Security Analysis: The Discipline Behind a View

Security analysis is the discipline of reaching a view on an instrument from evidence somebody else can check, with every assumption written down and the strongest case against it worked through. The difference from an opinion is not confidence or detail. The difference is that a reader can put a finger on the one sentence where they part company with the writer.

Views about shares are the cheapest thing in finance. Ten people at a wedding, asked what they think of a company, produce ten answers, several of them delivered with total conviction, none of them costing the speaker anything. Views are free, they are unlimited, and the supply of them expands to fill whatever space is given to it. So the interesting question was never how to have one. The interesting question is what has to be true of a view before anybody, including the person who holds it, should treat it as work.

The job as it looks from the outside is already settled: the process, the note it produces, the peer set it rests on, and the instrument the whole thing points at. Underneath all of that sits the standard the work is held to, and what a written argument has to carry before it earns the word analysis. The standard is a short one. Most of the difficulty is not in understanding it. The difficulty is in noticing how often perfectly serious work fails it while looking completely respectable.

One warning first. Everything below runs on Sarvani Coatings Limited, an invented company whose every rupee was chosen so the arithmetic could be checked line by line. Building a view and testing whether it is defensible are the work below, and the road halts one step short of the sentence in which a view would be written down. The halt is where the teaching ends and something else begins.

What is security analysis, as distinct from having an opinion?

Start with the shortest useful definition. Security analysis is a view about an instrument together with a route to that view which another person can walk. Both halves matter and the second one is the whole of the difficulty. Anybody can produce the destination. The discipline requires that the road be laid down beside it, in public, with every turn marked.

The blunt version of the test is settled earlier: could somebody disagree with this using the same document. The blunt version is the right test, and in that form it is easy to pass with a shrug, so it is worth sharpening now. The sharp version follows. A reader must be able to point at one sentence and say, that one, that is where the disagreement starts. Not disagree with the mood of the note. Not feel unconvinced. Locate the disagreement at a specific claim. The argument that follows is then about that claim rather than about which party is generally more sensible.

Feel the difference on something that has nothing to do with shares. Two people are arguing about a wedding. The first says the whole thing is going to be far too expensive. The second says the hall is Rs 4,00,000/- and the catering Rs 900/- a plate for three hundred plates, or Rs 2,70,000/-, so those two alone come to Rs 6,70,000/- against the Rs 5,00,000/- set aside, and the gap is Rs 1,70,000/- before anybody has bought a single garland.

Now the two versions under argument. With the first, there is nothing to grip. A reader can say they disagree, and then there are two people with two feelings and no way forward. With the second, there are four separate places to stand. The hall quote can be called stale. Three hundred can be called high, and put at two hundred and forty. Rs 900/- a plate can be said to include service, with the comparable figure at Rs 760/-. Or every number can be accepted and only the conclusion disputed, on the grounds that the Rs 5,00,000/- was never the whole budget. Four doors, and each one leads somewhere. Because a shared object sits between the two parties, the disagreement can be settled, or at least narrowed.

The shared object is the entire distinction, and it survives translation into finance intact. An opinion has no shared object; it reports an internal state. Nothing sits between two opinions that either party could examine and be moved by, so two of them can be exchanged and compared for confidence but never resolved. Opinions can only be traded. Analysis can be argued with, and that is why analysis can improve while an opinion can only be repeated more loudly.

The test does not measure length. A note running to three printed sides can be entirely opinion, and a four sentence one can be perfectly good analysis. The test does not measure arithmetic either: twenty exhibits attached to an unstated assumption produce a note nobody can argue with, and that failure has its own block below. And the test does not measure confidence, where the relationship runs backwards. The more certain a note sounds, the more likely that the places a reader might have pushed back have been smoothed over rather than exposed.

What does the discipline require that an opinion does not?

Three things, and the useful way to learn them is not by memorising the list. Each absence produces a different kind of damage, at a different time, to a different person, so the way to learn the three is to look at what each one does when it is missing.

Stated assumptions, which are what make a view updatable

An assumption is a thing treated as settled in order to get on with the work. Every piece of analysis has them, without exception. The question is never whether a written argument has assumptions. The question is whether they are written where a reader can see them, or left inside the writer's head where nobody can.

