The Spread Thesis: A View on Credit, Stated Testably

A spread thesis is a claim about what a price is paying for, written so that somebody else can test it. The thesis names the reading the price gives at a stated assumption, the conditions that would have to hold for that reading to be the right one, and the observation that would break it. A thesis of this kind says nothing about what anybody should do.

Here is the awkward thing about a credit view. By the time an analyst sits down to form one, the price has already formed one. Somebody agreed to lend and somebody agreed to pay, and the rate they settled on is their reading of the borrower, written in a number. An opinion arrives second, so a thesis cannot start from one. It starts by pulling the reading out of the price, and only then asks which conditions are holding that reading up.

The tea stall outside a busy office works this way. The stall runs monthly tabs for the regulars, and it charges the tab customers a little more than the ones who pay in coin. The extra is not payment for the bookkeeping. The extra is the stall owner's own reading of two things at once: how often a tab walks away, and how much of it is gone when one does. Either half of that reading can be argued with. The extra rupee on its own cannot be argued with. The rupee is only where the two halves landed.

What is a spread thesis a claim about?

A spread thesis is a claim about what a price is paying for. A thesis is not a claim that a spread will narrow. A thesis is not a claim that a spread will widen. And it is very deliberately not a claim that anybody should do anything.

The distinction between the two claims is not fussiness. The distinction decides whether a reader can do anything with the note at all. A claim about what a price is paying for can be picked up by somebody holding the same three figures and worked through line by line; they will either land where the writer landed or they will find the line where the two parted company. A claim that a spread is attractive can only be nodded at or shrugged at. One of those two claims can be examined and the other can only be agreed with.

A thesis in this sense is made of three written parts, and all three are written down at the same sitting. The reading, taken from the price. The conditions that would have to hold for the reading to be right. The breaking observation, the specific thing that would make the writer drop the claim. With any one of them missing, what is left is an impression with a number in it.

One thing is missing from those three panels: there is no direction in them. A thesis about what a price is paying for holds still while the world moves around it; a thesis about where a price is going has to be right about the world. Being right about the world takes a record of how spreads have moved through time, and the second kind of claim cannot be built without one.

What does the price already say, before anybody has a view?

Palash Cements Limited, an invented cement maker, borrows for five years at a contracted rate of 9.10 per cent a year. The five year node of the invented government SPOT rateA rate that runs from today to one named future date, with nothing paid out in between. Every rate in this material is annual and says SPOT or FORWARD. curve stands at 6.90 per cent a year. Both rates run on annual compoundingThe clock ticks once a year, so nothing gets added in mid-year steps. The same digits on a twice a year clock price a schedule differently, so the convention gets stated., which is the clock every rate below keeps, and the choice of clock matters enough to state rather than assume.

Rates hide the size of things, so now put the gap in money. The rest of this borrower's figures carry a face amountThe amount the borrower has agreed to repay at the end. Rs 1,000.00/- is used across this material so figures can be compared throughout. of Rs 1,000.00/- for this borrower, and the same amount is used below rather than a fresh one.

So: why would anybody hand over Rs 22.00/- a year that they did not have to? Not for the trouble of lending, and not as a reward for bravery. The extra is somebody's arithmetic on two questions and nothing else. How often does the money stop arriving? And when it stops, how much of it is gone for good?

Answer the second question and the first one falls out of it. Suppose 40 per cent of the amount owed would find its way back. Then 60 per cent stays gone, and the part that stays gone is what the loss given defaultHow much of what was owed stays unrecovered once a borrower stops paying. Built up in its own right elsewhere in this material. measures. Dividing the 220 basis points by it gives the annual failure rate at which the extra Rs 22.00/- is exactly used up.

The division does not terminate. Two point two zero over zero point six zero is exactly eleven thirds of a per cent a year. The fraction is carried, and six places are printed as a display: 3.666667 per cent a year, which most people would read aloud as 3.67 per cent. Run the fraction back through the same 60 per cent and it returns 2.20 percentage points, because that is the number it was divided out of in the first place.

