Variant Perception: Where Your View Differs From Consensus

A variant perception is a view that differs from what other people already assume, stated together with a measurement of how far it differs. Nine estimates for Sarvani Coatings Limited, an invented paint and coatings maker, ran from Rs 255 crore to Rs 284 crore of profit, so a view of Rs 280 crore sits inside ground somebody already occupies and is not variant at all, however strongly it is held.

The idea rests on three things already established. A consensusThe compiled set of published estimates for a company, gathered and averaged by a data provider. A consensus is a collection of separate opinions, not one opinion. is a compiled set of estimates with a middle and two ends, settled in the earnings work and used here rather than rebuilt. The reverse arithmetic that runs from a share price back to the growth the price contains was settled in the valuation work, and it supplies the second and better reading of what everybody else is working from. The thesis drivers already named the three variables the worked claim about Sarvani Coatings rests on. The measuring stick itself is the new part.

What is a variant perception, and which half of it do people skip?

Two halves, and almost everybody keeps the first one. The first half is the view itself: profit will be higher, or the margin will not hold, or the new line will take three years rather than eighteen months. The second half is the measurement: what other people are working from is written down, in numbers, and the distance between their figure and the analyst's is stated in the units they are using.

A difference nobody has measured cannot be acted on and cannot be checked later, so the second half is what separates a variant perception from an opinion. Think about a household deciding whether a wedding will cost more than the neighbours say. Saying it will cost more is a feeling. Saying that caterers in this locality quote about Rs 1,100/- a plate while the budget is Rs 1,450/-, a quarter more, is a measured difference. When the bills arrive, exactly one of those two statements can be marked. The other one can be argued about for years.

Research works the same way. An analyst whose figure for Sarvani Coatings is Rs 280 crore of profit after tax, and who has never written down what anybody else expects, does not have a variant perception. A figure with nothing beside it is a number and a mood. The moment the other side is written down, three things happen at once: whether there is a difference at all becomes visible, so does its size, and so does whether the gap is large enough to be worth anything.

What exactly is the difference measured against, and how much of it is visible?

The comparison is harder than it looks, and the honest answer is that only part of what people assume is ever visible. The obvious candidate is the compiled set of published estimates. The published set is real, it is quotable, and it has a middle and two ends that can be pointed at. The set is also thin. A compiled set of estimates covers one period, a small number of estimators and a handful of lines, so the set is only a partial view of what people assume, and almost everything a price reflects sits outside it.

Look at what the Sarvani Coatings set actually contains. Nine estimators. One line, profit after tax. One period, the year just ended. The set says nothing about what anyone assumes for year five, nothing about the exit rating anyone applies, nothing about the volume path, and nothing at all about the people who bought and sold the share without publishing anything. Nine notes is not the market.

The less comfortable and better answer is that the price itself carries an assumption set, and reading it backwards gives a fuller picture than any published estimate can. The reverse arithmetic was built earlier in the valuation work and is applied here rather than rebuilt: the analyst supplies a required returnA yearly return figure the analyst sets before any arithmetic begins. Nobody can look it up, and two readers will pick different ones. and an exit multipleThe assumed price buyers will pay for each rupee of earnings on the day of sale. A forecast of future opinion, and the most movable input in the whole calculation., and out comes the implied growthHow fast earnings would have to rise for a quoted price to make sense, given two inputs the analyst supplies. Implied growth falls out of the arithmetic and is printed nowhere. the price appears to contain. The backward reading covers every year to the exit, it covers everybody who traded rather than nine people who wrote, and it is available on any day.

Why is a view inside the range not variant, however firmly it is held?

Because a compiled set is a distribution and not a number. The word consensus makes it sound like one figure that everybody agreed on in a room. Nobody agreed on anything. Nine people wrote nine numbers, a data provider stacked them up, and the middle of that stack got a name. Somebody in the set is already standing on that ground, so a view that sits inside the existing range is not a variant perception, no matter how firmly it is held.

For Sarvani Coatings the set ran from Rs 255 crore to Rs 284 crore of profit after tax. The rangeThe distance from the lowest to the highest figure in a compiled set. The range describes how far apart the estimators were. The middle of the set hides that completely. is therefore Rs 29 crore wide, which is about 10.8 per cent of the mean of Rs 268 crore. Rs 29 crore is not a tight huddle. The spread is wide enough that a reader who puts profit at Rs 280 crore and calls that a differentiated view has said something one of those nine estimators had already put in print, with Rs 4 crore of room still above.

