How to write an equity research note, step by step

A research note is written in a fixed order: the claim first, then what the claim rests on, then the evidence, then the case against it, then the assumptions, then the conditions that would change it. The claim is composed last and placed first. A note opening with company background has spent the only attention it can count on for the least useful thing in it.

The claim-last, claim-first order is not a house style and it is not a matter of taste. Almost nobody reads a note the way it was written, and the order follows from that single fact about how notes are consumed. A note is skimmed, and a skim goes first paragraph, then the headings, then the titles on the exhibits, and then out. Whatever sits in those places is the note, as far as most of its readers are concerned. Everything else is available to the small number of people who go looking.

So the writing order is built backwards from that. The work gets done in whatever sequence the work demands. Then the analyst decides what the work actually shows, compresses it to one sentence, and puts that sentence where the attention already is. Every step below is a decision about placement and evidence, and not one of them is a decision about analysis. The analysis happened before any of this. Making the analysis reachable is a separate craft, and it gets skipped more often than any other part of the work.

In what order is a research note actually written?

Two orders run against each other, and that is the first thing to hold on to. There is the order in which the parts are produced, and there is the order in which they are laid out in the note. The evidence is gathered first and appears third. The claim is composed last and appears first. The claim feels unearned when it is stated before the reader has seen anything that supports it, so nobody finds the arrangement comfortable the first few times.

The same behaviour shows up everywhere else. A friend who is going to ask for something puts the ask at the end of a long message, and the recipient reads the first two lines and then hunts for the ask. A caterer quoting for a wedding puts the total on the third sheet under the menus, and every person receiving that quote turns to the third sheet first. A test report that buries its finding under a paragraph of method gets read from the bottom up. People reading anything they did not choose to read go straight to the point, and if the writer has not put it anywhere findable they invent one from whatever they saw first.

The invented claim is the trap, and it is worth sitting with. A reader who cannot find the claim does not conclude that there is no claim. The reader concludes that the claim is whatever the opening paragraph was about. A note that opens with the founding of a company and the range of products it makes has told its reader, without meaning to, that this is a note about a solid old company with a broad range. The ninth sheet was never reached, so the impression survives whatever the ninth sheet says.

The seven steps in full. Each is an instruction about where things go and how they are recorded, and that is all any of them is. Analysing anything is a separate subject, and none of the seven touches it.

1. Write the claim, in one sentence, and write it lastThe claim says what the analyst expects to happen next, and why. It is composed after the work is done, because until the work is done the claim is not yet known. A claim that needs more than one sentence is usually two claims, and a note carrying two claims is read as carrying none.
   Checking: can it be said out loud in one breath, and does it contain a because?
2. Place the claim first, above everything elseThis is answer first, then support, which belongs to Barbara Minto, The Pyramid Principle, 1978, and the attribution travels with the rule. The first paragraph is the only part of the note that can be counted on being read, so it holds the conclusion and nothing else.
   Checking: if a reader stopped after paragraph one, would they have the view and its reason?
3. State what the claim rests on, separately from the evidence for itThat is the chain of cause and effect which would have to hold for the claim to be right. It is a mechanism, written in words, with no figures in it. It gets its own section because it is a different kind of thing from the numbers underneath it.
   Checking: could somebody disagree with this section without disputing a single figure?
4. Lay out the evidence so a reader can check itEvery figure carries its source and the period it belongs to, beside the number rather than in a footnote at the back. A figure that came out of the analyst's own model is labelled an estimate, in the same line, in the same size type as everything else.
   Checking: could a reader verify any figure here without leaving the line it sits on?
5. Write the case against the view, in its own sectionName the two or three things that would most damage the claim, say honestly how strong each looks, and say what evidence would settle it. Write it properly rather than as a token paragraph that concedes nothing.
   Checking: would somebody who disagreed recognise their own argument in this section?
6. Record the assumptions in a list, with a value and a source for eachA list, not a paragraph. Each row carries what was assumed, the number, and where the number came from, including the rows where the source is the analyst.
   Checking: is every number in the working reachable from a row in this list?
7. Write down what would change the view, in observable termsNot a worsening, not a deterioration, not a loss of confidence. A specific published event: a margin below a stated level, a decline over a stated number of quarters, a change in what the company discloses.
   Checking: could two people who disagree about everything else still agree on whether this has happened?

