Collateral Agreements: How Counterparty Risk Is Reduced

Security is a written claim over named assets of the borrower, taken before there is any trouble. Security leaves the debt exactly where it stood and changes only what the lender keeps after a failure. Loss given default falls by the share of the exposure those assets reach once discounted. For counterparty C1 that share is 43.3 per cent, and expected loss drops by the identical figure.

Everything below sits on one separation, so make that separation now and hold it throughout. Any exposure has three properties and they belong to three different things. There is a size. There is a likelihood attached to the name that borrowed. And there is a severity, meaning how much of the money never comes back on the day that name stops paying. Security touches the third one only. Security does not make the borrower stronger and security does not make the loan smaller. Security changes what the lender walks away with afterwards.

Ordinary life carries the same lesson, even where nobody has put it in these words. A jeweller lends against a gold chain kept in his safe. The chain does not make the customer more likely to repay and it does not shrink the amount handed over the counter. The chain decides what happens on the morning the customer never comes back. Deciding that morning is the whole of what security does, at every scale, and the arithmetic below is only that sentence written down carefully enough for somebody else to argue with.

The worked case throughout 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 amount is in Rs crore. C1's exposure at defaultHow much is expected to still be outstanding at the moment the other side fails, which is not the same as the balance visible today. is Rs 2,160 crore, built from Rs 1,680 crore drawn, Rs 360 crore of a Rs 720 crore undrawn line at the bank's own 50.0 per cent assumption, and Rs 120 crore on the derivative position. The build-up of that figure is settled separately and used here as a fact about C1.

What is a collateral agreement, and what does it actually change?

A collateral agreementA contract giving one side a claim on identified assets of the other if the other fails to pay. is a contract, signed in advance, that gives the lender a claim on named assets of the borrower if the borrower fails. Three words in that sentence carry the load. Contract. The claim exists on paper and can be enforced rather than negotiated on the worst day of the relationship. Named. A claim over unspecified property is a claim over nothing. And in advance. A lender asking for security after a borrower is in trouble is asking for a favour rather than exercising a right.

Vindhya Commercial Bank Limited holds a registered chargeThe legal instrument by which a lender registers its claim over an asset of the borrower, so the claim sits on a public record and ranks against other claimants. over the inventory and receivables of C1. The registered charge is the artefact. The charge names a class of assets, it sits on a register, and it was created while the relationship was healthy. A charge over assets is not the only shape security takes: credit support provided by a third party, and an obligation backed by a sovereign, arrive at the same place in the arithmetic by a different route. A charge is what Vindhya Commercial Bank Limited actually holds against C1, so the charge is the shape worked through here.

Where a charge lands in the expected loss computation

Expected lossProbability of default times loss given default times exposure at default, being three numbers multiplied and never added. on any exposure is three numbers multiplied together, never added. The probability that the name fails inside a year. The share of the exposure that is genuinely lost when it does. And the amount outstanding at that moment. On C1 the bank uses its internal grade 4, whose one-year probability of default is 0.45 per cent, its own assumed loss given defaultThe share of an exposure that is actually lost when the borrower fails, after everything recovered. A severity, not a likelihood. of 40.0 per cent, and the Rs 2,160 crore exposure at default. Multiply the three and the answer is Rs 3.89 crore.

Now ask where a charge over inventory can possibly enter that product. Not the probability. A grade 4 name with security is still a grade 4 name. Security says nothing about whether the borrower will pay, only about what follows if it does not. Not the exposure at default either, at least not in the way this bank chooses to run it: the facility is the same size, the undrawn line is the same size, and the trades are the same trades. The charge reaches loss given default and nothing else. The ceiling on what any collateral agreement can do is therefore fixed before anybody counts how much collateral there is.

Why will the bank not count the collateral at what it is worth?

The charge over C1's inventory and receivables is valued at Rs 1,440 crore. The bank counts Rs 936 crore of it. The Rs 504 crore in between is the haircutThe percentage a lender knocks off the market value of collateral before counting it, to allow for the value falling or for the asset being hard to sell quickly., and understanding why a lender does this is the difference between reading the arithmetic and believing it.

Two things can go wrong between the day a valuation is written down and the day the bank has to sell. The value can fall. Inventory and receivables of a steel and alloys maker are not fixed objects, and a receivable from a customer who has also stopped paying is worth a fraction of its face. And the sale can be slow or forced. A bank taking possession of steel stock is not a steel trader, and it will accept whatever a real steel trader offers in the week it needs the cash. A haircut is not pessimism about the valuation; it is an allowance for the two things that stand between a valuation and a recovery, and a lender that skips it has quietly assumed both away.