The second trap in the same set is the middle itself. The meanThe arithmetic average of a compiled set. A summary of nine separate positions. Not one of the nine need have written that figure down. of Rs 268 crore is a summary of nine positions, and it is entirely possible that not one estimator wrote Rs 268 crore. The medianThe middle value once a set is sorted, with as many figures above it as below. Less disturbed by one extreme estimate than the average is. has the same property. So when a report says the company beat consensus, it means the outcome landed above an average of positions. Landing above an average is a different and much weaker statement than surprising the people who held those positions.

How is the difference measured rather than felt?

Three steps, in this order, and the order matters because taking one's own figure first invites shopping for a comparison that flatters it. The other side is written down. The analyst's own figure is stated. The distance is expressed in the units the other side used.

A measured difference has to pass two separate tests: it must be large enough to matter and specific enough to be settled, and most differences fail one of those rather than both. A difference of half a crore against a mean of Rs 268 crore is about 0.19 per cent, or roughly two paise a share. Two paise a share is perfectly specific and it will be settled precisely. The rounding in the analyst's own model is bigger than the gap, so the difference is still worth nothing. A difference amounting to a belief that the company is better run than people realise fails the other test: it may be enormous, and there is no year in which anybody can mark it.

One warning about units. A difference should not be built by subtracting one printed percentage from another. The record prints the year three effective tax rate as 25.1 per cent, but the two absolutes underneath it are a tax charge of Rs 93 crore set against profit before tax of Rs 371 crore, and that division gives 25.0674 per cent. Every printed percentage is a rounding of something, and differences built from roundings inherit them and then look precise. The arithmetic runs on the absolutes twice, and the two results are differenced.

What does the measurement look like run twice on one company?

The same claim gets measured twice below, first against the nine published estimates and then against the assumptions inside the price, and the two measurements do not say the same thing.

First attempt, against the compiled set

The set covers year three profit after tax and nothing else. Nine estimates, Rs 255 crore to Rs 284 crore, mean Rs 268 crore. On an assumed 24.00 crore shares that mean is Rs 11.17/- a share. The record does not publish a share count, and the assumed count is worth checking forward rather than back. Rs 278 crore over 24.00 crore shares gives Rs 11.58/- a share and Rs 1,486 crore of net worth over the same count gives Rs 61.92/-, and both of those are the published per share figures. Two independent checks landing together is what makes the assumption usable.

Now the two lessons the set actually teaches, and they arrive together. A result inside a range that already existed is not a surprise in any useful sense, however the beat is reported afterwards. One estimator in the set was already above Rs 280 crore, so a view of Rs 280 crore held before publication would not have been a variant perception. The published Rs 278 crore beat the mean by Rs 10 crore, about 3.7 per cent, and cleared seven of the nine estimates. The published figure also landed Rs 6 crore below the top of the set. Somebody wrote that number down in advance and was not surprised at all.

Second attempt, against the assumptions in the price

The second measurement is better, and it is also the one that needs its limitation stamped on it. The illustrative price is Rs 486/- against published earnings of Rs 11.58/- a share, or 41.9689 times. The multiple is the starting point rather than the finding. Under two assumptions the reader supplies, a required return of 12 per cent and 25 times earnings in five years, the Rs 486/- carries an assumption of roughly 24.2 per cent earnings growth in each of five years. The route is the one the valuation work built: Rs 486/- compounded five years at 12 per cent is Rs 856.50/-, that divided by 25 needs Rs 34.26/- of earnings a share in year five, and climbing from Rs 11.58/- to Rs 34.26/- across five years needs roughly 24.2 per cent compounded each year.

The worked case puts volume at 6.0 per cent for the year and realisation at 3.0 per cent. Volume and realisation compound rather than add: 1.06 times 1.03 is 1.0918, so revenue grows about 9.18 per cent, or 9.2 per cent to one decimal, and not the 9.0 per cent that adding the two rates would give. The difference is about 15.0 percentage points a year. Both sides of that subtraction were computed unrounded and then differenced. Differencing the two printed figures instead would have matched here, and matching here is luck rather than method.

