How to build an equity research peer map

A peer map is a set of companies chosen so that a difference between them means something. The question is named first, then each candidate is tested on what it sells, who it sells to, how it earns and what drives its cost. A set assembled for convenience reports a difference that belongs to the selection rather than to the companies.

The panel opens on the two members the worked case below uses, compared on the two year gross margin gain. Two readings make an even count, so the median is the two halved. 2.4 and 3.6 give 3.0 points, and Sarvani Coatings' own gain is 3.0 points exactly, so the gap is nil. Removing one member changes that. Without Nandivarman Paints Limited the median becomes 3.6, so the subject reads 0.6 points below its set. Without Kesaria Surface Solutions Limited instead it becomes 2.4, so the subject reads 0.6 points above. Two findings 1.2 points apart, both reachable by removing one name, and not a figure in any of the three companies moved between them.

The panel keeps a running check on itself for that reason, comparing the largest move any single exclusion would make against the gap the set is being used to state. At the opening setting the gap is zero and one exclusion moves the median 0.6 points, so the move is the larger of the two. Tick Thottam Chemicals Limited and Perungudi Powder Coatings Limited in as well, and with four members the largest single exclusion moves it 0.3 points against a gap of 0.9, so the check goes quiet. The check says how much one name is carrying, and nothing about whether the set answers anything in particular.

Underneath the whole exercise is one uncomfortable property of every comparison: the answer is relative to whatever was compared against, and no average knows where its members came from. The judgement that decides the reading is made before any figure is opened, by whoever wrote the list, and it almost never gets printed. A peer map is a way of dragging that judgement into the open where somebody can argue with it.

Two things are settled elsewhere and not rebuilt here. The order research runs in, question before evidence, is covered separately, and multiples are arithmetic assumed already in hand. The new ground is narrower: who goes on the list, and how a reader checks the work.

What is a peer map actually for?

Something domestic first. A household's electricity bill last month was Rs 3,400/-. Is that high? Unanswerable until the comparison is named. Against the same household's bill a year ago it is a question about that household. Against the flat next door it is about how the two households live. Against the block average it folds in every one bedroom rental with a single occupant, and makes a four person household look extravagant every time. Same bill, three comparisons, three findings, none of them wrong. The three comparisons are answers to three different questions.

A peer map is that choice made deliberately and written down: companies assembled so that when one differs from another, the difference answers the stated question. The set is not a list of companies that resemble each other; it is a list that holds constant everything except the thing being asked about. That is why a set is built for one question rather than for a company.

The case this guide runs on. Sarvani Coatings Limited is a listed maker of decorative paints and industrial coatings. Its gross marginWhat is left of every hundred rupees of sales after the cost of the materials that physically went into what was sold, before wages, freight, advertising and everything else. went from 43.0 per cent to 46.0 per cent over two published years, a gain of 3.0 percentage points. Did Sarvani Coatings do something, or did something happen to everybody who makes paint?

Notice what that question demands. The question is about an input price move, so it demands companies that buy roughly the same inputs. Size, exchange and reputation as a competitor do not enter it, and those three are the properties a set gets built on when nobody wrote the question down. The question chooses the tests, and the tests choose the set, in that order and never the other way round.

What is the sequence, start to finish?

Six steps, and they are steps rather than ideas: each produces something that can be handed to somebody. The difficulty in peer selection is discipline rather than technique, so none of the six requires a mechanism the analyst did not arrive with.