Why does a seven step order produce a six section note?

Because two of the steps act on the same thing. Step one composes the claim and step two decides where it sits, and between them they produce one section of the finished note: the opening. The remaining five steps produce one section each. So the writing order has seven steps, the note has six sections, and the two counts are both right about different things.

The split between step one and step two is worth holding on to rather than letting the two counts blur. Composing a claim and placing a claim are different acts, they fail in different ways, and a writer can be good at one and hopeless at the other. A writer who cannot compress a view into one sentence has a step one problem, and no amount of rearranging will fix it. A writer who has a perfectly sharp one sentence view and puts it on the ninth sheet has a step two problem, and a step two problem is fixed by moving a paragraph.

What goes in the first paragraph, and what must never go there?

The claim goes there, with its reason attached, and nothing else does. Context must never go there: the founding of the company, the range of what it makes, the size of the sector, the recent movement in the share price. Every one of those is a legitimate thing to put in a note. The first paragraph must carry what the note is telling the reader, and none of those four is that.

The rule has a name and the name is part of it. Answer first, then support, belongs to Barbara Minto, The Pyramid Principle, 1978, and it says that business writing puts its conclusion at the top and arranges everything beneath it as support for that conclusion. It is thirty years older than most of the notes written under it and it is still the single highest return change a research writer can make to their own output.

A resistance left unnamed keeps winning, so the resistance to answer first is worth naming. Stating a conclusion before the evidence feels like asserting. Stating it feels like skipping the part where the work is demonstrated. Writers who have spent three weeks on an analysis want the reader to walk the same three weeks, and putting the answer on line one feels like throwing that away. It is not. The work is all still there, one paragraph down, in a form a reader who cares can check. The reader who does not have three weeks now has the answer instead of a first paragraph about a paint plant.

Consider what the alternatives actually deliver. Take one company, Sarvani Coatings Limited, an invented paint maker, and three openings written for it. The first opens with history: when the company was set up, where its plants are, what it makes. The second opens with the market: the share price, the earnings multipleThe share price divided by the earnings attached to one share. The multiple is a ratio between two numbers, so on its own it says nothing about whether either number is right. of 42.0 times, the market value, the free floatWhatever slice of the share count sits outside promoter hands and can therefore be bought and sold on an exchange. Size and slice move independently.. The third opens with the claim. All three are competently written. Only one of them is a note.

What is the claim itself supposed to say?

A claim says what the analyst expects to happen next, and why. Three parts: a subject, a direction, and a reason. Strip any one of them out and what is left is not a claim, whatever it looks like in the note.

Here is a sentence that fails the test and fails it in the most common way. Sarvani Coatings has improving margins and a strong balance sheet. Read it slowly. The sentence has a subject. Improving describes what already happened rather than stating what happens next, so there is no direction. There is no reason at all. And two separate assertions sit joined by the word and, so a reader who accepts one and rejects the other has no idea what to do with the note. A sentence that lists good things about a company is a description, and a description cannot be wrong in any way that matters. Being impossible to get wrong is exactly why a description is comfortable to write.

Now a sentence that passes. Sarvani Coatings Limited's gross marginWhat is left of a rupee of sales after the direct cost of making the thing, and before wages, rent, interest and tax come out of it. gain is a level shift rather than a durable rate, so it will not repeat in year four. One subject, one direction, one reason, one sentence. The sentence can be wrong. Somebody can read it and say no, and know exactly what they are saying no to. The possibility of a flat no is the whole test.

Why are what a claim rests on and the evidence for it two different sections?

Because they are two different kinds of statement and they can fail independently. A claim rests on a mechanism: a chain of cause and effect written in words, with no figures in it, describing what would have to be true of the world for the claim to hold. The evidence is the set of measured things that support a belief that the chain is holding. A note can have a sound mechanism and thin evidence. A note can also have abundant evidence attached to a mechanism that does not follow. A note that puts both in the same paragraph makes those two failures indistinguishable.

The everyday version: a mechanic says the car will stall again, because the fuel pump is weak and a weak pump starves the engine under load. The mechanism is the second half of that sentence. The evidence is the pressure reading he took. Given only the reading, the owner has a number and no reason to care. Given only the reason, the owner has a story with nothing under it. Both are needed, and which of the two is in doubt has to be visible.

