Defining the peer set

Every relative valuation you will ever read rests on a hidden foundation, and the foundation is almost never examined. Before anyone says a company is "cheap" or "expensive" against its peers, someone had to decide who the peers are - and that decision, made in a few careless seconds by copying an index constituent list or a screener's sector bucket, silently determines the answer. A peer set built from the label the exchange files a company under, rather than from the way the company actually earns money, does not merely add noise to the comparison. It poisons it at the root, because every multiple you then average, every "discount to peers" you then compute, is measured against companies that were never comparable in the first place. illustrative

This module is the first in Part Six, and it comes before the modules on normalising the numbers, reading the valuation spread, and finding the low-cost operator, for the same reason the forensic mindset came before the specific accounting games: it installs the stance the rest of the part depends on. The stance is simple to state and surprisingly hard to hold. Comparability - the property that makes two companies legitimately comparable - is a fact about their economic model, not about their sector name. Two companies can sit on the same shelf at the exchange, in the same index, screened into the same bucket, and be as different underneath as a jeweller and a foundry. And the moment a non-comparable company enters your peer set, the comparison it corrupts will look exactly as authoritative as an honest one - a number, averaged, presented with decimal places - while quietly pointing the wrong way.

So the question this module answers is the one that sits underneath all of relative value and is usually skipped: who actually belongs in the comparison, and who only appears to? Get this wrong and everything downstream - the normalisation, the spread, the judgement of which name is genuinely mispriced - is arithmetic performed on the wrong set. Get it right and the rest of Part Six has something solid to stand on.

The reason a lazy peer set is so dangerous is that it feels rigorous. A screener returns "Sector: FMCG" and hands you twenty companies; an index groups thirty names under "Financial Services"; a sell-side comp sheet lists eight "peers" in a neat table. Each of these looks like the comparison has already been done for you, and each is built on a taxonomy designed for a completely different purpose - classifying companies for indices, for regulation, for data vendors - not for judging whether their valuations can be read against one another.

Relative valuation is the practice of pricing a company by reference to what the market pays for similar companies - an EV/EBITDA a turn below the peer group, a price-to-book at a discount to comparable lenders. It is powerful precisely because it side-steps the heroic assumptions of an absolute model and asks a narrower, more answerable question: is this business priced in line with genuinely similar ones? But that power is entirely conditional on the word "similar" being true. The technique cannot tell the difference between a company that is cheap because the market has mispriced it and one that only looks cheap because a structurally higher-margin, higher-multiple business was wrongly averaged into its peer group. The multiple has no idea what companies produced it. Feed it a false peer and it will happily report a false discount.

This is why the peer-set decision has to be made deliberately and first. The later modules in this part will teach you to normalise for one-offs and accounting differences, to read the spread of multiples across a group, to work out who the genuine low-cost operator is - and every one of those techniques assumes the group it operates on is real. Normalising the numbers of a company that should not be in the set at all does not rescue the comparison; it applies careful surgery to the wrong patient. The discipline, then, is to treat "who are the peers?" as the primary analytical act of relative valuation, not a clerical preliminary to it - and to build the set from what the businesses do, tested against the axes on which their economics actually differ.

To build a peer set honestly you stop asking "what sector is this?" and start asking "how does this business actually earn its money, and consume its capital?" Two companies are peers to the degree that they answer those questions the same way. In practice, comparability lives on a small number of structural axes, and a valid peer must share a company's position on the ones that dominate its economics.

Business mix and scale. A company that earns 70% of its profit from one product line is not a clean peer for a diversified group that happens to compete in that line as a fifth of its business - the blended multiple of the diversified group prices four other things too. Scale matters in the same way: a national player with distribution reach and procurement power earns structurally different margins than a regional challenger, so a size gap of several times is a reason to adjust, and sometimes a reason to exclude. Neither of these is visible in the sector label.

Integration: upstream versus downstream. Where a company sits in its own value chain reshapes its entire margin and capital profile. A vertically integrated producer that makes the raw material it converts captures a margin a downstream buyer never sees - and carries the capital, cyclicality and commodity exposure of that upstream operation. Put an integrated player and a pure converter in the same peer set and you are averaging two different businesses wearing one industry name; the integrated firm's fatter margin and heavier balance sheet distort every ratio the group produces.

Asset intensity. How much capital a business must sink into fixed assets to earn a rupee of revenue - its asset intensity - decides which multiples even apply. An asset-light distributor with negligible fixed assets and a high return on capital is not comparable, on book-based multiples, to an asset-heavy manufacturer whose entire value sits in plant; the light business will always look "expensive" on price-to-book and the heavy one "cheap", and neither fact means what it appears to. Comparability requires a shared answer to how the capital is deployed, not just what is sold.

