Normalising before comparing

Put two competitors side by side. They make broadly the same product, sell into the same market, and file under the same accounting standard. One reports a 20% EBITDA margin, the other 16%. Four hundred basis points is a large distance in most industries — the difference between a business you would call excellent and one you would call ordinary — and the temptation is immediate and strong: the first company is simply the better operator, and the comparison is over before it has begun.

It is not over. It has not started. Because the two margins were not built the same way, subtracting one from the other measures how the companies report at least as much as how they trade. The first firm leases its plants, which under current lease accounting pushes a cost that used to sit inside EBITDA down below it; the second owns its plants outright, so no such lift exists. The first books the interest on its cash pile inside its operating line; the second reports it separately, as it should. The second absorbed a genuine one-off write-down this year that the first did not. Peel those apart and the 400-basis-point chasm narrows to almost nothing — the two businesses, underneath, earn about the same. illustrative

This is the single most common error in relative value, and it is committed by careful people every day: comparing raw reported numbers across companies as though the numbers were built to be compared. They are not. A reported figure is the end of a long chain of choices — which costs to expense and which to capitalise, how long to depreciate an asset over, whether to lease or own, where to park other income, when the year happens to end — and two companies make those choices differently. Until you have undone the differences and rebuilt both figures on one basis, every peer comparison you draw is a comparison of accounting policies wearing the costume of business quality. Normalisation is the work of undoing them, and it is the precondition for everything else in this part of the book.

Normalisation is not a single adjustment. It is a checklist you run on both companies before you let their numbers touch each other, and each item removes one source of false difference. There is nothing clever in it — it is mechanical, unglamorous, and skippable, which is exactly why it gets skipped and why the reader who does it has an edge over the reader who does not.

Strip the one-offs. A year is not a run-rate. A large land sale, a restructuring charge, an insurance receipt, a write-down — each pushes a single year's profit above or below what the business sustainably earns. The exceptional items line is where the honest ones are flagged, but many are buried in ordinary cost lines and must be dug out of the notes. You add back the one-off losses and strip out the one-off gains on both companies, so that what you compare is the recurring engine, not the year each happened to have.

Align the accounting policies. This is the subtle work, because the differences are legal, disclosed, and invisible unless you go looking in the accounting-policy note. Two firms can report different profits from identical operations because one assumes a longer useful life for its assets and so charges less depreciation each year; because one practises heavier capitalisation, parking costs and interest on the balance sheet that the other expenses through the P&L; because one values inventory on a different basis, which moves the cost of goods when input prices are rising; or because one leases the assets the other owns, which under Ind AS 116 shifts the cost out of operating profit and into depreciation and interest below it. None of these is manipulation. All of them make the raw numbers incomparable.

Align the year and the cycle. Companies do not all close their books on 31 March, and even when they do, they may sit at different points of the same cycle — one reporting a peak year, the other a trough, in an industry where that swing is structural. A margin gap that is really the distance between one firm's good year and another's bad year is not a quality gap at all. Most short-term movement in a number is randomness wearing the costume of information — and a single year's peer spread, read as though it were permanent, is often just the two companies caught at different phases of the same wave.

Hold standalone and consolidated consistent. Compare like with like: one company's consolidated accounts against another's standalone will fold whole subsidiaries — their revenue, their debt, their margins — into one side and not the other. Decide which basis answers your question and use it for every peer, or you are adding a business to one company that the other does not carry.

Treat other income the same on both. Other income — interest on cash, dividends, gains on investments — is not operating profit, and where a company chooses to present it can flatter or depress the operating margin. Pull it out of both companies' operating lines and, if it matters, judge the treasury separately from the business.

The abstraction becomes concrete the moment you draw the bridge. Take the two composite peers from the opening — call them Larkspur and Marigold, two mid-cap makers of the same industrial product. Larkspur reports a 20.0% EBITDA margin, Marigold 16.0%. Now walk each number back to a common basis. illustrative

Larkspur's margin is lifted by two things that have nothing to do with how well it makes its product. It leases its plants, so under lease accounting the rental cost sits below EBITDA as depreciation and interest rather than inside it — worth about 150 basis points of headline margin that Marigold, owning its plants, does not get. And it books treasury income inside its operating line, worth another 50. Strip both and Larkspur's comparable margin is 18.0%. Marigold, meanwhile, is depressed by a genuine one-off write-down of about 150 basis points, and its year ended in a seasonally weak quarter that cost roughly another 50 against Larkspur's stronger close. Add those back and align the cycle, and Marigold's comparable margin is also 18.0%. The 400-basis-point gap was never a business gap. It was four accounting and timing differences stacked in the same direction.

Here is the checklist itself, in the order you would run it, with what each item hides and where to find it in the report. Note the last column: several adjustments touch both the numerator and the denominator of your ratios, which is the reminder to normalise both sides.

The checklist is universal, but the one adjustment that matters most moves from sector to sector — and this is the inversion. Run the same instruction, "normalise before comparing," on four different industries and it points you at a completely different line each time. A reader who learns one sector's decisive adjustment as the adjustment will scrub the wrong thing everywhere else: fussing over a retailer's other income while ignoring its leases, or stripping an exporter's one-offs while leaving the currency untouched. The posture is constant; the line it lands on is what you must relearn for each business.

