Equity Research: What the Job Actually Involves

Equity research is the work of forming a defensible view on a listed company's future profits, and on what the market price is already assuming about them. The output is a written argument with the evidence attached and the assumptions named, not a number. Most of the working day is reading disclosures, building and correcting a model, and narrowing the case down to the two or three questions the answer actually turns on.

Much of that paragraph is about the second thing rather than the first. Almost everybody arrives at this subject holding the first half only: that an analyst studies a company and works out what it will earn. The forecasting half is real work and genuinely difficult. A share price is not a blank sheet waiting for a forecast, so on its own that half settles nothing. The price already contains somebody's forecast. The whole of equity research is the comparison between what the evidence supports and the growth the price already carries, and an account that teaches only the forecasting half has taught the arithmetic and skipped the question.

Equity research sits on top of the profit ladder, the balance sheet, the cash flow statement, the multiple and the discount rate, and on top of industry structure and competitive position, each of them built elsewhere. The shape of the work that uses all of them runs: how research starts, what a working week contains, how the written case is put together, how a claim about several companies at once is tested, and where the line runs that this work is not allowed to cross.

What is an equity research analyst actually trying to find out?

The everyday version removes the mystery in about thirty seconds. A house on a familiar street is up for sale at a stated asking price. A week could be spent working out the value of that house: the plot, the age of the building, what the roof will cost in three years, how loud the road gets at eight in the morning. Working the house out like that is real work. But the asking price already contains somebody's answer to all of it. So the only question that can change a buyer's decision is not the value of the house in the abstract. The question is whether the answer already inside the asking price is one the evidence supports.

A listed share works the same way and is harder only because the assumptions are less visible. Sarvani Coatings Limited, an invented maker of decorative paints and industrial coatings running as the case throughout, trades at an illustrative Rs 486/- a share on a stated date of 25 August 2026. The quoted price is not a valuation anybody published. The price is the level at which the last buyer and the last seller disagreed by exactly nothing, and inside it sits a set of assumptions about how fast profits grow, for how long, and at what margin. Nobody wrote those assumptions down. Extracting them is therefore work rather than lookup.

So the job has two halves that must not be collapsed into one. The first half is evidentiary: what do the published statements, the disclosures, the segment splits and the industry's own behaviour support as a view about future profits? The second half is inferential and runs backwards: given the price, what must somebody already believe? Research is the comparison of those two, and the finding is always a difference rather than a level.

Extracting them is why the work reads as slow from outside. An analyst who has spent three days on Sarvani Coatings has not spent three days predicting a profit figure. The second half of the job is worthless if the first half is soft, so most of those three days has gone on establishing what the published record can and cannot support. A confident inference about what the price assumes, set against a shaky reading of what the evidence supports, produces a confident difference between one guess and another.

What does a week of the work actually look like?

Newcomers picture the job as modelling with some reading around it. The proportion is closer to the reverse, and the reason is worth understanding rather than just noting. A model is a machine for turning assumptions into consequences. A model is fast, once the assumptions that matter are known. Finding out which assumptions matter is reading, listening and asking, and there is no shortcut through it. The question a company actually turns on is almost never the question its own presentation is built around.

The drawing below sets an illustrative fifty hour week against the shape a newcomer expects. Neither bar measures anybody's actual week. The proportion is what the two bars carry, and the proportion is the part that surprises people. The model is the shortest block in the honest version and the longest block in the imagined one, and that single inversion explains most of what goes wrong in a first year of doing this work.

Take each block in turn. Each does something different. The twenty hours of reading are annual reports, quarterly statements, the notes rather than the face, and the transcript of the results callThe recorded question and answer session a listed company holds after publishing a quarter's or a year's figures. A written transcript is usually put out afterwards, which is what makes it something to read rather than something to attend.. The face of a statement is a summary, and the notes are where the summary was made, so the notes are where the interesting material sits. The twelve hours of writing are not a final flourish. Writing is where a paragraph that seemed believable turns out not to survive being written down in order, and analysts who write late discover their own gaps late.

