Collateral vs Guarantee: A Thing, and a Second Party

Collateral hands the lender a claim over one identified thing the borrower holds. If payment stops, the thing may be sold and the proceeds applied to what is owed. A guarantee instead brings in a second party who has promised to pay when the borrower does not. Collateral argues about how much is lost after a failure; a guarantee argues about how many parties must fail first.

Two words turn up together so often that most readers file them as two strengths of the same idea, one a bit sturdier than the other. The two words are not that. Collateral and a guarantee are two different objects that happen to arrive in the same paragraph of the same document, and the sentence worth carrying away is the one about where each of them sits. One sits inside the borrower. The other sits outside it. Everything else follows from that.

The ordinary version is easier to feel than the bond version. Start there. A household wants to borrow to fix a leaking roof, and the lender is uneasy. There are two ways the household can make the lender less uneasy. The household can hand over the papers of the gold it already keeps in a locker. If the repayments stop, the gold can be sold and the money used to clear the loan. Or an uncle, who is not borrowing anything and does not need the roof fixed, can put his name on the paper and say that if the household does not pay, he will. The gold is a thing that already sits inside the household; the uncle is a second person standing outside it, and no amount of gold turns into an uncle.

Collateral comes first on its own terms, then a guarantee on its own terms, and only once both are standing are the two laid side by side.

One warning belongs at the very top, before a single figure appears. Palash Cements Limited, an invented cement maker, carries a five year bond with a 9.10 per cent annual coupon on Rs 1,000.00/- of face, on an annual compounding convention. The bond carries no security and no guarantee, and it has no credit rating either. The arithmetic below runs on a stated assumption and shows what a protection would do to a price. Showing what a protection would do to a price is a different claim from saying a bond has one, and a weaker one.

What is collateral, and where does its claim actually sit?

CollateralA specific thing belonging to the borrower over which a lender is given a claim, so that it may be turned into money if payment stops. is a specific thing belonging to the borrower over which the lender is given a claim. If payment stops, the thing may be turned into money and the money applied to what is owed. The definition repays slow reading. Three separate ideas are packed into it, and each one does work further down.

The first idea is that the thing is specific. Not the borrower generally, not whatever happens to be lying around when the dust settles, but an identified item written into the paper. A machine with a number on it. A building with an address. A block of receivables that can be listed. The act of giving the lender that claim over an identified item is a pledgeThe act of giving a lender a claim over an identified thing, so the claim attaches to that item rather than to the borrower in general., and a pledge attaches to an item rather than floating over the borrower as a whole.

The second idea is that the claim sits inside the borrower. The thing is already there. Palash Cements Limited does not go out and acquire anything new in order to pledge it; it takes something already sitting on its own side of the line and hands the lender a claim over it. Nothing arrived from outside. Nothing new was created. The pledge changed who has first call on one item that was already there.

The third idea is the one people skip, and it is the one this whole guide turns on. Nobody new is being asked for a promise. A pledge does not add a party. A pledge does not produce a fresh signature from anybody who was not already in the deal. A pledge produces a route instead. If payment stops, this item can be sold and the money applied. And a route is only worth what the item fetches when it is finally sold. Nobody knows that number on the day the bond is issued.

The arithmetic further down comes straight from that last sentence. Suppose the pledged item is a kiln. On the day the bond is issued, the kiln is working, cement is moving, and nobody is selling anything. On the day the kiln actually has to be sold, cement may be moving badly for everybody, and the set of buyers who want a second hand kiln may be exactly the set of people who are also having a difficult year. The item is the same item. The price it fetches is not the same number. Nobody has measured that gap for this bond, so every recovery figure below is labelled an assumption in the same sentence in which it is used.

Here is the consequence, and it is blunt. Collateral does not change whether the borrower fails. A pledged kiln does not sell one extra bag of cement, does not collect one overdue invoice, and does not make one payment arrive that was otherwise going to be missed. Collateral changes what is left over afterwards. A borrower with a pledged kiln and a borrower without one miss the same payment on the same day for the same reason; the two lenders differ only in what they can do next.

What is a guarantee, and who has to perform?

A guaranteeA promise by a second party to pay what is owed if the borrower does not. is a promise by a second party that if the borrower does not pay, that second party will. The centre of gravity in that sentence is the phrase second party, and once that phrase is held on to, the rest of the comparison builds itself.

Where does the claim sit? Outside the borrower entirely. The claim sits against another party, and against whatever that party has. A guarantee is not a claim over an item on Palash Cements Limited's side of the line, nor over anything else on that side of the line. If the borrower turns out to hold nothing whatever when the claim is finally resolved, the guarantee is unaffected. The promise never depended on what the borrower holds.