Watch what happens when they get collapsed. Take the specimen claim and give it a reason drawn from the numbers: the margin gain is a level shift, because gross margin rose from 43.0 per cent in year one to 46.0 per cent in year three. The sentence reads like an argument and it is not one. In this company's profit ladder, gross profit is revenue less the cost of materialsWhat a maker paid for the physical inputs it consumed during the year. For a paint maker that is mostly pigments, resins and additives. and nothing else, so the three point rise in gross margin and the three point fall in the materials share of revenue, from 57.0 per cent to 54.0 per cent, are one event described twice. The sentence gives as a reason the very thing it is supposed to be explaining.

The actual mechanism has to say something the numbers do not say by themselves. A materials share of revenue can fall for two opposite reasons: because input cost came down, or because realisationThe average price actually received for a unit sold, after discounts and after the mix of what was sold. Realisation moves when prices move and it also moves when the mix changes. rose faster than input cost did. The two reasons have opposite futures, and the published statements separate neither. The sentence about realisation and input cost is the mechanism: it contains no figures, and somebody can argue with it without disputing a single number in the note.

How is the evidence laid out so a reader can check it?

By putting the source and the period beside the number, on the same line, in the same size type. Not in a footnote. Not in an appendix. Not in a source list at the back that a reader has to hold a finger in while they read forward. The reader is not being asked to trust the writer, and a note built to be trusted rather than checked has already failed at the thing a note is for.

Three things sit beside every figure. The value. The period it belongs to. For an Indian company that means the year or quarter ended on a stated date. And where it was published: the statement or the filing it came off, rather than the name of the company. Where a figure did not come off a published statement at all, a fourth thing is needed: the word estimate, in the line, saying that this number came out of a model rather than out of a filing.

The fourth case is the one writers fudge, and the fudge is almost never deliberate. A number gets computed in a spreadsheet, gets carried into the draft, and by the third draft it is sitting in a table beside four published figures wearing exactly the same typeface. Nothing in the note tells the reader that one of those five rows is a guess. An estimate that has lost its label is indistinguishable from a fact while being nothing more than the writer's own opinion, and that makes it the most dangerous single item in a research note.

A figure without a period attached is not merely incomplete. The entire question anybody has about Sarvani Coatings is which year a margin belongs to and which direction it moved, so a figure with no period attached is unusable. Gross margin of 46.0 per cent says nothing on its own. Gross margin of 46.0 per cent in the year ended 31 March of year three, against 44.0 per cent the year before and 43.0 per cent the year before that, is the whole subject of the note.

Where does the case against the analyst's own view go?

Into its own section, with its own heading, written properly. Not a sentence at the end conceding that risks exist. Not a bullet saying that the outlook could change. A real section that names the two or three things that would most damage the claim, says how strong each of them looks, and says what evidence would settle each one.

The reason is structural rather than moral. Every other section of a note supports the claim. The evidence supports it, the mechanism supports it, the assumptions are the conditions under which it holds. If nothing in the note tests the claim, then the note contains no test, and a document containing no test of its own central assertion cannot be argued with. The counter section is not a courtesy to the reader and it is not modesty; it is the only part of the note that could show the claim to be wrong, and without it the note is a statement rather than an argument.

There is a second reason, and it is the one that decides whether the section is any good. Only the writer knows which figure had to be reached for, which assumption felt uneasy at eleven at night, and which alternative explanation was never chased. The writer is therefore the person best placed to find the weakest point in the work. A stranger writing the counter case has to guess at all three. If the writer writes it and it comes out thin, that is worth noticing on its own.

Here is what a real counter section looks like against the specimen claim, using the same published figures the note already cites. First, the claim asserts a mechanism the statements do not show: they show the materials share of revenue falling from 57.0 per cent to 54.0 per cent, and they do not show why, and if the reason is the company's own pricing then the gain is a rate rather than a level shift and the claim is simply wrong. Second, two comparable makers gained too. Across the same two years the gain came to 2.4 points at Nandivarman Paints Limited and 3.6 at Kesaria Surface Solutions Limited, against 3.0 at Sarvani Coatings, so the move ran across the whole field, and a condition that has held for two years may hold for a third. Third, a level shift is still a level: nothing in the claim says 46.0 per cent falls, so the note can be right about the mechanism and wrong about what a reader would do with it.