Vindhya Commercial Bank Limited applies 35.0 per cent to this charge. The 35.0 per cent is this bank's own choice, not a requirement, a market convention or an industry norm. A different bank looking at the same steel stock could reasonably set a different figure, and the whole of the arithmetic below would move with it. After the discount the bank counts its eligible collateralThe value the lender is willing to count after the haircut, which is always less than the market value on the valuation report.: 1,440 times 0.65, being Rs 936 crore.

How much of the exposure does the security actually reach?

Eligible collateral of Rs 936 crore sits against an exposure at default of Rs 2,160 crore. Dividing one by the other gives the coverage ratioEligible collateral divided by exposure at default, being the share of what is at risk that the security actually reaches.: 936 over 2,160, or 43.3 per cent. The coverage ratio carries into everything else. No other figure on a collateral file does as much work.

The coverage ratio is easy to read as something grander than it is. Read it carefully. The ratio does not say the bank is 43.3 per cent safe. The ratio says that if C1 fails tomorrow, 43.3 per cent of what is outstanding has something identifiable standing behind it and 56.7 per cent has nothing standing behind it at all. The bank's assumption for the unsecured part has not changed: on that part it still expects to lose 40.0 per cent. On the covered part, in this bank's own model, it expects to lose nothing. So the blended loss given default is simply 40.0 per cent applied to the uncovered 56.7 per cent, or 22.67 per cent. The coverage ratio has walked straight into the severity without touching anything else.

Written as one line it stops looking like a trick. Loss given default after security equals 40.0 times one less 0.4333, or 40.0 times 0.5667, or 22.67 per cent. The 40.0 per cent is what the bank loses on a rupee with nothing behind it. The 0.5667 is the share of rupees in that position. The product of the two is the average loss across the whole exposure.

Why does the fall in expected loss come out as the coverage figure again?

The answer below surprises most people once and then never again. Predict it before reading on.

Six points off that line, so the numbers survive without touching anything. The spacing between the six points is the finding. Read them as a set.

Look at the last column. Every step of Rs 720 crore in the valuation takes exactly Rs 0.84 crore off expected loss: 3.89 to 3.05 to 2.20 to 1.36 to 0.52. The relationship is a straight line and not a curve, so the last rupee of collateral before full cover buys precisely what the first rupee bought, and then the line stops dead because there is no exposure left to cover. Beyond about Rs 3,323 crore the bank is holding security it cannot count. Holding it is a commercial decision rather than a risk one.

And now the identity. Expected loss without the charge is 2,160 times 0.45 per cent times 40.0 per cent, or Rs 3.89 crore. With the charge it is 2,160 times 0.45 per cent times 22.67 per cent, or Rs 2.20 crore. The fall is 1.69 over 3.89, or 43.3 per cent, and that is the coverage figure again. The match is not a coincidence and it is not an approximation: because the charge enters only through loss given default, and because loss given default after security is the old severity multiplied by one less the coverage, the whole product is multiplied by one less the coverage too. Reducing a number by 43.3 per cent and multiplying it by 0.567 are the same operation, and coverage is the only thing that moved.

Two routes, and they are not two answers

There is a second way to run the same arithmetic that many lenders prefer, and it looks completely different. Instead of leaving the exposure at Rs 2,160 crore and cutting the severity, deduct the eligible collateral from the exposure and leave the severity alone: 2,160 less 936 is Rs 1,224 crore, at an unchanged 40.0 per cent. Then expected loss is 1,224 times 0.45 per cent times 40.0 per cent, or Rs 2.20 crore.

The same number. Of course it is: 2,160 times 0.567 is 1,224, so applying the factor to the exposure or to the severity multiplies the same product by the same thing. Both routes are correct and they are one answer. Presenting them as two is the only real error here, and it leaves a reader believing a bank got two different results from one set of facts. One route is chosen, named as the route in use, and the other stated as the same figure rather than a second opinion.

How does a bank find out which exposures actually carry security?

Everything so far has been one counterparty. A credit officer holding a book of them has a different problem: not the arithmetic itself, but which names the arithmetic can even be run on. A map answers that question, and a map is a boring object that repays being built properly.

How to map Collateral and Netting

Four columns and nothing more. The counterparty. The security described against it. Whether a netting agreement is recorded. And the derivative current exposure carried. Netting bites only where derivative positions exist. Two of those columns are about the record rather than about the exposure, and that is the point: a map of risk mitigation is really a map of what the bank has written down about its own protection.