Who has to perform? A named party, and the standard that party has to meet is higher than it first sounds. The guarantorThe second party who gives the promise to pay, and who must be both able and willing at the moment of being asked. has to be able and willing at the moment of being asked. Able, meaning it actually has the money then, not that it looked comfortable when the paper was signed years earlier. Willing, meaning it pays rather than contesting whether the promise applies to this particular circumstance. And the moment of being asked is, by construction, a bad moment: nobody calls on a guarantee during a good year.

Go back to the household for a second. The everyday version does real work at this point. The uncle who signed is not asked for anything for years. He is asked in the month the household stops paying, and that is often a month in which something has gone generally wrong. If the uncle happens to work in the same trade, or has lent to the same people, or has money tied up in the same place, then the month that ruins the household is quite likely to be a difficult month for him too. A guarantee is worth least in exactly the circumstances that make it necessary, and that is a property of the arrangement rather than a criticism of any guarantor.

The consequence, stated as bluntly as the one for collateral. A guarantee does not change what a failed borrower leaves behind. A guarantee adds a party who would also have to fail before the lender is left with nothing. Two parties instead of one, standing between the lender and a loss.

An absence named rather than filled in

Palash Cements Limited has no credit rating anywhere on this platform. A rating scale, and the definition attached to each step of one, are set by the rating agencies and by the Securities and Exchange Board of India (SEBI) at sebi.gov.in, and they are revised, so a scale written out from memory is not merely stale on the day it changes, it is wrong. The row where a rating would sit is drawn empty below, with the reason printed inside it, and every sentence here holds without a rating.

The same applies to three other things a reader might expect. There is no measured default frequency here, no recovery study, and no series of past spreads behind any figure in this guide. An implied default rate is arithmetic on one price and one stated assumption, so it counts nothing that has actually happened.

Where do the two claims sit when they are drawn together?

Both sides are now defined, so they can be set against each other without either one being ranked. The cleanest way to see the difference is to draw the borrower as a boundary and ask where each claim lands relative to it.

Notice the arrow the drawing leaves out. No arrow runs backwards from either protection to the borrower to make the borrower pay. Both objects live entirely in the space after something has gone wrong. Neither one reaches into the months before, and neither one adjusts by a single rupee the amount the borrower can hand over on the due date.

Readers routinely treat a protection as though it made a borrower safer. Reaching only forwards is a general property of protections rather than a quirk of these two, and it deserves a picture of its own. A protection does not make the borrower anything. A protection changes what happens to the lender afterwards.

Which limb of the credit arithmetic does each one work on?

The arithmetic behind a credit spread multiplies two quantities together. One is a default rateA rate per year at which a borrower fails to pay, which is a different quantity from how much is lost when it happens., a rate per year that answers the question how often. The other is a loss given defaultThe share of the amount owed that does not come back, being one hundred per cent less the recovery rate on the same base., a share of the amount owed that answers the question how much. The named component behind the first is the probability of default, and the named component that sizes the base it applies to is the exposure at default; both are covered separately and are used here rather than rebuilt.

Now put each protection beside the limb it argues about. Collateral works on the second limb and a guarantee works on the first, and a reader who holds that distinction can follow any claim anybody makes for either.

Collateral is an argument about the recovery rateThe share of the amount owed that comes back after a default, an assumption here rather than a measurement. and therefore about the loss given default sitting on the same base. Collateral says nothing whatever about how often a failure happens. Take the pledged kiln again: it has no view on whether cement sells next year. The kiln has a view on what a lender walks away with in the year cement does not sell.

A guarantee is an argument about the first limb. A guarantee says that for the lender to lose anything, the borrower has to fail and then the second party has to fail as well. The promise has no view at all on what either of them leaves behind. Two protections, two different limbs, and the sentence that distinguishes them is short enough to remember.

The base and the period, named in the same sentence as the figure

Four quantities are about to be multiplied and divided, and they have four different bases and four different periods. Mixing them is the commonest error in this material, so they are named before they are used.

The multiplication that joins them is only meaningful once each one has been labelled. A recovery rate of 40 per cent of the amount owed and a default rate of 3.6667 per cent a year are not two percentages of the same thing over the same window; they are two different animals that happen to be written with the same sign after them.

What happens to the spread when the recovery assumption rises?

The arithmetic now runs, and it runs in the one direction that is not a trap. The starting figures come from the invented record used throughout this sequence. Palash Cements Limited borrows for five years at 9.10 per cent a year. On the invented SPOT curve, the five year government SPOT rate is 6.90 per cent a year. Both figures sit on an annual compounding convention, one discounting period a year. On a semi annual convention the same numbers would give different prices.