Take the household version first. A household plans the year on one salary. The plan is perfectly sensible and it rests on an assumption: that the salary continues at its current level. If that assumption is written on the fridge alongside the plan, then on the day the salary changes the household knows exactly which line to redo and roughly what happens to everything downstream of it. If it was never written down, the plan simply stops working one month and nobody can say why, and the conversation that follows is about blame rather than about arithmetic.

Stated assumptions are what make a view updatable, and a view that cannot be updated is a view that dies the first time the world moves. Missing assumptions are the failure that arrives late. On the day the note is written, they cost nothing at all. The note reads beautifully. A year later, when the margin has gone the other way, somebody asks which part of the reasoning broke, and the honest answer is that nobody knows, including the person who wrote it. Nothing is learned, and the same error is fully available to be made again next year.

Cited evidence, which is what lets a reader check instead of trust

Citing evidence means naming where each figure came from, precisely enough that a reader can go and look at the same thing. Not gesturing at the annual report. Naming the line, the year and the statement, closely enough that the reader's second look lands on the same number the writer's first look landed on.

Consider a street vendor's daily takings. A vendor who says business is good has offered a summary, to be either accepted or rejected. A vendor who opens the notebook where every day's takings are written lets the week be added up independently, and lets the two missing days be noticed. The summary would never have shown those two days. Uncited evidence converts a reader into a believer, and a believer cannot catch a mistake, so a believer is exactly what analysis does not want. An uncited figure fails on the day, not a year later, and its cost is that the reader's own knowledge is locked out of the work.

An examined counter case, which is how the strength of a claim is found out

The counter case is the strongest version of the argument against the writer's own conclusion, written out properly and taken seriously rather than set up to be knocked over. The counter case is the requirement people skip most, and they skip it for a reason that feels virtuous: they have already thought about it, so writing it down feels like padding.

Writing it down is not padding. Until the opposing case is written at full strength, nobody knows how strong the claim is, the writer included. The claim may survive it comfortably and show itself robust. The claim may survive only under one condition, in which case the condition has been found and belongs in the note. Or it may not survive, in which case the writer has been saved by the only person in a position to do it. A note without a counter case has not measured its own strength, so its confidence is decoration rather than a finding.

Where did the discipline come from, and what problem was it built to solve?

Security analysis as a systematic method is associated with Benjamin Graham and David Dodd, and with their book Security Analysis, published in 1934. The date is doing work. The book arrives immediately after a stretch in which prices had travelled a very long way from anything observable about the businesses underneath them, and then travelled a very long way back.

Graham and Dodd proposed no formula. The proposal was a change in what a price is treated as. Before it, the only comparison generally available was a price against another price: this share against that share, this price against the price last month, this price against what somebody expects the price to be. Every one of those comparisons stays inside the market. None of them ever touches the thing being priced.

The move was to treat a market price as a claim about a business. A claim can be examined against the business rather than only against other prices. Once that is said, an entire body of work becomes available and it is work anybody can do. The statements exist. The order book exists. The factory exists. Everything the business earned, everything it spent, everything it holds and everything it owes can be read, and the question can then be put: what would have to be true of that business for this price to be a sensible claim about it. The step goes outside the market and finds a second surface to stand on.

Graham also gave the discipline its most quoted idea, and the name is part of the term rather than a courtesy. Graham's margin of safetyBenjamin Graham's term for buying at a price far enough below one's own estimate of worth that being wrong by a normal amount still does no harm. How the estimate itself is built belongs to the valuation layer and is covered separately. is his, and the arithmetic of building the estimate it is measured against belongs to the valuation layer and is covered separately. The posture behind it is what matters: it assumes in advance that the estimate is wrong by some amount, and asks how much room has been left for that. The margin of safety is a discipline about the analyst's own fallibility rather than about the company.

The book is worth taking a posture from rather than a sentence: the standards a piece of work is held to should themselves be visible, so a reader can examine the standards and not only the conclusion. Everything else below is a working out of that one idea, and the original stands on its own.

