Fixed Income Research: What the Analyst Actually Produces

The work produces a note, not a number. Four things go in it: a single claim, the figures under that claim, the assumptions deciding those figures, and the limits on all three. Each figure travels with its base, its period and its unit, and a reader can rebuild it from the figures printed beside it. The note ends before an action gets named.

Here is why the output cannot be the number. Anybody with the inputs and a spreadsheet can produce a number, and increasingly anybody with a browser can produce it faster than an analyst can. Nobody else can hand the reader the chain running from the inputs to a claim, together with an honest account of where that chain is thinnest. That chain is written down or there is no chain. So the note is the product, and the arithmetic is one ingredient inside it.

Everything below is worked on made-up material: a curve built for teaching, a ten year bond with no borrower behind it, a five year borrower called Palash Cements Limited that carries no assessment of any kind, and a holding of Rs 5,000 crore that belongs to nobody. The craft is the transferable part, and the figures are there so the craft has something to be applied to.

So what does the analyst actually hand over?

The end of two weeks of work looks like this. There are files everywhere, a model that finally reconciles, and a sheet of scribbles. Somebody now has to read the result in four minutes and act on nothing until they have. Somebody hands them a note.

A note carries four things, and a reader can check for all four before reading a single line of the argument. One claim, stated first and stated whole. The figures that claim rests on, each of them rebuildable from the others. The assumptions that decide those figures, named at the place they get used rather than swept into a paragraph at the end. And the limits on what the figures will bear.

The order matters more than it looks. With the claim first, the reader spends the rest of the note testing it. Testing the claim is the only useful thing that attention can do. With the claim buried under the working, the reader spends the same minutes reconstructing what the writer already knew on the first morning. Nothing is being withheld for effect; a note is not a story with an ending.

Drop the fourth item and something quieter goes wrong. The claim still stands there looking the same, and the reader has no way to tell how much weight it will take. Limits are not modesty. Limits are the instruction manual for the claim.

What must a figure carry before it can go into a note?

Four things, and each of them takes about six words to supply. Those four carriers are the cheapest quality control in the whole subject, and skipping them is the commonest defect in written analysis at every level of seniority.

A ratio is a ratio of something to something, and the denominatorThe quantity written underneath in a fraction. Swap it and the same top number reads as an entirely different result. is never obvious from the top line alone, so the base comes first. Then the period, because a rate is a rate per something and a year and a half year are two different answers to one question. Then the unit, and this one bites hardest in this subject: 2.20 percentage points and 220 basis points are a single quantity written two ways, so a note that swaps them lands a hundred times off. Then the label naming which object it is. A rate says SPOT or FORWARD. Where it says FORWARD, only two facts separate one FORWARD rate from another, so the rate names both: the stretch of time it covers, and the date that stretch opens. A duration says MACAULAY or MODIFIED, and those two differ by one plus the yield, so a note mixing them is out by that factor. Anything netted says gross or net, and says net of what.

The rebuild test in that last row is the one people argue with, so here it is at street level. A shopkeeper hands over a bill with three items on it and a total that does not come to the sum of the three. A bill like that is not a bill with a small error in it. Nobody can pay it with confidence. No way is left to tell which of the four numbers on the paper is the one to trust. A note works exactly the same way, and the currency it is spending is the reader's willingness to believe what comes next.

The specimen sentence below belongs to nobody and makes no claim about any borrower. Building a coverage figure and working its arithmetic is set out under interest coverage. The difference on display is only the one between a figure with its carriers attached and the same figure without them.

How to use Duration and Convexity in Fixed-Income Analysis

Duration and convexity were both built and derived elsewhere, and neither gets rebuilt here. Using them in written analysis is a different job, where the question stops being how the measure is constructed and becomes what a sentence containing it is allowed to claim.

MODIFIED duration answers exactly one question: how far a price moves when its own yield changes by a little. MODIFIED duration answers it as a straight line. The relationship it is standing in for is not straight, and the amount by which it bends is what convexity puts a number on. So a note using MODIFIED duration on its own has quoted an instrument that is accurate for tiny movements and wrong in a known direction for anything larger.

