What a Credit Rating Claims and What It Cannot Claim

A credit rating is a published opinion that places one borrower, or one bond, above or below others in how likely the promised payments are to arrive. A rating ranks; it does not measure. A rank is not a probability, not a price and not an instruction to buy or sell, and Palash Cements Limited carries none.

What kind of statement is a credit rating?

Almost every mistake made with a credit rating comes from misreading what kind of sentence it is, so start there. A credit ratingA published opinion that orders a borrower, or one particular obligation, by how likely it is that what was promised gets paid. is an opinion about an ordering. The rating puts one borrower, or one specific bond, above or below other borrowers and other bonds in how likely it is that what was promised gets paid. The ordering is the entire content of the statement. A rating does not say by how much one sits above another, it does not say how likely either of them is to pay, and it is not the output of a measurement anybody else could sit down and redo.

The word for that kind of statement is ordinalRanking one thing above another without saying by how much. First, second and third are ordinal; a distance in metres is not.. An ordinal statement ranks. A cardinal statement measures. Most readers handle both without thinking about it, dozens of times a week, and keep them apart correctly right up until money is involved.

Here is the everyday version. Consider a queue outside a single office at nine in the morning. The queue states exactly one thing: who is ahead of whom. The queue states nothing about how long anybody will wait. A queue is an ordering and a wait is a measurement, so the person third in line might be inside in four minutes or in fifty, and the queue is silent on that either way. Given the queue alone and asked how long the fourth person will be standing there, the only correct reply is that the question cannot be answered from what was supplied. A credit rating supplies the queue, and the question readers keep asking of it is the question about the wait.

Set that against the other object this sequence has already built in full. A credit spread is a subtraction between two rates for the same length of time, and anybody holding those two rates arrives at the same answer. A spread is a measurement in the plain sense: it has a size, the size can be divided or multiplied, and two people who disagree about it are disagreeing about arithmetic rather than about judgement. An assessment has none of those properties. Two people who disagree about where a borrower belongs in an ordering are disagreeing about judgement, and no arithmetic settles it.

The difference between an ordering and a measurement is not a criticism of ratings. An ordering formed carefully by somebody who reads the borrower's books closely is worth having. One specific move does deserve criticism: taking the ordering and treating it as though it had a size. The rest of this guide is that one move, taken apart.

What exactly is being assessed, the borrower or the bond?

The second question is the one readers skip, and it decides what they are holding. An assessment can attach to two quite different objects, and the words used for them look alike enough that the difference slides past.

An issuer assessmentAn assessment attaching to the borrower in general rather than to any one bond that borrower has issued. attaches to the borrower taken as a whole. An issuer assessment is a view about that borrower's general capacity to meet what it has promised. An obligation assessmentAn assessment attaching to one specific bond, with its own terms, its own standing in the queue and its own security or lack of it. attaches to one specific bond. The bond has its own coupon and its own dates, its own standing in the queue if payment ever stops, and its own collateral or none. Two bonds issued by the same borrower can differ on every one of those, so the two assessments are being formed about genuinely different objects.

Take the household version. The point lands in three seconds. A shopkeeper known for years is, in general, good for money. The reputation is a view about the shopkeeper. Now the shopkeeper asks for two separate loans on the same afternoon: one where the lender holds the keys to his delivery van until repayment, and one where the lender holds nothing at all. The view of the shopkeeper has not changed between the two conversations. The view of the two loans has, and it should have. The second loan is a different thing from the first, even though the borrower is the same person on the same afternoon.

So a reader who has an assessment in front of them and does not know which of the two objects it attaches to does not yet know what has been assessed. The distinction sounds like pedantry until one notices how often the two get quoted interchangeably in a sentence that then draws a conclusion about a specific bond. An assessment formed about a borrower in general has said nothing about where any one bond stands in the queue, and an assessment formed about one bond has said nothing about the borrower's other promises.

Palash Cements Limited carries neither an issuer assessment nor an obligation assessment, and the absence is deliberate rather than an oversight.

Why is no rating scale printed here?

Because printing one would make the text wrong rather than merely old. The claim is sharper than it first sounds.