Subtract the second from the first. The gap is 2.20 percentage points, or 220 basis points over the five year government SPOT rate, for five years. The gap, and not any part of the 6.90, is what the rest of the arithmetic works on.

Now assume a recovery of 40 per cent of the amount owed. The 40 per cent is the assumption the whole sequence runs on, and no recovery study stands behind it. A loss given default is one hundred per cent less the recovery rate on the same base, so on that assumption it is 60 per cent of the same amount owed, or 0.60.

A relationship that only works in one direction has not been demonstrated, so the check runs backwards as well. 3.6667 per cent a year multiplied by 0.60 is 2.2000 percentage points, exactly where the arithmetic started.

Here is where the collateral argument enters, and here is the discipline that makes it work. Hold the implied annual default rate completely still at 3.6667 per cent a year, and move only the recovery assumption. Suppose a recovery of 70 per cent of the amount owed instead of 40 per cent. The loss given default falls from 0.60 to 0.30 on the same base.

Back-check it: 1.1000 divided by 0.30 returns 3.6667 per cent a year, the same figure step two produced. The default rate really did stand still while the spread halved.

The rupee reading makes the same point without any percentages at all, and it is worth having because a share floating free of its base is easy to misread. The amount owed on this bond is Rs 1,000.00/- of face. A loss given default of 60 per cent of the amount owed is Rs 600.00/- of every Rs 1,000.00/- owed. A loss given default of 30 per cent of the same amount owed is Rs 300.00/-. The quantity being shared out did not change; only the share did.

The rupee form of the loss per year was settled earlier in this sequence and follows from figures already given. An implied annual default rate of 3.6667 per cent a year applied to a loss given default of 0.60 on an exposure of Rs 1,000.00/- gives an expected credit loss of Rs 22.00/- a year. At a loss given default of 0.30 on the same exposure it gives Rs 11.00/- a year. Half again, the same halving the spread bar showed, expressed in rupees rather than in points.

Four control positions, printed so they survive without any interactive

The two positions worked above are the ends of a short range. Here are all four, with the implied annual default rate standing still at 3.6667 per cent a year at every one of them.

The units rule enforced in every line

A percentage point and a basis point are the same quantity written in two units. One basis point is one hundredth of a percentage point, so 2.20 percentage points is 220 basis points and 0.50 percentage points is 50 basis points. Every spread here appears in both units the first time it is written, and no figure in one unit is ever set beside a neighbouring figure in the other without both being labelled.

And no spread is ever written as a bare number. A spread is always over something and always for a stated maturity. So 220 basis points here means 220 basis points over the five year government SPOT rate, for five years. Drop either half of that and the figure stops meaning anything, because a spread over a different reference or for a different number of years is a different quantity that happens to be written with the same digits.

What does the spread do as the recovery assumption moves by hand?

The claim above is that collateral works on one limb and leaves the other untouched. A claim like that is easy to nod at without believing. Moving the assumption shows which of the two drawn objects actually moves. The control below shifts one thing only, the assumed recovery rate as a share of the amount owed, and everything else on the drawing is computed from it.

Across all four positions of the control, the bar shortens and the split under it shifts. Nothing the control touches has anything to say about how often a borrower fails, so the dashed marker on the right does not move by a pixel at any position. No borrower is getting safer here. One assumption about the aftermath is changing the price that assumption would justify.

Why can no guarantee be priced without a second party?

The collateral arithmetic above ran cleanly. A reader who has just watched a spread halve will reasonably expect the guarantee to get the same treatment two paragraphs later. The guarantee does not get it, and the reason follows.

Pricing a guarantee needs three things this platform does not hold, and each of the three is missing for a different reason.

1. A second party of any kind.There is exactly one issuer anywhere on this platform, which is Palash Cements Limited, and no other party of any description: no parent, no associate, no institution standing behind anything. A guarantee is a promise by somebody, and a single issuer leaves nobody to be the guarantor. Inventing one would mean inventing a whole second entity with a balance sheet and a story.
   One issuer, no second party. The requirement fails at the first step.
2. A basis for saying whether that party could perform.Even given a second party, pricing its promise needs some assessment of whether it would be able to pay when asked. No assessment of anybody exists on this platform, no rating is given to any party, and none may be invented, for the reason set out earlier about where rating scales come from and who maintains them.
   No assessment of any party exists here, and none may be written in.
3. A basis for saying whether the two failures would arrive together.This is the hard one, and it is the reason a guarantee cannot be priced by simply multiplying two default rates. A guarantor whose fortunes move with the borrower is a completely different object from one whose fortunes do not. The first adds much less than it appears to; the second adds much more. Nothing on this platform speaks to that question in any direction, so the arithmetic has no input for it.
   Whether the two failures arrive together or apart is unknown, and it decides the answer.