Say plainly what the move does not license. Treating a price as a claim about a business does not mean the method delivers the right answer, and it does not mean a business holds a single true number waiting to be found. The move means something narrower and more durable: there is now something to argue about other than the price, and the argument runs on evidence rather than on whose feeling is stronger.

What is the difference between analysis and forecasting?

Analysis and forecasting get conflated constantly, and the conflation is what produces most of the bad writing in the field. A forecast is a statement about what will happen. Analysis is a statement about what the evidence supports and how strongly it supports it. The two are different kinds of sentence, they carry different obligations, and one of them can be checked today while the other can only be checked later.

Here is the sentence most people find surprising the first time. An analyst can be highly confident about a mechanism and quite uncertain about a number, and saying both out loud is a finding rather than a weakness. The two confidences are separate quantities, and not two readings of the same dial.

A doctor makes this move every day without anybody thinking it evasive. The mechanism is an infection, and about that the doctor is not remotely uncertain: there is a test, and the test says so. How many days until the fever breaks is a different question, and because the answer depends on things nobody can see yet, the honest answer is a range. Nobody hears that and concludes the doctor is hedging. The patient hears a person who has separated what is known from what is not, and that separation is more useful than a single confident number would have been.

Applied to the case, the mechanism is one that supports confidence: if what each unit of output swallows in cost of materialsThe pigments, resins, solvents and packaging that physically go into the product, taken as a cost line. It is settled in the accounting layer and used here without being rebuilt. rises, gross margin narrows unless realisationRevenue divided by the number of units sold, so the average price a company actually collected once discounts and the sales mix are inside it. Two different things push it up: what was charged, and what was sold. rises by at least as much. The relation is close to arithmetic. It is not a guess about the future and it does not depend on anybody's judgement. Where realisation goes next year is a completely different question, and there the honest answer is a range with the reasons for its width written beside it.

Now the practical detection rule, applied to somebody else's work in about four seconds. A note carrying a single number and no range has usually smuggled a forecast in behind an analysis. Not always. Some quantities really are that tight. But a single point estimate for something genuinely uncertain is a claim about the world dressed up as a conclusion about the evidence, and the giveaway is that no width is offered anywhere in the note.

The reverse error is worth naming too, so the correction does not overshoot. A range wide enough to contain every outcome anybody could imagine has also produced nothing. Width is not virtue; it reports how much the evidence pins down. Where the honest width is enormous, the useful sentence says the evidence pins down very little here, rather than dressing that up as a range.

How does an analyst know when the evidence is enough?

Practitioners argue about this one most, and the argument goes nowhere because both sides want an absolute answer. There is no absolute answer. There is a very good relative one.

Evidence is enough when the uncertainty that remains no longer changes the ordering of the possibilities that actually matter. The test says nothing about how much evidence has been gathered, how many hours were spent, or how confident anybody feels. The test asks one question: if the next thing were learned, would the possibilities swap places? If they would not, the work is finished, whatever the pile on the desk looks like.

The everyday version is immediate. A parent is choosing between two schools for a child. One is eight minutes from the house and one is forty. Both are perfectly good, the fees are similar, and each school hands over a brochure of forty printed sides. The first four sides of each already set the ordering by the commute: forty minutes twice a day is what will actually shape the next several years. The remaining seventy two sides hold the canteen menu, the founding date, the colours of the sports houses. None of them moves the ordering. So the reading is not diligence. The reading is a way of postponing the decision while feeling responsible.

Two consequences fall straight out of this, and both change how the work is done.

The first is that sufficiency is relative to the question. Change the question and the same pile of evidence becomes sufficient or insufficient without a single sheet being added or removed. Ask whether Sarvani Coatings grew faster than its sector last year, and the published revenue figures settle it with nothing further needed: Rs 2,415 crore against Rs 2,120 crore is 13.9 per cent, the sector grew 11.0 per cent, done. Ask why the gross margin rose, and that same pile settles nothing at all, and it can be read until the eyes give out. Same evidence, two questions, two verdicts on sufficiency.

The second is uncomfortable and it is worth sitting with. An analyst who keeps gathering evidence after the ordering has stopped moving is not being careful; they are avoiding the writing. The charge sounds harsh and it is not meant as a character judgement. The extra hours buy nothing about the question and quite a lot of relief from the discomfort of committing something to paper. The tell is easy to spot: re-reading a document already read, rather than going and getting a document not yet read.