The six hours of conversations are the part outsiders find surprising and insiders find indispensable. A distributor, a supplier one step up the chain, a plant engineer, somebody who left the industry two years ago: none of them supplies a number that can be printed, and all of them tell whether the mechanism written down is the mechanism that operates. The value in this job is created almost entirely in deciding which questions matter, and almost none of it in the arithmetic that follows once the questions are settled.

How Public Equity Research Works: who makes it, and who reads it?

The analytical work described so far is the same wherever it is done. The destination differs sharply, and so does what attaches to the research on the way. Three settings cover almost all of it, and holding them apart saves a great deal of confusion later.

Research produced inside a broking house is written for people outside it, and a broking house is the sell sideThe part of the market that produces and distributes research, executes trades and arranges deals for other people, rather than managing money itself. The name comes from selling services to investing institutions., and its output is distributed: it goes to institutions that allocate money, it carries the writer's name, and because it leaves the building it carries conduct and disclosure requirements with it. Research produced inside an asset manager is written for the people in the next room, and an asset manager is the buy sideThe institutions that manage money and take positions with it, such as mutual funds, insurers and pension managers. They consume research rather than distribute it, and much of what they produce never leaves the building., and its output is usually never distributed at all. Research produced by an independent analyst working for themselves is distributed like the first and registered in its own right.

The same analytical work carries different obligations depending on where it sits, and distribution rather than analysis is what switches those obligations on. The split is not a technicality but the entire logic of how this activity is supervised: forming a view is thinking, and thinking is not a regulated act. Handing that view to somebody who may act on it is a service being provided to another person, and that is what a regulator takes an interest in.

How to start a public equity research process: what comes first?

There is a correct starting order and it is unintuitive, because the most impressive-looking step comes last. Here is the sequence, and then the reason it runs this way.

First, establish what the company sells, to whom, and how it gets paid. Sarvani Coatings sells decorative paints to households through dealers, and industrial coatings to manufacturers on contract. Decorative and industrial are two different businesses with two different cash rhythms sitting inside one profit ladder, and every later step differs depending on which one is under examination. A sheet of notes on this costs an hour and pays for itself repeatedly.

Second, read the last three annual reports oldest first. Reading the newest one first is the natural instinct and the wrong one. The newest report is written to explain the present. Reading them in the order they were written shows the direction of travel: what management said would happen, what then happened, and what they said about it afterwards. Reading in order is the single cheapest source of evidence about how seriously to take a stated intention.

Third, build the revenue line before anything else. Not the whole model. The revenue line, split as far as the disclosure allows: by segment, by volume against price where the company gives it, by geography if that is how it reports. Revenue is the only line that is not a residual of something else, and almost every serious mistake further down is a revenue mistake wearing a costume.

Fourth, write down the two or three things that would have to be true for the current price to make sense. Not a valuation. A written list of conditions, in sentences. For Sarvani Coatings that list is short: a gross marginRevenue less the direct cost of what was sold, expressed as a share of revenue. For a maker of paints the direct cost is dominated by pigments and resins, so the gross margin moves with input prices and with what the company can charge. that holds near 46.0 per cent, volume growth that stays above what the sector itself is doing, and the share gain of the last year repeating rather than being a one-off.

Fifth, and only now, open a model. By this point the model has three or four questions to answer instead of forty, and every cell in it exists because a question required it. A model built in this order is small, and a small model that answers the right question beats a large one that answers whatever it was designed to answer.

Why does starting with the model waste a week so reliably? Because a model answers whatever it was built to answer, and a model built before the questions are known was built to answer the generic ones. The model produces an output. The output looks precise, and precision is persuasive in a way that has nothing to do with being right. A model built before the questions are known answers questions nobody asked, and starting with the model is the commonest way to lose a week.

Which two questions is every piece of research separating?

This is the intellectual centre of the subject, and it repays slowing down. Every piece of equity research is trying to separate two questions that ordinary conversation runs together. The first is what will this company earn. The second is what is already assumed. The two feel like one question and behave like two, and almost every confusing thing about how share prices respond to news comes from the two being run together.