The bottom row of the drawing above is the whole idea in rupees, and it is worth a second look. The extra coupon is Rs 22.00/- a year. The expected lossA failure rate multiplied by the severity of a failure. Where it comes from and how it is built up is covered separately in this material. at 3.666667 per cent a year on a 60 per cent severity is also Rs 22.00/- a year. The rate was solved backwards out of the spread, so the two amounts are equal by construction rather than by coincidence. Calling it a check misunderstands what was done; calling it forced arithmetic has it right.

And now the obvious question. Is 3.666667 per cent a year a lot? A lot needs something to measure against, and only four things would serve: a counted record of how often borrowers have stopped paying, a record of what came back when they did, a run of the same spread through time, or a second borrower priced on the same day. The arithmetic above supplies none of the four. High and low are therefore not properties of the 3.666667. A note that calls it either has quietly imported a comparison it never made.

Which leaves one question that can actually be answered from what is on the table. Not whether the reading is generous. Under what circumstances is that reading the right one?

What are the honest limits on that number?

Three limits sit on 3.666667 per cent a year. The three limits are not caveats cleared once at the top of a note and then forgotten. The limits belong in the same paragraph as the figure, on every occasion the figure is written. A reader who meets the number without them has been handed a conditional reading dressed up as a measurement, and they will file it as evidence.

Limit one: the 40 per cent was declared, not measured

Where did 40 come from? Somebody typed it, and the typing is the whole of its provenance. A recovery assumptionA statement of how much of what is owed would come back after a failure. The writer of the note supplies the figure. What such an assumption is, and how one gets chosen, is covered separately. is supplied by the writer, and nothing kept here argues for one value against another. Move it and the reading moves with it, and it does not move gently.

The third column and the fourth are worth reading together. Across that whole range the reading doubles exactly, and it doubles because the severity halves from 80 per cent to 40 per cent. Meanwhile the money in the last column has not budged by one paisa, and it cannot: whatever rate the arithmetic lands on gets multiplied by the very severity that produced it, so the product is always the Rs 22.00/- the spread pays. An assumption nobody measured is doing more work on the answer than the price is.

Limit two: the whole spread was treated as payment for credit

Every rupee of that Rs 22.00/- was handed to the credit question. In a working market some of it is paid for something else entirely, most obviously the difficulty of getting out of a holding without giving a piece of it away. If part of the extra is buying that, then less of it is buying protection against failure, and the break-even computed from the whole spread is too high. Saying how much would take a traded volume, a bid, an offer or a cost of getting out, and none of those four is in the two rates the spread was built from. The direction can be said, and the direction is one way only.

Limit three: implied is a word doing serious work

An implied figure is what a price works out to under a stated assumption. No failure has been counted and no recovery measured anywhere in the arithmetic above. The 3.666667 per cent came out of a division, so what it describes is the price it was divided out of, not the borrower whose name sits above it. Drop the word implied and the sentence quietly changes species. The sentence stops being a restatement of a price and starts being a claim about how often Palash Cements Limited stops paying, and a division of one rate by another cannot support a claim of that kind.

What would have to be true for the thesis to hold?

Everything above was preparation for this part, and it is the part most notes skip. The reading goes at the top, in one line: the 220 basis points is compensation for credit and for nothing else, and at a declared recovery of 40 per cent it is exactly used up by failure arriving at 3.666667 per cent a year. Underneath it goes everything that would have to hold for that line to be the right one.

Four lines do it here. Each one is arguable on its own, and each one can fail without touching the other three. That separability is what turns a claim into something a colleague can argue with, because it gives them somewhere specific to stand.

Line three deserves a word. No earnings, no interest bill and no schedule of fixed charges has been stated for Palash Cements Limited, so there is no fixed-charge coverageEarnings measured against the fixed payments a borrower has committed to, on a base and a period both stated. How it is built and read is covered separately in this material. reading for this borrower anywhere. Line three instead shows what the condition looks like when the writer does have such a figure in front of them. The line names the base, names the period, and commits to both. A later reader can then tell whether the same figure is being compared with itself.

Look at what the four lines are not. None of them says the spread is wide. None of them says the borrower is sound. Every one of them is a statement somebody could take issue with and be specific about their objection. A note that ends with a feeling gives a reader nothing to disagree with except the writer.