One genuine exception, stated so the rule does not become an excuse to stop early. Evidence that leaves the ordering alone can still be worth gathering when it changes the size of the gap and the size is what the question turns on. Knowing one school is better differs from knowing it is slightly better rather than enormously better. But that is a different question, so the test has been applied correctly rather than suspended.

What does the discipline refuse to do?

A discipline is defined as much by its refusals as by its methods, and these three hold the whole thing together. Each is easy to state, and each is broken constantly by people who would recite it correctly if asked.

It refuses to treat price movement as evidence about a business

The share price went up, so something must be going right. The sentence appears in serious work more often than anybody would like. The reasoning is circular, and the circularity is worth seeing clearly rather than merely being told about. The task is to work out what a business is doing. The price is a collection of other people's answers to that same question. Using it as evidence means using other people's answers as the input to one's own. The price becomes both the question and the answer, and the loop never touches the business at all.

There is a serious idea in the neighbourhood and it should be named rather than described anonymously. How fast and how completely prices absorb information that is publicly available is the subject of Eugene Fama's work on market efficiencyEugene Fama's line of work asking how quickly and completely prices take in information that is already public. It is a question about the speed of the market, and it is covered separately., and it is a real question with a real literature behind it. Even the strongest version of it does not rescue the sentence above. Efficiency says that a price already contains what is publicly known. Efficiency does not turn a price into an observation about a gross margin, and it certainly does not settle which of three explanations for that margin is the right one.

The household test settles it. The neighbours all decide a house is worth more this year. Has its roof improved? The two questions are not related by anything, and no amount of agreement among the neighbours puts a single tile back.

It refuses to let confidence stand in for evidence

Confidence is a feeling about a claim. Evidence is a thing outside the writer that bears on the claim. The two can move together, and very often they do not. The confident writer who is wrong is not the dangerous case; that writer gets found out. The dangerous one is the confident writer who is right, whose confidence was never earned by the evidence and who now has one more reason to trust it next time. The practical form of the refusal is a habit: on noticing certainty, find the sentence the certainty rests on and ask what it cites. Often it rests on another sentence by the same writer, and that one on a third, and the chain never reaches anything a reader could check.

It refuses to reach a conclusion the stated evidence does not support

The third refusal is by a distance the hardest. The demand for a conclusion is almost always external, and it usually comes from somebody with a legitimate reason for wanting one. A decision has to be made. A meeting is at four. Somebody has asked a direct question and is waiting. Against all of that, the sentence that the evidence does not settle this yet sounds like a failure to do the job, and it takes a certain steadiness to write it.

But consider the alternative. A conclusion stated beyond what the evidence supports is not a conclusion at all; it is a guess in the costume of one, and the reader will now weigh it as the costume implies rather than as the guess deserves. The reader is worse off than if nothing had been said: the precision is false and it cannot be seen to be false. The honest output is not a shrug either. The honest output names which possibilities remain open, what evidence would separate them, and what that evidence would cost, and that is a genuinely useful answer to somebody deciding at four o'clock.

How is a piece of analysis judged when the outcome is already known?

Judging the work by the outcome is where most people go wrong, and the going wrong feels not merely reasonable but obviously correct. The call was right, so the work was good. The call was wrong, so the work was bad. Nothing sounds fairer than that.

The trouble is that a single outcome is one draw from a spread. The process that produced the view is a different object entirely, and it does not become better or worse depending on which way one draw landed. The outcome and the process are not the same thing being measured twice, so a view that turned out right can be poor work and a view that turned out wrong can be excellent work.

Judging a decision by how it turned out has a name. Annie Duke calls it resulting, in Thinking in Bets, published in 2018, and having the word available makes the error far easier to catch. Without a name for it the error is invisible: it does not feel like an error at all. It feels like being sensible.

The household version is unpleasant and clarifying. A household drops the fire insurance because the premium went up and money is tight. Nothing burns down. Was that a good decision? The outcome says yes and settles nothing. The same decision run through a hundred years produces a small number of ruinous ones, and this household drew an ordinary one. Only reading the reasoning separates the two.