Take the invented bullet the earlier work in this subject built. Ten years still to run. A coupon of 8.50 per cent a year written into it. The bullet changes hands at par, Rs 1,000.00/-, and par is what pulls the yield onto 8.50 per cent a year as well. A twice yearly convention laid over the same digits prices the identical schedule at something else, so compounding runs once a year throughout. The bullet's MODIFIED duration reads 6.5613 and the curvatureThe bend in a relationship, so that equal steps taken one way do not produce equal steps taken the other. figure attached to it reads 58.4702.

Now declare a movement, purely so there is arithmetic to read. Let the yield rise by 200 basis points. MODIFIED duration on its own predicts a fall of 13.123 per cent, simply 6.5613 doubled. Reprice the ten dated amounts at the new yield and the fall comes to 12.030 per cent. Mirror the declaration, letting the yield fall by the same 200 basis points, and the straight line predicts an identical 13.123 per cent the other way while the repricing gives a gain of 14.378 per cent.

The straight line therefore overstates the loss by 1.093 points and understates the gain by 1.255 points, and it runs that way round for every ordinary bond, every time. That is not a caution, it is a direction. A rule of thumb a stall holder uses to guess the weight of a sack behaves the same way: if the guess is always a little heavy on the big sacks and always a little light on the small ones, the guess is useful, and the person who says which way it leans is more useful still. A note that prints a MODIFIED duration without saying which way its error runs has handed over a tool and kept the error bar in a drawer.

How to compare Fixed-Income Maturity Structures

Comparing two maturity structures is where the labelling discipline stops being tidiness and starts earning money. A second invented instrument sits beside the bullet: a zero coupon bond, sitting at the same 8.50 per cent yield, whose maturity was set equal to the bullet's MACAULAY duration of 7.1191 years. The construction runs on the unrounded waiting time rather than the four place figure printed here, and nothing in the argument needs a price for the zero.

A zero pays once, so its MACAULAY duration simply is its maturity. Divide that waiting time by one plus the yield and the MODIFIED duration lands on 6.5613, the bullet's own figure to the last place, by construction rather than by luck. Somebody whose only tool so far has been duration will expect these two to move together, and they will be wrong.

The bottom row and the two above it show the trap in action. The rounded figures a note would print, 12.193 less 12.030, give a subtraction of 0.163. The same pair at six places gives 0.163735, rounding to 0.164. On the other side the printed pair gives 0.216 and the honest figure is 0.215. Two subtractions, both looking self-verifying, both landing one thousandth away from the truth, and in opposite directions. That is the reason the table prints six places in its body: a reader who checks must land where the note says they will land, and the only way to arrange that is to publish the precision the check needs.

Now the caveat. A caveat belongs beside the arithmetic rather than in a footnote at the end. Nothing on this platform has put a price on the bend. Both instruments were handed the same 8.50 per cent a year. Somebody made that decision while building a teaching record, and the decision is the only reason the extra bend arrives free here. Where money actually changes hands, buyers want the bend and bid for it, so an instrument carrying more of it would ordinarily be handed a lower yield in exchange. A note may print this arithmetic. The same note may not print a preference.

Scenario vs Stress Case: what actually separates them?

Every movement declared so far came from the writer, chosen so there would be arithmetic to read under it. The point is worth saying slowly. Declared movements are the single most misread convention in this whole subject, and the misreading runs one way: a reader takes a declared movement for a forecast.

A scenario is a movement the analyst declares in order to read the arithmetic under it, sized to something a reader would recognise as ordinary. A stress case is a movement the analyst declares in order to find out what breaks, sized deliberately past ordinary. The two differ in what the writer wants out of them and not at all in how they are worked, and both of them are declared rather than observed. Neither is a forecast. Neither carries a probability. A note presenting either one as what the writer expects to happen has quietly changed its own subject.