Customer and go-to-market: B2B versus B2C. A business that sells a branded product to consumers, who choose it by name and pay a premium for it, lives in a different economic world from one that sells the same physical output in bulk to industrial buyers who choose on price. The first has pricing power, advertising as its main discretionary cost, and a margin it partly controls; the second is a price-taker whose margin is a spread it does not set. They can make an almost identical product and still not be peers, because the thing being valued is the economics of the sale, not the chemistry of the good.

Cost-curve position. In any commodity or near-commodity business, the only durable advantage is being the low-cost producer, and where a company sits on the industry cost curve defines whether it survives a downturn and how it should be valued relative to rivals. Two producers of the identical commodity, one at the bottom of the cost curve and one near the top, are not really peers even though nothing could look more comparable on paper - the low-cost one earns through the cycle while the high-cost one only earns at the peak.

The move underneath all five axes is the same. Take the company you want to value, write down where it sits on the axes that dominate its economics, and admit into the peer set only companies that share those positions - not the ones that share its index, its screener bucket, or its two-word sector name. A peer set is a claim that "these businesses are similar enough that their valuations can be read against one another," and like every claim in this guide it has to be earned, not assumed.

Take a composite company to make this concrete. Call it Kanva Foods - a mid-cap maker of branded packaged snacks and spreads, sold to consumers under its own labels, buying its agricultural inputs on the open market, distributing through a light, outsourced network. Its economics are those of a branded consumer business: a decent gross margin it partly controls through brand pricing, advertising as its largest discretionary line, modest fixed assets, and a return on capital well above its cost of capital. illustrative

A defensible peer set for Kanva contains other branded, downstream, asset-light consumer-foods makers of broadly similar scale - companies that also sell to consumers by name, also buy rather than grow their inputs, and also earn a brand margin rather than a conversion spread. Against that set, Kanva's multiple can be read honestly: a discount might genuinely signal mispricing, a premium might genuinely signal quality. Now watch what one wrong inclusion does. Suppose we drop in a bulk edible-oil refiner - same broad "Foods/FMCG" bucket, similar market cap, a name that any screener would happily return next to Kanva. It is a commodity converter: it buys crude oil, refines it, and sells it largely in bulk and private label, its margin a thin spread over a raw-material cost it cannot control, its balance sheet heavy with processing plant.

Nothing about Kanva changed between the two columns. Its gross margin is still ~44%, its multiple still ~23x, its business still a branded, asset-light consumer maker. What changed is the mirror we held it against. The refiner's structurally lower margin and lower multiple pulled the peer average down, and against that contaminated average Kanva suddenly reads as "expensive" - a false signal manufactured entirely by one wrong inclusion. A reader who trusted the second column would mark Kanva as over-valued and pass on it, or worse, conclude the refiner is the "cheap" way to play the same theme, when the two are not playing the same theme at all. The refiner is cheaper because it is a different, lower-quality-of-earnings business, not because it is a bargain version of Kanva.

Here is the inversion that makes this module belong in this guide. The stance is universal - build the peer set from the economic model - but the specific axis that decides comparability is different in every sector, and a reader who learns one sector's peer test as the peer test will build false sets everywhere else. The question "what makes these two companies genuinely comparable?" has a different answer for a bank than for a consumer brand, a commodity producer, or an IT services firm, and using the wrong answer is how a peer set that looks careful turns out to be worthless.

The inversion is sharpest at the lender, and it is worth seeing why. Tell a reader to build a peer set for a consumer company and they will instinctively reach for brand and category, which is roughly right. Let that same reader loose on a bank and the instinct betrays them: they compare on size and loan growth, because those are the visible, headline numbers - and a bank will happily show a briskly growing book at an attractive multiple while its funding franchise tells a completely different story. What makes two lenders comparable is the shape of both sides of the balance sheet: a retail bank funded by a cheap, sticky deposit base is not a peer for a wholesale lender funded by flighty bulk borrowing, however similar their size, their sector, or their price-to-book. The axis that decides it - the liability franchise - simply does not exist on a consumer company's statements, so a reader who never learned to look for it will build a "peer set" of banks that share everything except the one thing that matters. Carry the stance across sectors; relearn the axis for each. That is the whole reason a peer set that looks impeccable in one sector can be nonsense in another - it was built on the wrong axis.