1. Write the one question the set exists to answerOne sentence, on paper, before a single company name goes on the list. It names the thing to be explained and the period it is explained over.
   Checking: could a stranger read that sentence and say what would count as a difference?
2. Run four tests on every candidate, in the same order, and record all fourWhat it sells and to whom. How it earns. What drives its cost. What it is exposed to. Write pass, partial or fail beside each, for every name, including the ones that seem certain.
   Checking: does every candidate carry four recorded results rather than an impression?
3. Label every partial peer on the face of the setA candidate that fails a test can still go in. It goes in carrying the label, the failed test named, and one line saying which half of it the analysis is using.
   Checking: can a reader see which member is partial without asking?
4. Fix the comparison fields before opening anybody's filingsGrowth, margin, capital intensity, working capital, return on capital employed. Write the list and date it. Then go and get the numbers.
   Checking: was the list written before the data, and can that be shown?
5. Publish the set beside the comparison, with the reason each name is in itThe set travels with the number, or the number travels alone and means nothing. One line per member is enough.
   Checking: could a reader rebuild the set from what was printed?
6. Write the exclusion record: who was considered, who was left out, and whyEvery name considered and rejected, with the test it failed. This is the part everybody skips and the part a careful reader reads first.
   Checking: does each rejected name carry a reason somebody could argue with?

What does a candidate have to pass before it goes in?

Four tests, and they are about the business rather than about the share. None of the four asks what a company is worth or how it is quoted. Worth and quotation are properties of the market in a company, and a set built on them holds constant nothing at all.

Test one, what it sells and to whom. Not the industry label, the actual product and the actual buyer. A household choosing a paint in a dealer's shop and a factory buying a coating on a twelve month contract are two demand curves wearing one word. The first responds to a repainting cycle; the second to somebody else's capital spending plan.

Test two, how it earns. Where the revenue physically comes from, line by line. A maker selling through a dealer networkThe chain of independent shops that stock a maker's product and sell it on to households. The maker sells to the dealer, and the maker's advertising spend buys the dealer's incentive to push one brand rather than another. earns on volume through thousands of small transactions. A maker supplying three large manufacturers earns on a handful of contracts, each of which can be renegotiated. Same revenue line, entirely different fragility.

Test three, what drives its cost. Usually one input, and usually the one that moves most. For a coatings maker it is a small set of pigments and resins whose prices follow crude oil, so an input move reaches every maker at roughly the same time. For a services business it would be wages. A margin question lives or dies on the cost driver test. Running that test means reading the notes, and that is why it is skipped most often.

Test four, what it is exposed to. The cycle, the customer concentration, the rule it lives under. Ten shops in one shopping centre share a single driver: if that centre loses its anchor tenant, all ten suffer together and none of them did anything. The five forces frame set out by Michael Porter in Competitive Strategy, 1980, lays this out, and the part that matters here is narrow: companies that look alike can be exposed to different things.

The tests are independent, and here is the part that catches people. What matters is not how many tests a candidate passed but which one it failed, so a candidate can pass three, fail the fourth and still be useless. Fail the cost driver test and it is worthless for a margin question and good for a growth one. Fail the buyer test and it is worthless for a demand question and possibly excellent for a cost one. The question decides which failure is fatal.

What is to be done with a company that is only half comparable?

The instinct is to throw it out, and the instinct is usually wrong. A candidate that fails a test is not contaminated; it is informative about some things and silent about others. The partiality is not what makes a partial peer dangerous. Putting one into a set and saying nothing about it is.

Think about wedding catering quotes: three for 400 guests and one for 90. Dropping the small one loses the only quote showing what fixed costs look like spread thin. Averaging all four silently gives a per plate figure no caterer would honour at either scale. Keeping it with a note reading 90 guests, different pricing structure, used only for the fixed cost point costs nothing. The label is what converts a contaminated average into a usable observation.

Kesaria Surface Solutions Limited is exactly that case: a smaller maker whose revenue is about 82 per cent industrial, sold to manufacturers on contract, against Sarvani Coatings at about 75 per cent decorative sold through dealers. On what it sells and to whom it fails. On what it is exposed to it fails again. Industrial capital spending and a household repainting cycle move on different clocks. On how it earns and on what drives its cost it passes outright: it sells coatings by volume, and buys the same crude linked pigments and resins from the same kind of supplier at the same time.

For a margin question the two tests it passes carry the question and the two it fails do not. So it goes in, and it goes in loudly: partial peer, fails on buyer and on cycle exposure, included because the shared input cost is the thing being tested. A reader who disagrees now has something specific to disagree with.