So how is a piece of work actually judged once the outcome is known? By reading the work and looking for four things, none of which is the outcome. Were the assumptions written down. Was the evidence cited well enough to be checked at the source. Was the case against written out at its full strength. And when the view broke, was the cause identifiable in the work as written, or was it something nobody could have seen.

The last question is the one that separates a research error from bad luck, and the distinction is not a comfort blanket. The distinction has a hard test. A research error is one where the failing assumption can be pointed at today, in the work as it was written at the time. Bad luck is where the work would have to have contained something that was not knowable when it was written. A full post mortem on an invented call about this company, reaching exactly that verdict, is covered separately.

What does one claim about Sarvani Coatings look like written both ways?

Enough description of the standard: here it is run on something. One claim about Sarvani Coatings Limited is written twice. Same writer, same figures, same honesty. Only the discipline differs.

Version one, written as an opinion

Sarvani Coatings has good margins and is a well run company. The numbers have been moving in the right direction for three years and management seems to know what it is doing.

Nothing in that is dishonest. The margins did move, in the direction described, and a person could write it in complete good faith. The sentence still hands a reader nothing. Try to disagree with it precisely and there is no word to contest. Good is not a quantity, right direction is not a measurement, and seems to know what it is doing reports the writer's impression, and nobody can check an impression unless they already share it.

Version two, written as analysis

Now the same territory, written so that every sentence can be argued with separately. Every figure below belongs to the invented case and each one is stated as a share of revenue in the year named.

Sentence six is the one that costs something to write. Sentence six is the counter case, and it says that if input costs across the whole field did the work rather than anything Sarvani Coatings decided, then the year three EBITDA marginProfit measured before interest, before tax and before the depreciation and amortisation charges, then set against revenue. It sits below gross margin on the ladder, so the employee and other expense lines are already out of it. of 18.47 per cent is a peak rather than a level. The sentence weakens the writer's own position, and it belongs in the note precisely because it does.

Apply the test from the opening block and the difference is not a matter of taste: in version two a reader can point at sentence three and say they disagree, and in version one there is no sentence to point at. Sentence three is the whole distinction, and the distinction survives every attempt to restate it as a difference of length, tone or arithmetic.

Sentence two looks like the least interesting line in the table and is doing something structural, so it is worth pausing on. Gross profit is nothing other than revenue with the cost of materials taken out. So the gross margin and the materials share must sum to 100 per cent in every single year, by construction rather than by coincidence. A three point rise in one is the same three point fall in the other, seen from the opposite side. The identity is not a finding. The identity is a check, and it is the kind of check that tells a reader instantly whether the writer is computing or copying.

Now run the sufficiency test on it

The three explanations and the published ladder are both in hand. The starting position is also the position the interactive below opens in, so it is worth stating in plain figures. With only the published statements in hand, all three explanations are consistent with everything available, and the ordering between them is flat. Gross margin 43.0 per cent rising to 46.0 per cent, the cost of materials 57.0 per cent of revenue falling to 54.0 per cent, and not one line anywhere in those statements that any of the three explanations fails to produce.

Now start adding things and watch which ones earn their place.

Reading the same three annual reports a fourth time changes nothing. It cannot. A document already consistent with all three explanations is consistent with all three however carefully it is read, and no amount of attention converts consistency into separation.

The volume and realisation split for year three is genuinely informative and it still separates nothing, and that is the more instructive case of the two. Revenue was 13.9 per cent higher and volume 6.0 per cent higher, putting realisation roughly 7.5 per cent above the year before. Materials per unit of output went the other way and finished roughly 3.6 per cent dearer. The split is a real finding, and it runs the wrong way round from the story most people expect: nothing became cheaper, and the margin widened anyway because realisation ran ahead of input cost. Tested against the three explanations, though, it separates nothing. A price rise across the whole field raises realisation. Sarvani Coatings raising its own prices raises realisation. A shift in mix towards industrial coatings raises realisation. All three produce exactly this reading, so the ordering does not move a millimetre. A piece of evidence can be entirely genuine, entirely new to the reader, and still separate nothing, and that is the case learners find hardest to accept.