Here is the household version. A household asks what the month looks like if the electricity bill lands at its usual size, and what the month looks like if the fridge dies in the same week the school fees fall due. The second question is not a prediction that the fridge is about to die. The second question asks which parts of the month have slack in them, and it is asked precisely because the answer is useful whether or not the fridge cooperates.

One rule falls out of that, and it is a writing rule rather than an arithmetic one. A declared movement is an input, so it gets labelled at the place it is used and not only at the place it was defined. Defined once on the first printed side of a note and used unlabelled on the fourth, it has stopped looking declared by that fourth side. An unqualified number looks like a fact about the world, and by then that is how it reads.

What can a Stress Case reach on this holding, and where does it stop?

Work it on the invented holding. Rs 5,000 crore of fixed income sits at a MODIFIED duration of 5.20, against a comparison basketA standing collection of holdings that somebody's results get measured against. whose MODIFIED duration is 4.80. The difference between those two sensitivities is 0.40.

Naming that 0.40 comes before doing anything with it. Mislabelling a difference between two sensitivities is the exact fault at issue here. It is a sensitivity, roughly a price move of 0.40 per cent for every 100 basis points. A sensitivity is not a length of time. Years are the unit MACAULAY duration carries; a MODIFIED duration reports a price response instead, and so does any difference between two of them, so a note writing 0.40 years has named a quantity nobody has ever defined.

Now declare the movement: a rise of 100 basis points applied to every rate in the structure by the same amount. The word parallel is doing exactly that work in the sentence. Two rupee answers come out of one declaration.

Thirteen times. Rs 260.00 crore and Rs 20.00 crore are both true, they answer different questions, and nothing in either figure announces which question it answered. So the base gets written into the sentence carrying the figure, without exception, or the reader receives a quantity that could be either of two things.

And now the honest half. A writer under deadline skips exactly this part. Nothing on this platform records the Rs 5,000 crore holding under any movement other than the parallel one. So no rupee figure can be given here for a structure that rises at one maturity while falling at another. The figure is not hard to obtain. The figure was never recorded. Where a reader expects it, the row below is left blank and the reason for the blank is written into the cell itself. A stated reason is a different object entirely from an estimate typed into a hole.

What makes a sentence a Liquidity Fact rather than an assumption?

Every sentence ever written about being able to sell something gets sorted by one test, and the test is whether confirming it needs the writer in the room. Two outcomes come out of it, and each one demands different writing.

The damage is done by grammar rather than by dishonesty. Somebody writes that the instrument can be sold in size on any day. Nobody lied. But that sentence is built like an observation, so it gets filed as one, and three weeks later a reader is defending a position on the strength of a sentence that was never anything more than the writer's opinion in the clothing of a measurement.

Nothing on this platform records a traded quantity, a bid and an offer, or a turnoverHow much of something changed hands over a stretch of time that has been named. figure for any instrument named above. No sentence anywhere in this guide says that any of them can be sold, and the single place a reader would look for one carries a blank row instead, with the reason set down inside it. Stating no liquidity fact at all is the honest position available here, and reaching for a plausible one instead is precisely the failure at issue.

How to analyse a Rating Action in Fixed Income

A rating action is a change in an agency's published assessment of one borrower's credit. Something appears on a screen, and a great many notes get written that afternoon which consist, once the tone is stripped away, of repeating what the agency said in slightly different words.

The work is four questions, and not one of them is what the new assessment says.

First, what changed at the borrower, in figures. Not in adjectives, in figures, and if the figures cannot be named then the honest sentence is that they are not yet known. Second, what the price had already said before the action landed. A spread is a price, and prices move ahead of publications rather than behind them. Third, what the action adds that the figures already held did not contain, often less than the length of the announcement suggests. Fourth, what the action stays silent on, and that silence is where the next work begins.

Nobody has published an assessment of Palash Cements Limited anywhere in this material, so the four questions are shown by their shape rather than on a live case. The meaning of a scale, the claim each of its steps makes, and what a publishing firm must put out when it moves one all belong to whoever publishes them. The Securities and Exchange Board of India (SEBI) keeps that wording, and sebi.gov.in is where the current text sits, with the publishing firm's own method document beside it.