The household version. A friend's daughter has scored 88 out of 100 in an examination. Is that good news? The question cannot be answered, and the reason is not a lack of information about the daughter. Nothing has been said about what everyone expected. If the school and the household had been quietly bracing for 70, it is a very good result. If the last four papers were all above 95, the same 88 is a difficult evening. The number carries no news on its own, and only the distance between the number and the expectation does.

Now the reveal, with Sarvani Coatings' invented figures. Nine estimates existed for the year, with a mean profit after tax of Rs 268 crore and a range running from Rs 255 crore to Rs 284 crore. The company reported Rs 278 crore. Against the mean that is a beat of 3.7 per cent, and it is also a figure that sits comfortably inside a range that already existed before anybody opened the results. The mean consensus estimateThe middle of the published forecasts put out by whoever follows a company. It summarises what a group of people currently expect, and it shifts whenever any of them changes their mind. for earnings was Rs 11.17/- a share against an actual Rs 11.58/-.

A result landing inside the range that already existed is not a surprise in any useful sense, however it is reported the next morning. The range was the market's own statement that it did not know the answer to within Rs 29 crore. The price at Rs 486/- was never carrying a number for one year's profit, so landing at Rs 278 crore inside that band resolved the first question and left the second exactly where it was. A price carries a rate of growth sustained over several years, and one year's result is a small piece of evidence about a claim of that shape.

The symmetric case looks unfair from outside, and it is the one worth holding on to. A company can miss every published estimate and its shares can rise, if the price had already been carrying something worse than the miss. Neither of those movements is irrational, and neither is evidence that the market is a mood. Both are what happens when a result is read against something that nobody wrote down.

Where did Sarvani Coatings' margin actually come from?

Meghna Iyer, an invented analyst, now makes a first pass on Sarvani Coatings Limited using only what is published. She is not forming a view today. She is finding the question. Finding the question is the correct work at this stage rather than a warm-up for the real work.

She reads the three years oldest first and writes down the ladder.

Three things jump off that table. Revenue grew 13.9 per cent in the latest year. Earnings before interest, tax, depreciation and amortisation (EBITDA) grew 31.2 per cent. Profit after tax grew 41.1 per cent. Each line down the ladder grew faster than the line above it, and that widening is the whole of the reading problem this case poses. Something is happening between revenue and profit, and finding out what is the first honest piece of work available.

Start where the money is. Between year two and year three the EBITDA margin rose from 16.04 per cent of revenue to 18.47 per cent, a gain of 2.43 percentage points. Decompose that gain across the three cost lines, each expressed as a share of revenue in both years, and it splits as follows: gross margin contributed 1.99 points, employee cost contributed 0.29 points by falling from 8.77 per cent of revenue to 8.49 per cent, and other expenses contributed 0.15 points by falling from 19.20 per cent to 19.05 per cent. The three contributions sum to 2.43 exactly.

Notice the claim the decomposition cannot support. Materials do not account for the whole difference: operating leverageThe effect of a cost that does not rise in step with revenue. When revenue grows and a cost line grows more slowly, that cost falls as a share of revenue and the margin widens without anything about the cost itself having improved. on the employee and other expense lines supplied 0.44 points of the 2.43, or about one fifth. Saying materials explains all of it would be a small overstatement that costs nothing today and costs a great deal in a year, when somebody checks. Getting the size of an effect right matters as much as getting its direction right, and the habit of saying entirely when about four fifths is meant is how a careful reader stops trusting an analyst.

So the question narrows to the gross margin, and here it gets genuinely hard. Over the two years from year one to year three the gross margin rose 3.0 points, from 43.0 per cent to 46.0 per cent. The record does not carry year one's volume growth, so the two year move cannot be decomposed per unit. The one year move can be. Between year two and year three, volume rose 6.0 per cent while revenue rose 13.9 per cent, so realisationRevenue divided by the units sold, meaning the average price actually achieved after discounts and mix. It rises when prices go up and also when the sales mix moves towards more expensive products. rose about 7.5 per cent. The cost of materials rose from Rs 1,187 crore to Rs 1,304 crore while output rose that same 6.0 per cent, so what each unit of output swallowed in materials ended the year about 3.6 per cent higher.