What observation would break it?

Not a feeling. Not a direction. An observation, specific enough that somebody who thinks the claim is wrong could go looking for it, and written down before anybody asks for it. The breaking observation is the cheapest sentence in a note and the one most often missing. Writing it costs the writer the option of quietly moving the goalposts later.

For the thesis above, three would do it. A coverage figure, computed the way the note defined it, falling below 2.6000 times. Any evidence that a material part of the 220 basis points is being paid for something other than credit. Or the appearance of a term that lets the borrower move the timing of repayment. Somebody who disagrees can go and check each of the three, and checkability by a disagreeing reader is the only test of a breaking observation that matters.

The 2.6000 times needs the same care as line three did. The level is one the writer draws in advance, on a coverage reading the writer has and the arithmetic above does not. No measurement produced 2.6000, and a different writer would draw the line somewhere else and say so. The value is not what makes it a breaking observation. Written down in advance, and checkable by anybody who wants to look, the level qualifies whatever number the writer picked.

How is a thesis recorded now so that it can be judged later?

Four things, written at the time and never reconstructed afterwards. The claim, in the words actually used. The figures, each with its base and its period. The assumptions, each one named as an assumption. And the observation that would have broken it.

Write those four and a year later somebody can ask whether the reasoning was sound. Skip them and the only question left is whether it turned out well, a different question wearing similar clothes. A thesis recorded properly can be judged on its reasoning. One reconstructed from memory can only be judged on its outcome, and outcomes are a poor teacher. The uncomfortable part is that the unrecorded version is the more comfortable one to live with, which is exactly why it is so common.

Where does the thesis stop, and what is on the other side of it?

The reading is written, the conditions are written, and the breaking observation is written. The next sentence a reader expects is the one telling them what to do about it. The sentence a thesis actually carries there reads: a spread thesis does not state whether to take the extra Rs 22.00/- a year, or leave it, or wait.

The refusal is not delicacy. The sentence needs three facts that no analysis of a spread contains. The rest of whatever the reader already holds. The date the money is needed on, and the purpose. And the thing they would actually do on the morning the breaking observation arrives. None of those three is in a price, and none of them is knowable from here, so a sentence built on them would be built on the reader's circumstances while pretending to be built on the evidence above it.

So a thesis that stops there is not half a note but a whole one. A thesis that keeps going has quietly changed subject without telling anybody, and the reader who acts on the last sentence thinks they are acting on the arithmetic in the first one.

Can two people hold opposite readings of one spread and both be honest?

Yes, and the fact that two people can is the neatest demonstration that testable and agreed are different things. Two writers take the same 220 basis points. The first declares a recovery of 20 per cent and reads a break-even of 2.750000 per cent a year. The second declares 60 per cent and reads 5.500000. Neither has made an arithmetic slip. Neither has hidden anything.

The two writers disagree about one stated thing, and a reader can see what it is and weigh it themselves. That is what a testable disagreement looks like. Now set it beside the alternative: two writers who have each put down that the spread looks about right for the risk. The two of them may be miles apart or may agree entirely, and nobody, themselves included, has any way of telling which.

Who actually works this way, and what do they get out of it?

A lender writing a credit paper for a committee uses the break-even as the opening line rather than the closing one. The committee's job is not to admire the number; it is to test the four conditions underneath it, and a paper that arrives without them gives twelve people nothing to do except take a view on the author.

An analyst covering borrowers writes the mind-changing line for a selfish reason. Six months later, when something has moved and everybody is arguing from memory, the person with the written record is the only one who can say what they actually claimed and on what basis. Everyone else is reconstructing, and reconstruction is always kinder to the reconstructor.

A household saving into anything at all meets the same shape without the vocabulary. The chit fund that pays a bit more, the friend of a friend who will pay a bit more, the deposit at a smaller institution that pays a bit more: every one of those extras is somebody's reading of two questions. Whether the extra is generous usually cannot be answered at all. The useful habit is to write down what has to hold for the extra to make sense, and what would be worth hearing about on the day it stops holding. That is a spread thesis in household clothes, and it costs nothing except the discipline of writing it before it is needed.