What gets compared once the set is fixed?

Five fields do most of the work in a first pass: revenue growth, margin, capital intensity, working capital, and return on capital employed. The list is not sacred. The timing is: the five fields go down on paper before anybody's accounts are opened.

Here is why that timing rule exists. Once four sets of numbers are on a screen, some comparisons separate cleanly and some are a muddle, and the clean ones are quite often clean because of an accident in how the set was assembled. Choosing the comparison after seeing the data is the moment the selection stops being analysis and becomes an argument looking for evidence, and nobody who does it experiences it that way.

The five fields, and what each is doing in a set of coatings makers:

How does the set quietly decide the answer?

Every comparison of this kind produces a statement of the form the subject sits here relative to the set. The subject's figures are fixed and the set's figures are fixed. The membership is not, and it is a judgement made by one person, usually quickly, and usually never printed.

So the conclusion is a function of a choice the reader cannot see. Not partly. In a relative comparison, the selection is not an input to the result; it is half of the result.

Run it on the real question; the panel at the top computes every figure below. Sarvani Coatings gained 3.0 percentage points of gross margin over two published years, Nandivarman Paints Limited gained 2.4 points over the same period, and Kesaria Surface Solutions Limited gained 3.6 points.

Build the set the question asks for, the two coatings makers. The average gain is 2.4 plus 3.6, halved, and 3.0 points is Sarvani Coatings' own gain exactly. The gap is zero: the gain looks like something that happened to everybody who makes paint.

Now build the set somebody assembles when no question was written down, almost always a set built on size. Nandivarman Paints, at Rs 14,490 crore against Sarvani Coatings' Rs 2,415 crore, is six times larger and gets dropped as not comparable. Thottam Chemicals Limited, at Rs 3,860 crore, is 1.6 times its size and looks like a natural fit. Kesaria Surface Solutions, at Rs 1,449 crore, stays. The size set averages 3.6 and 0.4, giving 2.0 points, and Sarvani Coatings now sits 1.0 point ahead of its field: the gain was company specific.

The only thing that moved was a list nobody printed: four companies, four unchanged sets of figures, two opposite conclusions. The second set is not a stupid set. Every name in it is a listed company of roughly the right size in coatings and coatings chemicals, and that is how a set gets built at four in the afternoon.

One boundary. Everything above runs on an operating measure, a margin gain, deliberately, and the identical problem attaches to any relative comparison, including one of price to earningsThe ratio of a share's quoted price to the earning attributed to it, printed on any exchange quote screen. How it is built and what it can carry are covered under valuation. multiples. A multiple carries the market's view of every member as well as the analyst's selection of them, so the case there is stronger rather than different. The selection problem is what travels, and it comes before whatever measure is plugged into it.

The tolerance panel below starts from that same set built on the question: Nandivarman Paints in, Kesaria Surface Solutions in as a labelled partial peer, Thottam Chemicals out. Two changes in order rebuild the size set exactly. Dragging the tolerance below 19 points drops Nandivarman out on distance, and ticking Thottam Chemicals in then gives the size set precisely: average 2.0 points, gap 1.0 point.

Why is the excluded list worth more than the included one?

An analyst asked why a company is in the set gives a description: same products, same buyers, same input. Similarity is easy to narrate after the fact, so every word is true and none of it informative. Asked why a company is out, the same analyst gives a decision. A decision is where the judgement sits.

An exclusion record is the plainest document in research: every name considered, and one line each on why it was left out. The record takes ten minutes, it is skipped almost universally, and its absence is invisible. A reader can see the set that was published; a reader cannot see the set that was nearly published.

There is a household version. When somebody says the fair price for a second hand scooter is Rs 42,000/-, the useful question is not which listings they looked at but which they threw out. Three dropped for high mileage and two for being in another city is a different average from three dropped for being cheaper than the number the person had already decided on.

What does the whole sequence look like run on one company?

Sarvani Coatings Limited, three candidates, one question. Running the sequence end to end makes the awkward decisions arrive in a fixed order rather than all at once.

