed over or charged by one side, so it costs the borrower real flexibility: assets pledged to one lender are assets no other lender will lend against. The difference in cost is why a borrower will often sign a netting agreement in an afternoon and argue about a charge for a quarter.

When does the difference between them change a decision?

Everything above is a comparison of two techniques on one counterparty, and a comparison of that kind is only useful if it changes what somebody does on Monday. The comparison changes two things.

The first is where negotiating effort goes. A relationship manager has a limited number of conversations with a borrower before goodwill runs out, and on C1 the arithmetic says which conversation is worth having. Rs 1,68,48,000 sits behind the security discussion and Rs 6,48,000 sits behind the netting discussion. The gap is not a close call and it is not a matter of taste.

The second is that the answer flips completely with the shape of the book, so the ranking on C1 is not a general result and must never be carried to the next counterparty unexamined. Where two sides face each other under dozens of contracts running in both directions, nearly the whole exposure is amounts owed both ways, netting reaches almost all of it, and it does so without asking the counterparty to pledge a single asset. In that relationship netting is both the larger technique and the cheaper one. On C1, where 94.0 per cent of the Rs 2,196 crore is loan, line and add-on, netting reaches a corner and security is the only technique that touches the money. Same two techniques, opposite conclusions, and the deciding factor was the shape of the exposure rather than any merit in either method.

What is worth carrying away from all of this?

One sentence holds the whole comparison. Netting changes what is owed and collateral changes what is recovered, and because those are different objects at different moments, the percentages attached to them are never comparable without their denominators. On counterparty C1 the two work out at Rs 6,48,000 against Rs 1,68,48,000, a ratio of exactly twenty six, and that ratio is simply Rs 936 crore of eligible collateral divided by the Rs 36 crore of trade value that got set off.

The second thing to carry is a habit rather than a fact. When a percentage arrives in a credit paper, a treasury note or a supervisor's return, the first question is not whether it is correct. Correctness can be taken as given. The question is what it is a share of, and whether the number sitting next to it is a share of the same thing. Here two perfectly correct figures, side by side, understated the difference between two techniques by a factor of about seventeen, and no auditor checking the arithmetic would have found anything wrong.