What does the whole practice look like on one borrower?

Palash Cements Limited carries no assessment, and serves as the material a note gets written about rather than as the subject of one. At the five year government node this platform records 6.90 per cent a year. The debt Palash Cements Limited put out for five years carries 9.10 per cent a year. Both figures run on a once a year compounding clock. Nobody wrote the distance between them into any document at all; a reader performs that subtraction and arrives at 2.20 percentage points. Written in the unit a note more often reaches for, a hundred basis points to the point, the same distance reads 220 basis points. One quantity, two units, and a note prints whichever it means and never the other.

The bottom two rows are worth more than the four above them. Both routes are arithmetically correct; only one starts from a display figure. A number printed for a reader is an output, and feeding it back in as an input leaves a visible residue, here two thousandths of a percentage point. Naming that residue as rounding leaves the reader trusting the rest of the note more, not less.

The 3.67 per cent a year is what the price says at an assumed recovery of 40 per cent. No forecast is being made anywhere in that sentence. Nobody measured a frequency to arrive at it. The probability of this borrower failing is a separate quantity, and nothing on this platform computes it. The severityThe share of what was owed that stays missing once everything recoverable has been recovered. of 60 per cent behind it came from a recovery figure this platform simply asserts, with no evidence anywhere here standing behind it, and shifting that recovery shifts the result in step. Part of a spread may also be paying for something other than credit, and wherever that happens the implied rate reads high. Three conditions, and all three belong in the paragraph that prints the figure.

Where does the note stop?

Analysis states a claim and everything that would have to be true for it. Advocacy states a claim and everything that supports it. Read quickly, those two produce almost identical prose. They are opposite practices, and advocacy is the one that has quietly dropped the conditions that would break the claim.

The practical form of the whole distinction is one sentence long: what would change the writer's mind goes in the same paragraph as the claim, before anybody asks for it. Not in an appendix. Not when challenged. Beside the claim, where a reader who stops after one paragraph still gets it.

The same test works at a fruit stall. One seller names the fruit, names its price, and says which of it will not last past this evening. Another says it is the best fruit in the market. Both are confident, both are friendly, and which of the two a buyer is dealing with is audible inside a single sentence. The difference is not honesty and it is not tone. The difference is whether the seller volunteered what works against them.

And this is where the note ends. Everything above sets out what a claim rests on, how far the figures will stretch, and where they stop. The next sentence, the one naming an action to take, is not written here and is not written anywhere on this platform. Anybody who arrived for that instruction has come to the wrong shelf: a note that spends itself on keeping a claim conditional and then names an action has undone the whole of its own argument in one closing line.

Who reads a note, and what do they do with it first?

A credit officer at a lender already has money out to the borrower and cannot unlend it this afternoon. A credit officer forms a view of their own on the claim, so the claim is not what they want from a note. The assumptions section is. Their exposure is already fixed, and the only live question is which of the writer's assumptions, if it moved, would move the claim. A note organised so that answer takes ten minutes to assemble gets read once.

An analyst on the other side reads it as a checkable object. The second reader will pick two figures at random and try to rebuild them from the rest of the note, and what they learn in ninety seconds is not whether the writer is right but whether the writer is careful. Both of the figures they pick will have a base, a period, a unit and a label, or they will not, and there is no third outcome.

Somebody six months from now, who has inherited the file and never met the writer, is the third reader and the one most often forgotten. Everything conditional left unwritten is invisible to them, and every declared movement labelled only once, at its first appearance, has by then turned into a fact about the world in their reading of it. Writing for that person is the entire discipline, and it costs about six words a figure.

Smaller scale, same habit. Somebody comparing two savings products reads that one pays 7 per cent and the other 3.5 per cent, and the second is quoted for six months while the first is quoted for a year. The person who asks per what before comparing has just performed the whole of this guide.