Nothing got cheaper. Sarvani Coatings paid more for the pigments and resins in each tin, and realisation outran that cost, so the margin widened anyway. The finding is the opposite of the intuitive story. Check the arithmetic yourself: 56.0 per cent multiplied by 1.0364 and divided by 1.0747 gives 54.0 per cent, exactly where the materials line landed.

Which leaves the question Meghna Iyer actually writes down at the end of her first pass. Why did realisation outrun input cost by nearly four percentage points? Three explanations fit, and the published statements separate none of them.

So Meghna Iyer's first pass ends with a list of what would separate the three, rather than with a view. She writes down three items. The split of revenue growth into volume and realisation, given by the company and already used above. The segment split of revenue between decorative and industrial, testing the mix explanation directly: a mix shift has to appear as industrial revenue growing faster than decorative. And what management said about pricing on the results call, evidence about intention rather than about outcome, to be weighed as such.

She has not opened a model and she has not formed a view, and at the end of a first pass that is exactly correct. She has a question sharp enough that the next three days can be spent on it rather than on everything. The difference between a productive week and a wasted one is almost entirely whether that sentence got written down on day one or on day four.

At its default setting the drawing reproduces the published position exactly. Sarvani Coatings' profit after tax of Rs 278 crore grown at 9.2 per cent a year for five years reaches Rs 432 crore. The same figure grown at 24.2 per cent a year reaches Rs 822 crore. The difference is Rs 390 crore, and the price path is 1.90 times the evidence path. Where does 24.2 per cent come from? The 24.2 per cent comes from running a reader's own two assumptions backwards: a required 12 per cent a year from the shares, and a valuation of 25 times earnings in five years. Both of those belong to whoever sets them. The output of that arithmetic is a written assumption that can be tested, and a written assumption is never a target.

How to write a public equity research case study: what are its parts?

A case study is where the reading turns into something somebody else can examine. Six parts, and the order matters because the claim goes first. Putting the conclusion at the top rather than building up to it is Barbara Minto's answer first rule, set out in The Pyramid Principle in 1978, and it is the right structure here for a specific reason: a reader who disagrees with the claim needs to know it on line one, so the rest can be read looking for the weak point.

Take the parts one at a time. The claim is a sentence, not a paragraph, and it has to be the kind of sentence that could turn out to be false. Written for Sarvani Coatings it might read: the margin gain is a one-off step up rather than a rate that keeps repeating, so whatever earnings growth the price appears to assume will not arrive from margin. A claim written that way exists to be taken apart, and a claim nobody can take apart is not a claim.

The next part, what the claim rests on, makes the shortest section in a note and the most revealing one. For that sentence the list runs to three items: whether the gross margin holds at 46.0 per cent, whether volume growth holds above the sector's 4.5 per cent, and whether the 0.13 percentage point share gain repeats. Everything else in a model is detail. Nobody including its author can keep track of which variable is doing the work, so a case that rests on more than a handful of them is usually resting on none.

The evidence for carries a source next to every figure. Sourcing every figure sounds like bookkeeping, and a source beside a number is what lets somebody else check the work. The evidence against is the section people skip, and skipping it is not laziness but a genuine psychological difficulty: writing down the strongest version of the case against one's own conclusion is unpleasant, and it is also the only way to find out whether the conclusion survives it.

The part headed what would change the view is what turns a written argument into a case. Written in advance and specifically. For the Sarvani Coatings claim above, two observations would break it: a gross margin still standing at 46.0 per cent four quarters later while what each tin costs in materials keeps climbing, and another year delivering a share gain of 0.13 percentage points or better. Both were set down in advance of anything happening, and that ordering is the whole difference between a test and a commentary written afterwards.

The closing part, what is not known, ends the case honestly. For Sarvani Coatings that section is not empty: nothing in the published statements can tell a pricing environment shared across the whole field apart from what the company did with its own prices, and no amount of further reading of those statements will change that.

How to research an equity theme: what changes when it is not one company?

A theme is a claim about several companies at once. Industrial coatings demand is about to rise sharply. Repainting cycles are shortening. Raw material prices have peaked. Themes are attractive because one piece of thinking appears to pay off many times, and they are dangerous for exactly the same reason: a wrong theme is wrong everywhere at once.