The segment reportingThe requirement that a company splits its revenue and results across the distinct parts of its business, so a reader can see the pieces rather than only the total. What must be split, and how, is set out in the accounting layer. split does move it, and it moves it decisively, by arithmetic. Industrial coatings were Rs 604 crore of Rs 2,415 crore in year three, or 25.01 per cent, against Rs 510 crore of Rs 2,120 crore in year two, or 24.06 per cent. The mix shifted 0.95 percentage points. The mix shift is a one year figure, so it has to be tested against the one year gross margin gain of 2.0 points, from 44.0 to 46.0, and never against the two year headline of 3.0. Dividing 2.0 points of margin by a 0.95 point shift in mix shows that mix could only have done this if the two segments' gross margins differed by about 209 percentage points. A gross margin cannot exceed 100 per cent. The requirement is impossible by a factor of two, so mix is out, and it is out on arithmetic rather than on judgement.

The peer gross margin record moves it too, and only part of the way. Across those same two years Sarvani Coatings picked up 3.0 points of gross margin, Kesaria Surface Solutions Limited picked up 3.6 points and Nandivarman Paints Limited 2.4. Every one of the three rose. A pricing environment shared across the field explains a gain that every maker had; one company's own pricing does not explain what happened at the other two. So the field wide explanation moves ahead. The field wide explanation does not win. Sarvani Coatings sits between its two peers rather than outside them, and nothing in that record separates whatever it did on its own from what everybody was doing together.

Which evidence would separate those last two? Per unit realisation for each maker, or a channel checkGoing to the people who actually buy or sell the product, such as dealers or contractors, and asking what they paid and what they saw. A research technique covered separately, and named here only as a kind of evidence. across the same dealers through the year. Neither of those is in the published statements, and neither is in the notes to the accountsThe pages behind the main statements where a company sets out the detail: what sits inside a total, what the policies are, and what happened that the face of the statements does not show. either. Stopping there is perfectly respectable, and the honest sentence is that the question narrowed and did not close.

Two of the buttons produce no movement at all, and they produce it for completely different reasons. The difference between the two reasons is worth more than the rest of the interactive put together. A third reading is redundant, meaning it was capable of separating something and had already spent that capacity; the share price is inadmissible, meaning it was never capable of separating anything about a business in the first place. Both feel like work. Only one of them was ever going to be.

And here is where the work stops

Everything above narrows a question. None of it produces a view about Sarvani Coatings Limited. The honest written output at this point is a paragraph saying that the gross margin gain is more consistent with a pricing environment shared across the field than with anything the company did alone, that a shift in mix is ruled out by arithmetic, that the two remaining explanations cannot be separated by anything published, and that per unit realisation across the field would separate them. The paragraph is analysis. Deciding what somebody should do about it is a different matter, and the discipline set out above does not settle that.

Where does thorough looking work still fail?

When does doing this work become a regulated activity in India?

Everything above is about the thinking. The moment the thinking is written down and passed to somebody else, a second set of considerations attaches to it, and they come from a different place entirely.

In India, the moment analysis is handed to somebody else it becomes a regulated activity, and the requirements fasten onto the handing over rather than onto the thinking. Working out in private what a company's margin did attracts no rule at all. Publishing that work, sending it to clients, or putting it in front of anybody who might act on it brings registration and disclosure requirements into play, and they are ones the Securities and Exchange Board of India writes. The rule set is the research analyst regulations.

The conduct rules take the same shape as the discipline itself, and both care about what a reader can see. The discipline asks the analyst to expose their assumptions so a reader can check the reasoning; the conduct rules ask the analyst to expose their interests so a reader can weigh the reasons for writing. Neither is about being trustworthy. Both are about being checkable, and checkable is far more useful.

Who actually uses this discipline outside a research desk?

All this reads easily as professional manners for people who write about shares. The standard is really about making a claim checkable, and that problem turns up wherever somebody decides on incomplete evidence and has to account for it afterwards.

The reason the discipline travels this well is that it was never really about securities. The discipline is a way of making a claim examinable by a person who was nowhere near the room where it was formed, and that problem is older than markets and shows up everywhere they are not.