Step one, the question. Written before any name: was Sarvani Coatings' gross margin gain of 3.0 percentage points, over the two published years, something that happened to it or something it did? The question rules out a great deal: it does not ask whether the margin is good, and it does not ask about growth.

Step two, the four tests. Meghna Iyer, the analyst working the case, records all four for all three candidates rather than only the doubtful ones. A recorded pass is what lets somebody disagree with her later.

Step three, the labelling. Kesaria Surface Solutions goes in carrying its label, as set out above. Nandivarman Paints needs none: it is larger and almost entirely decorative, but size and mix are not among the four tests and neither is a reason to attach a warning.

Step four, the fields. Written and dated before any filing is opened: revenue growth, gross margin, capital employed to revenue, working capital days, return on capital employed. Only the second is load bearing here. Dropping the ones expected to be dull is how a list becomes a case, so the other four go in anyway.

Step five, the published set. Two names, each with its one line reason, printed beside every comparison figure the note carries.

Two things in that table deserve a sentence. The set average gain of 3.0 points is Sarvani Coatings' own figure exactly, and the match is the finding: the gain sits in the middle of a field where everybody gained. The set average gross margin of 42.8 per cent is nearly useless, being the midpoint of a decorative heavy maker at 47.5 per cent and an industrial heavy one at 38.0 per cent, with no company anywhere sitting at 42.8 per cent for a reason. An average of two numbers that differ for a known reason is not a benchmark; it is a coincidence with a decimal point.

Step six, the exclusion record. Thottam Chemicals Limited, out: it supplies resinsThe binder in a paint. Resin is what makes the film stick to a wall and stay there, and along with pigment it is the largest part of what a coatings maker physically buys. and additives to coatings makers rather than making coatings. And here the exercise turns: the excluded name is the most useful company in the case.

Why is the excluded company the most useful one in the case?

Thottam Chemicals Limited fails as a peer, and it fails cleanly. Thottam sells no coatings, sells to none of the same buyers, and earns by selling chemical intermediates by the tonne. Nobody should put it in a set built to compare coatings makers.

And it is the single company whose figures would do most to answer the question the set was built for. The gain at Sarvani Coatings is the gap between what it charged and what its inputs cost, and those inputs come from companies like Thottam Chemicals. If the input price move was passed straight through the chain, Thottam's own margin barely moves. If that margin widened sharply over the same period, value moved upstream and the coatings makers gained despite their supplier. A peer shows whether a move was shared across a field; a supplier shows where in the chain it happened.

The honest ending is that the evidence here cannot resolve it. Thottam Chemicals' gross margin gain over the two years is 0.4 points, and a small ambiguous move at the supplier does not separate a field wide pricing environment from Sarvani Coatings' own pricing. The supplier's figure narrows the live explanations instead. Evidence that narrows a question without closing it is still evidence, and a note reporting it as a resolution has broken something.

When is a peer map the wrong tool?

Three situations, and the first is the hardest to accept.

Where a company genuinely has no comparable. A maker with a process nobody else uses, a licence nobody else holds, or a buyer nobody else sells to. Forcing a set on it does not fail loudly. The tool returns an average like any other, with a gap beside it that looks like a finding. Nothing in the output says the tool should not have been used, and the silence is what makes it dangerous rather than merely limited.

Where the differences are larger than the similarities. Two companies can share an industry label and share almost nothing that matters. The honest move is to stop calling it a comparison and start calling it two descriptions, less satisfying and more useful.

Where the question is about the company's own history. To know whether Sarvani Coatings' realisationThe average price actually received for each unit sold, after discounts, and blended across whatever mix of products went out of the door. Realisation moves when prices move and also when the mix moves. is holding, the comparison required is Sarvani Coatings against Sarvani Coatings across quarters. A peer set answers a different question and will answer it whether or not it was asked. Segment reportingThe split of revenue and profit a company publishes by line of business, so a reader can see how much came from each part rather than only the total. is the better tool there, and it needs no peers at all.