The procedure differs from the single company procedure in four steps, and the fourth is the one that does most of the work.

One: establish what the theme actually asserts, in a single sentence. Most stated themes collapse at this step. Written down as one sentence, they turn out to assert either nothing testable or three separate things. "Industrial coatings demand rises" is testable. "The coatings sector is entering a new phase" is not a claim at all.

Two: find who gains and who loses if the theme is true. Both halves. Demand and prices move within a system, so a theme with only winners has usually not been thought through. Among the invented peers in this case, Kesaria Surface Solutions Limited is industrial heavy and would gain most from an industrial demand theme, Nandivarman Paints Limited is almost entirely decorative and would barely notice, and Thottam Chemicals Limited supplies resins and additives one step up the chain and would gain or lose depending on whether the theme moves volumes or input prices.

Three: check whether the gain lands in revenue or in margin. Revenue gains and margin gains are different claims with different sizes. A theme that adds volume adds revenue at the existing margin. A theme that lets makers price ahead of costs adds margin on the existing revenue. Margin effects are usually larger and always less durable: a price advantage invites a response and a volume advantage does not necessarily.

Four: test whether the theme is already visible in the reported numbers of the companies concerned. Most themes fail at this fourth step, and failing it is not a small problem. If industrial coatings demand were already rising sharply, it would show in reported segment revenue. Sarvani Coatings' industrial revenue went from Rs 510 crore to Rs 604 crore across the two most recent years, a rise of 18.4 per cent against 12.5 per cent for decorative, so industrial did grow faster and the industrial share of revenue moved from 24.06 per cent to 25.01 per cent, a shift of about 0.95 percentage points. Everybody else can read the reported numbers too, and has had since the day they were published, so a theme already visible in them is usually already in prices.

What separates research from an opinion about a share?

Everybody has an opinion about a share and opinions are free. Research is separated from an opinion not by confidence, not by length, and not by the amount of arithmetic in it. The separation is whether somebody else could disagree with it using the same document.

The disagreement test is sharper than it looks. A piece of writing that names its assumptions, cites its evidence and states what would break it hands a reader everything needed to argue back. A piece that says the shares are attractive because the business is strong hands the reader nothing to grip. The second one might be right. There is no way to find out from the document, and no way afterwards either. When the year is over there is nothing written down to check the reasoning against. The test of research quality is whether a reader could have disagreed with it using the same document, and a piece nobody could disagree with has hidden its assumptions rather than avoided having any.

Which leads to the part most treatments of this subject leave out. A view can be wrong and still be good work. Take the invented post mortem this case supplies. At the end of year two, when the published gross margin stood at 44.0 per cent, an analyst wrote that Sarvani Coatings' gross margin would compress, because input cost per unit was rising and the company had limited room to raise prices. Over the following year that margin went the other way and ended at 46.0 per cent. The call was wrong.

Now read the work rather than the outcome. The mechanism was right: input cost per unit did rise, by about 3.6 per cent, exactly as the analyst said it would. The magnitude assumption was wrong: realisation rose about 7.5 per cent, far more than the analyst had allowed for, so the margin widened instead of narrowing. The failing assumption is identifiable in the work today and could have been examined at the time, so the miss is a research error rather than bad luck. The error is also narrow rather than a broken case, and the difference matters: the analyst's model of what drives this company's margin survives, and one input to it was set too low.

What is the work not allowed to do?

Two answers, one general and one specific to India, and they are separate things.

The general answer is about what research can honestly claim. Nobody knows what a company will earn, and research cannot claim to. Research can claim that a set of assumptions is more consistent with the evidence than another set, and it can show its working. Every sentence that describes a future as though it were settled has quietly moved from the first claim to something the evidence cannot support.

The specific answer is about conduct. In India, research that reaches other people is a regulated activity, and the obligations attach at the moment of distribution rather than at the moment of thinking. The instrument is SEBI's research analyst regulations, and what it covers is registration, the disclosures that must sit alongside a piece of research, how conflicts are handled and what records are kept.