A credit rating is expressed on a scale. The scale, the steps that sit on it, and the definition attached to each of those steps are set by the rating agencies and by the Securities and Exchange Board of India (SEBI) at sebi.gov.in. The scale and the definitions are published, they are the agencies' own material, and they are revised. A reference work that writes such a scale out from memory has done something worse than going stale. A stale figure used to be right. A rating scale copied from a writer's recollection was wrong from the moment the scale changed, and nothing in the text would ever announce it.

The authority is named and the row is left standing empty rather than filled in. Seven separate rule sets on ratings sit in the jurisdiction section below, every one of them named at its source and not one of them written out.

The same refusal governs the borrower. Palash Cements Limited carries no credit rating anywhere on this platform. Inventing a rating would mean inventing a step on a scale, and inventing a step would mean inventing the scale. Inventing a scale is the exact thing that cannot be done honestly. And the invention would not be harmless. A rating printed cleanly beside a company name reads as a finding to almost every reader who meets it cold.

Three further absences travel with that one, and they matter to what any figure in this guide can be used for. No default studyA count of how many borrowers at each step of a stated scale failed to pay, over a stated period and across a stated population., no transition study and no measured frequency of anything stands behind the arithmetic below. Counting how often borrowers at a given step failed to pay is a separate job, done by somebody else and published separately, and arithmetic on assumed figures never amounts to a count.

Is a credit rating a probability of default?

No, and the reason is worth more than the answer. A reader who meets an assessment naturally wants a number behind it, and that instinct is right rather than lazy. Ordinal statements are hard to act on. Wanting the queue converted into a wait is entirely reasonable.

One document would do the converting, and the document has a name. A default study is a count rather than an opinion. To be worth anything a default study has to say four things at once: the scale it counts against, the population of borrowers it watched, the period over which it watched them, and how many at each step failed to pay. A failure rate over three calm years and a failure rate over three difficult ones are different numbers about different worlds, so strip any one of those four out and the count means nothing.

No default study stands behind anything above, and no transition study either, so no count exists of how often borrowers moved between steps. No step of any scale is attached to a number.

Hold on to the distinction that does the work: the number would come from counting, and an assessment on its own is not a count of anything. An agency forming a view about one borrower has not thereby counted how many similar borrowers failed. Somebody else does the counting, afterwards, and publishes it separately. Where that counting has not been done, or has not been read, the step stays a step and never becomes a rate.

What happens when a rating is set beside a spread?

Here the two objects meet, and the useful thing is that they refuse to reconcile. The price side is short and has to reach a number before the refusal means anything, so it runs first and in full.

Palash Cements Limited promises 9.10 per cent a year on Rs 1,000.00/- of face over five years, annual compounding throughout. On the invented government curve behind this sequence, money lent for the same five years earns a SPOT rateThe price of money handed over now and returned at one named date in the future, quoted as one rate for that one date. of 6.90 per cent a year. Subtract the second from the first and 2.20 percentage points remain, the same quantity written as 220 basis points. The subtraction is the whole of the observed part.

Now the assumption, and it is announced rather than smuggled. Suppose that when a borrower stops paying, 40 per cent of the amount owed eventually comes back. The 40 per cent is an assumed recovery rateThe share of the amount owed that comes back after a borrower stops paying, measured on the amount owed rather than on the price paid., and no recovery study anywhere stands behind it. The assumption is supplied so that the arithmetic has a second quantity to work with.

With both quantities in hand the relationship the spread claims to be can be turned around. A spread is being treated here as one year's worth of expected loss, and expected loss is a rate at which promises break multiplied by the share of money that breaks with each one. So divide.

The label on that figure matters as much as the figure itself. The figure is an impliedSolved backwards out of a price under a stated assumption, rather than counted from what happened or forecast about what will. annual default rate. The word is doing structural work and travels with the number every single time it appears. The rate was solved backwards out of a price. Nobody counted a default to produce it.

And that is exactly where the assessment side stops. The price side began with two quoted rates, took one subtraction and one announced assumption, and arrived at a figure anybody can reproduce. The assessment side begins with a published opinion, expressed as a step on a scale, and to reach a rate it would need the fourth link from the section above. No count of failures stands behind any step of any scale. So the row between the two objects stays empty, and no arrow crosses it.

Does the arithmetic close when it is run backwards?

The arithmetic does close, and running it backwards is not optional. A relationship shown in one direction is a recipe somebody has to trust. A relationship that closes on its own starting figure has been checked in front of the reader, and that check costs one line.