The inversion, stated plainly: the same metric — say EBITDA margin — is inflated by leases in retail, by capitalised interest in infrastructure, by a weak rupee in exporters, and by treasury income in cash-rich IT, and the adjustment that fixes it in one sector does nothing in another. Tell a reader to normalise a retailer and the whole weight falls on the lease treatment; tell them to normalise a half-built infrastructure company and it falls on capitalised interest and CWIP; tell them to normalise an exporter and it falls on constant currency; tell them to normalise a cash cow and it falls on other income. Carry the checklist, but ask first, in every new business: which single line, left un-normalised, would most distort a comparison here? That line is where you start.

Doing this on a real pair of filings is slower than it sounds and quicker than you fear, and it follows a fixed order. illustrative

Start with the accounting-policy note in each annual report, side by side. This is the single most valuable page for normalisation, and almost nobody reads it comparatively. It states the depreciation method and asset lives, the inventory valuation basis, the lease and capitalisation policies — exactly the differences that make raw margins incomparable. Where the two companies differ, you have found an adjustment to make; where they agree, you can compare that line directly. Write down each difference before you touch a number.

Next, hunt the one-offs. The exceptional items line catches the flagged ones, but the material ones are often folded into ordinary cost lines and revealed only in the notes or the management discussion — a write-down inside cost of materials, an insurance receipt inside other operating income. Read the two companies' notes for the same year and list every item that will not repeat, on both sides, adding back losses and stripping gains so the run-rate is clean for each.

Then check the basis and the calendar. Are you holding one firm's consolidated numbers against another's standalone? Do the year-ends align, and if the industry is cyclical, are both companies at the same phase? A steel or sugar peer comparison drawn between one firm's peak year and another's trough is measuring the cycle, not the companies. And for the specific sector, add the adjustment that carries the weight — the lease treatment for the retailer, the currency for the exporter — as the last and largest step.

Finally, rebuild both ratios end to end. Recompute the margin on the normalised profit, and — the step most readers skip — recompute the return on the normalised capital, stripping CWIP and surplus cash from the base so the denominator is as clean as the numerator. Only now, with both companies on one basis on both sides of every ratio, is the spread you see a spread you can trust. Everything before this was preparation; this is the comparison.

Normalisation makes numbers comparable; it does not make them true, and it does not make the comparison the answer. Its limits matter as much as its power.

It is itself a set of judgements, and judgements can be wrong or self-serving. Deciding what counts as a one-off, which asset life is "right," how to restate at constant currency — each is a choice, and two honest analysts can normalise the same accounts to slightly different numbers. The discipline is to be transparent about every adjustment and to apply the same test to every company, so that your normalisation is consistent even where it is uncertain. The danger is the opposite: over-normalising until you have scrubbed away a real difference. If one firm genuinely runs its plants harder, keeps its costs structurally lower, or earns a durable currency advantage, adjusting that away in the name of comparability erases the very edge you were trying to measure. Normalise the accounting artefacts; keep the business reality.

It also cannot repair bad data. If a company's underlying numbers are misstated — profit inflated, a cost hidden, a receivable that will never collect — normalising them produces a clean, comparable, wrong figure. Normalisation assumes the reported numbers are honest and merely built on different policies; where that assumption fails, the forensic work of the earlier parts comes first. A beautifully normalised comparison of two sets of accounts, one of which is fabricated, is worse than useless, because its tidiness lends the fabrication credibility.

And a normalised comparison, however careful, tells you only how two businesses stand relative to each other today. It does not tell you which will compound, which management will allocate the next decade's cash well, or what either is worth. Relative value is a lens, not a verdict; it narrows the field and frames the question. The judgement of quality, durability and price is still yours to make, and the next modules take it up.

The first way people are fooled is by comparing headline numbers straight off the screen — reported margin against reported margin, one firm's ROCE against another's, trailing P/E against trailing P/E — as though a shared label guaranteed a shared basis. It does not. The numbers look comparable precisely because they carry the same name, and that surface similarity is the trap: the label is identical and the construction underneath is not. A reader who ranks a screen of peers on raw EBITDA margin has ranked their accounting policies and cycle positions, and mistaken the ranking for a league table of quality.

The second is accepting management's own normalisation without checking it. "Adjusted EBITDA," "underlying profit," "normalised earnings" — companies present pre-cleaned numbers, and the cleaning is done by the party with the strongest interest in the flattering version. The commonest abuse is the recurring "exceptional": a restructuring or impairment charge added back every single year, which by its recurrence is a structural cost, not a one-off. Use management's adjusted number as a starting hypothesis to test against the record, never as a fact — and strip the same items yourself, on the same rule, for every company.

The third is normalising the numerator and forgetting the denominator, which we have met twice already because it is that common. You clean the margin, the peers converge, and you stop — leaving the capital base, the share count, the enterprise value still built differently underneath your return and valuation ratios. A converged margin on an un-normalised denominator is a half-finished comparison that feels complete, and feeling complete is what makes it dangerous.

The fourth is the standalone-versus-consolidated mismatch, the quietest of the four. Pull one company's consolidated revenue and profit and set them against another's standalone, and you have added a portfolio of subsidiaries — their sales, their debt, their margins — to one side of the comparison and not the other. The numbers will look like peers and describe different-sized animals. Always confirm you are on the same basis before the first subtraction.