Here is that map across the ten largest single-name exposures of Vindhya Commercial Bank Limited, ranked by funded exposure as the bank publishes them.

Now the sentence that has to sit at the top of any map like this one, and which most maps leave out. A blank in the security column means the record does not say, and it does not mean there is no security. C3 may hold a first charge over its port cranes that nobody entered on this sheet. C3 may hold nothing instead. The map cannot tell, and a reader who takes silence as an absence has read a claim into the sheet that the sheet never made. Six of the ten rows are silent here, and six rows of ignorance is a perfectly respectable finding as long as it is labelled as ignorance rather than as unsecured lending.

The household version of the same filing problem is one everybody has had. A person knows the insurance policy exists because it was paid for; whether the papers are in the steel almirah or at a brother's house is a different question, and on the day a claim has to be made, the second question is the one that matters. A charge that is real and unrecorded protects the bank in law and does not protect it in any report anybody reads.

What does the map show once it has been drawn?

Four things, and only the first is the one a reader expects. Security is described against four of the ten: C1, C2, C4 and C5. A netting agreement is recorded against exactly one, C1. Five of the ten carry derivative current exposure: C1 at Rs 96 crore, C3 at Rs 24 crore, C5 at Rs 12 crore, C7 at Rs 36 crore and C10 at Rs 60 crore. Which means four counterparties carry derivative exposure with no netting agreement recorded against them, and between those four names it comes to Rs 132 crore.

In this bank's own record Rs 132 crore already means something else: what C1's derivative position is worth before netting, the plus Rs 132 crore trade standing alone. Say the figure carefully every single time. Two different objects wearing one number. Name the set, always. Rs 132 crore across C3, C5, C7 and C10 with no netting agreement recorded is one thing; C1's un-netted current exposure of Rs 132 crore is another; and a reader who meets the bare figure in a sentence has no way to tell which one arrived.

The fourth finding is the one that changes where an officer would look next. Set each name's derivative exposure against its own total exposure and the ranking inverts.

C10 is the most derivative-heavy name in the book at 7.7 per cent and it is the smallest name on the list, so the exposure most in need of a netting agreement is not the one anybody looking at a size ranking would reach for first. The inversion is the everyday shape of a mis-set priority. A shopkeeper checks the lock on the front shutter because the front shutter is the big one, and the small side door with no lock at all is where the loss comes from.

What does a charge do that a netting agreement does not?

Both techniques appear in the same map and they act in completely different places. A bank bothers with both for exactly that reason. A charge acts on loss given default: the exposure stays what it was and the severity falls. A netting agreement acts on the exposure itself, by collapsing a set of trades with one counterparty into a single claim so that a losing position is allowed to offset a winning one instead of counting for nothing.

On C1 the difference in force is not close. Start from the position with no netting at all, where the derivative reading is C1's un-netted current exposure of Rs 132 crore, exposure at default is Rs 2,196 crore and expected loss is Rs 3.95 crore. Now let the netting agreement do the most it could ever do and cancel the winning trade completely. An add-on is taken on contract size rather than on value, so C1's potential future exposure of Rs 24 crore is left standing. Exposure at default falls to Rs 2,064 crore and expected loss to Rs 3.72 crore. The fall is 6.0 per cent, and it is a ceiling rather than a typical outcome. The charge on the same starting position covers 936 over 2,196, being 42.6 per cent, and takes expected loss to Rs 2.27 crore.

So on this counterparty no value of the netted trade lets netting catch the charge. The crossing simply is not there. Netting as a technique in its own right, and the full comparison between the two, is covered separately; what matters here is only the map column and the fact that the two act on different factors.

Where does a collateral agreement actually go wrong?

The consequence is smaller than people fear and larger than they expect. Work it through. Revalue the same charge at Rs 1,080 crore rather than Rs 1,440 crore. Eligible collateral becomes 1,080 times 0.65, or Rs 702 crore. Coverage falls from 43.3 per cent to 702 over 2,160, or 32.5 per cent. Loss given default rises from 22.67 per cent to 27.00 per cent. Expected loss rises from Rs 2.20 crore to Rs 2.62 crore. Not one thing in the credit system changed, C1 is still grade 4 and the facility is still Rs 2,160 crore, and the answer moved 19.1 per cent.