So the guarantee row below is drawn empty, with those three requirements printed inside it, beside a collateral row that carries the arithmetic worked above. Two rows, one filled and one blank, and the contrast between them is the finding rather than a gap.

One temptation catches careful readers rather than careless ones, and it is worth naming. Having been told that a guarantee works on the first limb, a reader may reach for two default rates and multiply them together, reasoning that both parties have to fail. Multiplying the two rates is exactly the step the third requirement blocks. Multiplying two rates together assumes the two failures are unrelated events, and whether they are related is the single most important thing about any guarantee. Neither rate on its own answers that question, so the multiplication has no licence.

What does each protection need at the moment it is called on?

The two protections stop resembling each other at the moment they are called on, and this is the part a reader can actually use on a document they are handed. Ask both protections the same four questions and the answers come out different every time.

Collateral needs three things at once. Collateral needs a thing that still exists. An item can be consumed, sold, moved or worn out in the years between the pledge and the claim. Collateral needs a claim over that thing that still holds when the claim is finally resolved. And collateral needs a buyer willing to pay for it at that moment rather than at the moment the bond was issued. People forget that third requirement, and the whole recovery assumption stands on it.

A guarantee needs three things too, and they are different things. A guarantee needs a party that still exists. The guarantee needs that party to be able to pay. And it needs that party to actually pay rather than dispute whether the promise covers this particular circumstance. Collateral fails when a thing turns out to be worth less than anybody assumed; a guarantee fails when a party turns out to be unable or unwilling, and those are not two versions of the same failure.

Several of those requirements are governed by rules set elsewhere. Whether a pledge holds when a claim is finally resolved, and the order in which claims are met, are set by the insolvency authority at ibbi.gov.in. Disclosure about security pledged against a bond, disclosure about a guarantee supporting one, and the duties placed on a trustee acting for the holders are all set by SEBI at sebi.gov.in. How a guarantee is treated when a credit exposure is measured is set by the Reserve Bank of India at rbi.org.in. Every one of those moves on its own schedule, so the authority is named instead, and each appears again as a blank row further down.

What goes wrong when the recovery assumption is moved against a held spread?

Now the error. The people who make it are the people who have understood the material, and that earns it more attention than most errors. The error is not made by somebody who has misread the definitions. The error is made by somebody who has learned the triangle correctly and then moved one input without saying which of the other two quantities they were holding still.

How does a lender or an analyst actually read a protection?

So which of the two is better?

The two cannot be ranked against each other, and the reason comes with the finding.

The two are not two strengths of the same protection. The two protections sit in different places. Each is called on from a different party. And they come up short in different ways. A pledged thing may fetch far less when it is sold than anybody assumed it would. A guarantor may turn out to be unable or unwilling at the one moment it matters, a circumstance rather than a deception. Ranking them would need a measurement of both failure modes on the same scale, and this platform holds neither and invents neither.

The comparison leaves a pair of questions instead, to ask of any protection anybody ever describes. Where does the second claim sit, and who has to perform? Those two questions place a protection, identify which limb of the arithmetic its supporters are arguing about, and show what has to be true on the day it is finally called on. A ranking would have to be re-derived for every document, and these two questions do not, so the questions are more useful than a ranking.

The three honest limits on every implied figure above

None of these three is optional and none of them is a footnote.

First, the 40 per cent recovery is an assumption. No recovery study stands behind it, and moving the assumption moves the answer with it. Holding the 220 basis point spread completely still and changing only the recovery assumption produces four different implied annual default rates out of the same unchanged price.

Same price, four answers, and the assumption is doing that much of the work. Notice also that this table is the failure block viewed from the correct angle: it is legitimate here precisely because the spread is named as the thing being held still, in the heading of its own column.

Second, the whole spread is being treated as compensation for credit. In a real market some part of a spread pays for not being able to sell the bond easily, and every basis point of that read as credit makes the implied default rate too high. Split 0.40 percentage points off the 2.20 as payment for a reason other than default and 1.80 percentage points remain; 1.80 divided by 0.60 gives 3.0000 per cent a year rather than 3.6667 per cent a year. Separating the credit part of the spread from the rest needs a second price, and the gap is stated rather than worked around.

Third, an implied default rate is what the price says. It is not a forecast and it is not a measured frequency of anything. Nobody counted defaults to produce 3.6667 per cent a year; it was solved backwards out of one spread under one assumption, and reading it as the probability that Palash Cements Limited fails has misread the arithmetic that produced it. The word implied is attached to the figure at every single use for exactly that reason.

Palash Cements Limited, its five year bond and the government SPOT curve used here are invented.
Educational material. Not advice on any investment, tax, budget or market position.