Two of those three attack the mechanism rather than the company. Attacking the mechanism is what a counter case has to do. Arguing that Sarvani Coatings is smaller than Nandivarman Paints, or that its EBITDA marginEarnings before interest, tax, depreciation and amortisation (EBITDA), taken as a share of revenue. A rough operating figure that ignores how the assets were paid for. of 18.47 per cent looks high, attacks nothing at all, because the claim never said anything about size and a margin looking high to somebody is not a fact about the world.

How are the assumptions recorded?

In a list, with a value and a source on every row, including the rows where the source is the writer. Not in a paragraph, not woven into the argument, not implied by a number appearing in a table. A list, with columns.

The reason is almost embarrassingly simple. An assumption inside a sentence disappears, and an assumption in a list can be attacked, and being attackable is the entire point of writing it down. A reader skimming a paragraph will not notice the clause that says assuming volume growth holds. A reader looking at a five row table with a column headed source will notice immediately that three of the five rows say estimate, and that is the useful thing they could possibly notice about the note.

Think about a household deciding whether it can afford a larger rent. The reasoning happens out loud over dinner and it contains four assumptions: that the second income continues, that the school fee does not rise this year, that the car lasts another two years, that nobody falls ill. Said out loud in a flowing conversation, none of the four is examined. Written down as four lines on the back of an envelope, the second income line gets looked at hard. Seeing it alone on a line is what makes it look like something that could not happen. Nothing changed except the format.

Each row needs three columns and there is a fourth that is worth having. The assumption itself. The value. Where the value came from, and that third column is the honest one: disclosed by the company, computed from a published statement, or the writer's own estimate. And, where the note supports it, what happens to the claim if that row is wrong. A consensus estimateThe average of the forecasts several analysts have put out on one company. Averaging a set of opinions does not turn them into a measurement. put into an assumption row is somebody else's estimate, and the row should say so rather than dressing it as data.

What would change the view, and how is that written down?

As a published event with a date attached. The test is not whether the condition sounds serious. The test is whether two people who disagree about everything else in the note could still look at the same filing and agree on whether the condition has happened.

Run that test on a phrase everybody has written at some point: the view would change if the outlook deteriorates. When did that happen? On what day? Which document records it? There is no answer. Deterioration is not an event but a description somebody applies afterwards. A condition nobody can observe cannot be tested, and a view that no possible event would overturn is not a view at all, however carefully it is argued.

Now the same condition written properly: gross margin below a stated level for two consecutive quarters. The condition has a date. The margin sits on a published quarterly statement. Anybody can look it up, including somebody who thinks the claim was nonsense from the start, and both will get the same answer. Observability is the whole requirement, and it is a low bar that most change conditions in most notes do not clear.

Three conditions is usually enough and they should include at least one that would prove the claim right as well as ones that would prove it wrong. Against the specimen claim: gross margin holding at 46.0 per cent across four consecutive quarters while input cost per unit rises would break the claim. A repeat of the 0.13 percentage point share gain would break it too. Volume growth falling below the 6.0 per cent recorded in year three would support it. All three are things that get published on a date.

When is the note finished?

When five conditions are met, and not before, and not after. The claim is one sentence. Every figure carries a source and a period. The case against is written. The assumptions are listed with values and sources. The change conditions are observable. The five together are the stopping rule, and the rule says nothing about length.

Length is the criterion writers use by default, and it is the wrong one in both directions. A note can be four hundred words and finished. A note can be four thousand words and unfinished, carrying three thousand words of context and no change conditions anywhere. Length measures how much was written, and the stopping rule measures whether the reader can check the work, and those two things are not related.

There is a practical version of the rule that is easier to apply at midnight than the list is: what would a reader who disagreed need in order to argue back properly? A disagreeing reader needs to know what the note claims, why it claims it, what it is looking at, what has already been thought against it, what was assumed, and what would change the view. If any of those six is missing, they cannot argue, and the note is not finished. If all six are there, a seventh sheet of background makes the note longer and nothing else.

What must never appear in a note built this way?

A target for the share price. A recommendation to buy or to sell. A rating. A statement that a share is cheap, dear, attractive or overvalued. None of those follows from the six sections above, and the reason is a boundary rather than a caution: every step in the order produces something a reader can check, and a rating is not something a reader can check.