The facts you need to build an honest peer set are structural, and they sit in the filings, not in the screener. Start with the company you want to value and reconstruct its position on the axes that dominate its economics - then go find the companies that share those positions, wherever the exchange happens to file them. illustrative

The segment note is the first place to look, because it tells you what the company actually is beneath its blended headline: which business lines earn the profit, how concentrated the revenue is, and whether a company the label calls a "peer" earns its money the same way. A company reporting 80% of profit from one segment is a focused play; a conglomerate reporting five segments is not its peer even if they overlap in one. The geography split, usually alongside the segment note, tells you whether two companies with the same product face the same demand and the same currency - an exporter and a domestic player in the identical category are only partial peers. The revenue-mix and product disclosures separate the branded from the bulk, the premium from the value, the high-value service from the commoditised one. For a lender, the disclosures on deposit mix and advances composition are where the loan-book mix and funding franchise live - the axes that decide bank comparability. And the fixed-asset and capital-employed lines tell you the asset intensity and, read with the notes on captive plants and raw-material sourcing, the degree of integration.

The habit to build is to assemble the peer set before you look at anyone's valuation, so the set is chosen for comparability and not, unconsciously, for the answer you were hoping to reach. Write down your company's fingerprint on the five axes; list candidate peers; and for each candidate, check its segment mix, its integration, its asset intensity and its go-to-market against your company's, admitting only those that match on the axes that matter and consciously noting where even an admitted peer differs. The output is not a long list - it is usually three to six genuinely comparable names, sometimes fewer, and a short, honest note of who you excluded and why. A small, clean peer set beats a large, contaminated one every time, because the whole value of the comparison depends on the members being real.

A correctly built peer set is necessary for honest relative valuation, and it is nowhere near sufficient. It locates the right comparison; it does not, by itself, deliver the verdict, and treating it as more than it is creates its own errors.

It cannot make the comparison itself valid if the numbers going into it have not been normalised. Two genuine peers can still be un-comparable on raw reported figures - different accounting choices, one-off items, a different year-end, an acquisition mid-year - and cleaning those distortions is the work of the next module, not this one. A perfect peer set fed dirty numbers produces a clean-looking, wrong answer.

It cannot tell you why a genuine peer trades where it does. When the peers are real and the numbers are normalised and a company still trades at a discount to its group, that discount is a question, not a conclusion - it may be a mispricing, or it may be the market correctly pricing a worse business, a governance problem, or a structural headwind the multiple is right to reflect. The peer set earns you the right to ask the question honestly; the answer comes later in Part Six, where the whole point is that the cheapest name is usually cheapest for a reason.

And comparability is a matter of degree, not a binary, so a peer set can be over-engineered as easily as it can be lazy. Insist on perfect matches on every axis and you will end up with a peer set of one - the company itself - which tells you nothing, because relative valuation needs a group to be relative to. The skill is judgement about which axes dominate this particular company's economics and matching hard on those while tolerating difference on the rest. A reader who rejects every candidate for some minor dissimilarity has not been rigorous; they have made the comparison impossible, which is its own way of getting it wrong.

The first and most common way is trusting the pre-built list. A stock screener's sector filter, an index's constituent list, a data vendor's "industry" tag, and the sell-side note's comp table all present a peer set that someone else assembled - and every one of them was built for a purpose other than valuation comparability. The screener groups by a coarse taxonomy; the index by size and free float; the vendor by a standardised classification; the sell-side table often by whatever is convenient and covered. Accepting any of them as your peer set is outsourcing the single most important judgement in relative valuation to a process that was never trying to make it. The defence is to treat every pre-built list as a source of candidates to test, never as a peer set to use.

The second way is comparing on market cap, which feels like a comparability axis and is not one. Two companies of the same size can be economically opposite - a large asset-heavy commodity producer and a large asset-light platform can share a market cap and share nothing else - and pairing companies because they are "similar size" is a subtler version of the same-label error. Size is worth adjusting for once the business models genuinely match; it is worthless as the basis for deciding whether they match.

The third way is letting one glamorous or one distressed name distort the group. A single peer trading at an extreme multiple - a market darling on an enormous multiple, or a broken company on a collapsed one - drags the group average and, through it, your verdict on the company you are valuing. The average of a peer group with one outlier is not a robust benchmark; it is a number one company is dictating. This is why you look at the dispersion first, why the median often beats the mean, and why an outlier is a prompt to ask whether it belongs in the set at all rather than a value to quietly average in.

The fourth, and the most seductive, is building the peer set to reach the answer you already want. If you like a company and want it to look cheap, there is always a higher-multiple "peer" you can include to lift the average; if you dislike it, there is a lower-multiple one to drag the average down. Because the peer-set choice is made early, feels clerical, and is rarely shown, it is the perfect place for a conclusion to sneak in before the analysis begins. The only guard is to fix the set on comparability grounds, in writing, before looking at valuations - and to keep the list of who you excluded and why, so that a peer set assembled to flatter a thesis cannot hide behind the authority of a table.