One caution about rounding, given in advance rather than after a reader has been tripped by it. Dividing 2.20 by 0.60 gives exactly three and two thirds, a figure that never terminates. Carrying four decimals inside the multiplication, 3.6667 times 0.60 returns 2.2000 points. Rounding first to the two decimal reading of 3.67 per cent a year and multiplying that back returns 2.2020 points instead. The reading and the working figure are different things, and an account that prints only the reading sets its reader up to check the arithmetic and conclude the error was their own. So both appear here: 3.6667 per cent a year is what the division gives and what any further arithmetic uses, and 3.67 per cent a year is how it is read aloud.

What does a credit rating not claim?

Here is the list, plainly set out, because the list is what readers are short of. Everything on it is something a rating gets asked for and does not supply.

A rating does not claim a probability. The probability question is the whole of the section above, and it stands first on the list because it is the misreading that costs the most.

A rating does not claim a price. An assessment was never a statement about a price to begin with, so it says nothing about whether a bond is cheap or dear at the spread it happens to trade on. Two bonds an agency has placed at the same step can trade at spreads far apart, and neither the agency nor the assessment has been contradicted by that.

A rating does not claim anything about how much would come back if payment stopped. Recovery is a separate assumption entirely, and it is the single assumption doing the most work of all, as the failure block below shows.

A rating does not claim to be a recommendation. An ordering is not an instruction, and no ordering turns into one by being quoted.

And it makes no claim about how the price of the bond will move. An assessment is about whether what was promised gets paid, and a price moves for reasons that include payment and then keep going. A price also moves for reasons of its own: the government curve behind it moved, sellers appeared, the amount of the issue changed hands more slowly than it used to. None of that is what an assessment is about.

Which limits travel with the implied figure?

Three, and they are gathered here rather than left as a note, because a limit in a note is a limit nobody carries away.

First, the 40 per cent recovery is an assumption, and no study of recoveries supports it. Moving it moves the answer with it, and the ladder above is the proof: at an assumed recovery of 30 per cent the figure reads 3.1429 per cent a year, at 40 per cent it reads 3.6667, at 50 per cent it reads 4.4000, and at 70 per cent it reads 7.3333. One price, four answers, none more true than the others. Whoever states the answer states the assumption in the same sentence, or does not state the answer.

Second, the whole spread has been treated as compensation for credit and nothing else. In a real market some part of what a borrower pays over a reference is compensation for the plain difficulty of selling the bond again on demand, and every basis point of that read as credit pushes the implied rate too high. Carving 0.40 percentage points out of the 2.20 on that reasoning leaves 1.80 points of credit, so the implied figure falls from 3.6667 to 3.0000 per cent a year. The subtraction does not separate the two. Notice the direction is fixed: because the entire spread was divided as though it were all credit, this error can only ever run one way.

Third, an implied default rate is what the price says. The figure is not a forecast and not a measured frequency of anything. Nobody counted a default to produce 3.6667 per cent a year, and it describes the price rather than the borrower. Changing only the assumption changes the figure while Palash Cements Limited carries on exactly as before. The figure is a property of the assumption, not of the borrower.

So what is a credit rating honestly good for?

Less than most readers expect, and still more than nothing. A rating is a second opinion, formed by somebody with access the reader does not have, expressed as a position in an ordering, at one moment in time. Every clause in that sentence is a limit and the sentence as a whole is still a useful object.

Used as a prompt to go and ask what changed and why, it earns its place easily. Somebody with better access has formed a view, and a view formed with better access is worth knowing about even when it cannot be audited. Used as a starting point for an analyst's own reading, it earns its place too, and it shows which questions somebody thought were the hard ones.

Used as a number, as a price, as a forecast, or as permission to stop reading, it does not earn its place, and each of those four is a different way of asking an ordering to carry a size. The sentence above is the one to carry away, and it is the same sentence the queue outside the office was teaching at the start: the ordering is real, the ordering is useful, and the ordering is silent about the wait.

A reader arriving wanting to be told which assessment is safe will find no scale, no step, no definition of any step, and no assessment for any borrower. Each of those could only be supplied by inventing it, and an invented scale printed beside an invented company reads exactly like a finding to the next person who quotes it.