More survives that boundary than people expect. Notice what is left. The claim survives, a statement about a mechanism that can be argued with. The evidence and its sources survive. So do the counter case, the assumptions and the change conditions. Everything that makes a note checkable survives the boundary. Converting a checkable argument into an instruction to somebody about their own money is the one step that does not survive, and it needs knowledge of who the reader is, what else they hold, and what they can afford to be wrong about, none of which a note contains.

A note written in a professional setting and sent to other people carries a further set of obligations that have nothing to do with craft. Who paid for the work. The holdings of the writer and the writer's employer in the company. The other business the employer does with the company. When the note was first sent and to whom. All four are conduct disclosureThe statements a person distributing research has to attach to it: who paid for the work, what they hold, and what else they do for the company. Set by the regulator rather than by the writer. requirements, set by the regulator rather than by the writer, and they change.

What do the six sections look like written out on one company?

Below is the skeleton of a note on Sarvani Coatings Limited, with all six sections written out so there is something concrete to take apart.

1. The claim

2. What the claim rests on

Gross profit in this company is revenue less the cost of materials and nothing else, so the three point rise in gross margin between year one and year three and the three point fall in the materials share of revenue, from 57.0 per cent to 54.0 per cent, are the same event described twice. The materials share can fall for two opposite reasons: input cost coming down, or realisation rising faster than input cost. The two have opposite futures. A fall driven by input cost reverses when input prices reverse. A rise in realisation that outruns input cost is a rate the company controls, and a rate can continue.

The one year in this record that can be decomposed per unit is year two to year three, where the materials share fell two points, from 56.0 per cent to 54.0 per cent. On volume up 6.0 per cent, revenue per unit rose about 7.5 per cent and materials cost per unit of output rose about 3.6 per cent. Input cost per unit did not fall. Year one volume is not in the record, so whether the same holds across the earlier year cannot be worked out from what is published. The two year headline and the per unit figures are kept apart for that reason rather than paired.

3. The evidence

Read one row across before reading the table down. Reading across is the only way the evidence layout earns its space. Revenue ran Rs 1,840 crore, Rs 2,120 crore and Rs 2,415 crore across the three years ended 31 March, and every one of those three came off an audited statement. Gross profit ran Rs 792 crore, Rs 933 crore and Rs 1,111 crore, and not one of those three came off a statement: each is revenue less the cost of materials, worked out by whoever wrote the note. Three of the nine rows in that table were reported, five were computed and one is an estimate, and a reader who did not look at the last column would take all nine as published fact.

Look at the last row and at the way it is dressed. The row sits in the same table as eight figures a reader would take as published, and the only mark separating it from them is the word estimate and a shaded cell. The row is the writer's own work, and the note says so on the row itself, where a reader will see it.

4. The case against the claim

• The mechanism is asserted rather than shown. The statements show the materials share of revenue falling. They do not show why it fell. If the reason is Sarvani Coatings' own pricing rather than a condition it was handed, the gain is a rate and the claim is wrong at its root.
• The move ran across the whole field. Measured over those two years the three gains were 3.0 points at Sarvani Coatings, 2.4 at Nandivarman Paints Limited and 3.6 at Kesaria Surface Solutions Limited. Sarvani Coatings sits between the other two, so nothing in that comparison separates its own pricing from anything else, and a condition that has held two years may hold a third.
• A level shift is still a level. Nothing in the claim says gross margin falls back. If 46.0 per cent simply holds, the note has been right about the mechanism and wrong about anything a reader would do with it.

5. The assumptions

Volume up 6.0 per cent and realisation up 3.0 per cent compound rather than add, so on those first two rows together revenue growth works out at about 9.2 per cent. Three of the five rows say estimate. A reader who wanted to attack this note would start there, and the format is what makes that easy.

6. What would change the view

• Gross margin at or above 46.0 per cent across four consecutive quarters while input cost per unit rises. That would break the claim, because a level shift that keeps holding against a rising input cost is behaving like a rate.
• A repeat of the 0.13 percentage point share gain, or better, in the year that follows. That would break it too, for the same reason.
• Volume growth falling below the 6.0 per cent recorded in year three. That would support the claim rather than break it, and it is listed for that reason.

All three were written down before any of them could be observed, and being written first is what makes them worth anything. A condition added after the fact is a description of what happened, not a test of what was thought.