The deeper version of the same failure is worse than staleness, and it is the reason a haircut on this particular asset should feel uncomfortable. Inventory and receivables of a steel and alloys maker are worth least in exactly the market where a steel and alloys maker fails. Demand collapses, the stock is a drug on the market, and the receivables are owed by customers who are themselves under. The valuation is not merely old; it was measured in a world where the borrower was fine and it is being relied upon in a world where the borrower is not, so the error runs in a direction rather than in both. A household knows this instinctively: the gold in the locker holds its price in a bad year, and the second-hand value of a family's trade tools does not.

Which loss given default is the whole case running on?

Every figure above rests on the bank's assumed 40.0 per cent loss given default, so it is worth saying out loud what that assumption is doing. Vindhya Commercial Bank Limited assumes 40.0 per cent on every non-defaulted grade, and separately holds provision coverage of 68.0 per cent against the exposures that have already defaulted. The 40.0 per cent and the 68.0 per cent describe the same event from either side of it, and they have never been reconciled. The gap is a finding about Vindhya Commercial Bank Limited rather than an error in the arithmetic.

The consequence reverses a sign. At 40.0 per cent, the model asks for Rs 343.6 crore across grades 1 to 9 and the bank holds Rs 494.4 crore of standard asset provisions, a cushion of Rs 150.8 crore. Run the identical computation at the 68.0 per cent the bank actually experiences and the model asks for Rs 584.1 crore, a shortfall of Rs 89.7 crore. Break-even sits at about 57.6 per cent, between the bank's own two figures, so any statement of the cushion is meaningless until somebody says which severity it is measured on. The charge over C1 cuts that severity by 43.3 per cent whichever of the two is the starting point. 68.0 per cent times 0.567 is 38.53 per cent just as 40.0 per cent times 0.567 is 22.67 per cent. A charge cannot settle which severity is the right one.

Who actually picks up a coverage ratio, and what do they do with it?

Four people read the same 43.3 per cent and do four unrelated things with it. Follow all four and the purpose of the ratio becomes obvious.

The credit officer inside the bank uses it to decide what more can be done with this name. C1 is the largest single-name exposure in the book at a reported total of Rs 2,496 crore, and with Nirjhar Alloys Private Limited, a wholly owned subsidiary the bank reports inside the group line rather than as a separate name, the group comes to Rs 3,168 crore. The record carries a collateral valuation for C1 and for nobody else. The officer cannot compute a coverage ratio for the subsidiary at all. The correct entry against every other name is that the measurement was not made, and an officer who scales C1's 43.3 per cent across the book to fill the column has manufactured comfort out of nothing.

The analyst outside the bank uses it to judge whether the disclosure can be trusted. The haircut, the drawdown assumption and the severity are all internal, so she cannot recompute any of them. She can ask three questions that cost nothing: on what date was the collateral last valued, what discount was applied and who set it, and how many of the disclosed exposures carry a described security at all. A bank that answers all three has thought about its measurement; a bank that publishes a secured percentage with no valuation date has published a number that ages silently.

The board committee uses it to decide where the security policy is actually binding. Four names described, one netting agreement recorded, and six rows silent is not a portfolio statement, it is a statement about a filing standard, and the fix is a data requirement rather than a credit one.

The mechanism is identical at every scale, and here is the household version. A person taking a loan against jewellery is doing exactly what C1 is doing. The lender is not saying the borrower is more reliable and is not lending less. The lender has valued the jewellery, knocked something off for the price falling and the trouble of selling, and lent against what is left. If gold prices halve, nothing about that person's job or salary has changed and the lender's position has still got worse. Security is a claim on a value that moves, so a coverage ratio is only ever as fresh as the last time somebody looked at the value.

Where do the rules on eligible collateral and haircuts come from?

Two bodies stand behind the treatment of collateral, and the order in which they are named matters more than most readers expect. The idea of recognising security and discounting it originates with the Basel Committee on Banking Supervision at the Bank for International Settlements, bis.org. The Committee publishes the credit risk mitigation framework, the standardised and internal ratings approaches within which collateral is recognised, and the treatment of counterparty exposures. Basel is where the concept comes from and where the reasoning is set out.

But a global standard is not what binds an Indian bank, and naming only the first source is a confident and common error. An Indian bank's own list of eligible collateral, its discounts, its valuation and re-valuation obligations, and how all of that interacts with provisioning, large exposures and capital, come from the Reserve Bank of India at rbi.org.in. Name the standard, then name what binds, then send the reader to the text of both. State no figure from either.