The 24.2 per cent measures earnings and the 9.18 per cent measures revenue, and those are not the same quantity, so there is a bridge to state before the subtraction is allowed to stand. Subtracting one from the other is allowed only where the margin stays put and no line beneath earnings before interest and tax (EBIT) does anything, and asserting precisely that is what it means to call the margin gain a level shiftA step up that happens once and then holds where it landed. Set a level shift against something that repeats. A repeating gain compounds into a far larger number by year five. rather than a rate. State the bridge, or a reader who checks is left staring at an apparent unit mismatch.

Now the limitation, stated as plainly as the finding. The second measurement is against a price and against two inputs the reader chose. No part of it is a measurement against anything anybody said. Writing the gap up as a 15 point difference from what analysts think would be a misattribution. The size of the analyst's own contribution is easy to see. Hold the price and the required return, change only the exit rating, and at 15 times the same price carries roughly 37.6 per cent a year, a gap of 28.4 points. At 35 times the same price carries roughly 16.1 per cent, a gap of only 7.0 points. The company did not move between those three readings. The analyst did.

What makes a difference worth holding rather than merely different?

Not boldness, and not the size of the number. A difference is worth holding only when there is a reason the other side does not have: a mechanism they have not traced, a period they have not covered, or arithmetic they have not run. Absent one of those three, a difference is a guess with a decimal place attached, and the decimal place is doing the work that evidence should be doing.

The everyday version is a street vendor deciding to buy an extra sack of onions. A reason of prices feel like they are rising is a guess. A reason of the wholesale market shutting for two days next week for a festival, with the shopkeepers nearby not having checked the calendar, is a period the other side has not covered, and it settles itself in nine days. Same conclusion, entirely different quality of claim.

On the worked case the candidate reason is the third kind. The published statements show gross margin rising over one year, year two to year three, while input cost per unit of output also rose about 3.6 per cent. The whole of the gain therefore came from realisation outrunning input cost. Anybody can read that. Fewer people run the arithmetic on what happens if the realisation gap does not repeat. Arithmetic on a realisation gap that does not repeat is the sort of work not everyone has done, and a difference can honestly rest on it.

How is a variant perception told apart from being different on purpose?

By which came first, the view or the evidence. A variant perception is what remains after the work: the company was examined, a view was formed, what other people assume was written down, and the view turned out to sit somewhere else. Being contrarianHolding a position mainly because other people do not hold it. The word describes where a view sits relative to a crowd, not how the view was reached. on purpose runs the same steps in reverse: what other people assume is read first, a spot away from it is picked, and support for that spot is then found.

Choosing a view because it is unlike other people's inverts the process. The evidence then gets selected to support a position that was already chosen, and the claim quietly stops being about a company and turns into a claim about a crowd. The tell is what happens to disconfirming evidence. A view formed from the work has room to be wrong, and contrary evidence is information. A view chosen for its unpopularity has no such room. Abandoning it means rejoining the crowd, and the crowd was the thing being avoided.

The notes themselves carry the give away. If the sentences are mostly about the company, the mechanism and the numbers, the process ran forwards. If the sentences are mostly about what the market has missed, what everyone is ignoring and how nobody is looking at this properly, the process ran backwards, and the company has quietly become a supporting character in a story about other people.

What happens when everybody arrives at the same view?

The difference disappears and the view stops being variant. The whole purpose of a measured difference is that it should eventually be settled. Being joined is one of the two ways the settling happens, so it is the point rather than a loss. The other way is that the view turns out to be wrong, which is also settled and also useful.

The sequence is worth watching in slow motion. The view starts Rs 12 crore from a mean nobody actually holds. Evidence arrives, others revise, the mean travels towards it, and the span narrows around what used to be a lonely corner. On the day the middle of the set reaches that figure, the measured gap is nil, and all of the value in that gap has been collected. Nothing has gone wrong. A shop that opened first in a new locality does not consider it a failure when the locality fills up.

A research process that needs its conclusions to stay unpopular is the one thing all of this rules out. A process like that has confused the difference with the claim. The difference was only ever the measuring instrument, showing how much of the view was not already in the price; it was never the thing believed about the company.

What if the difference is not in a number at all?

The awkward and interesting case is a difference with no number anywhere to hang it on, and the worked example on Sarvani Coatings lives there. Attribution rather than a forecast is what divides the two sides here, and attributionWhich cause produced an observed movement. Two people can agree completely on what happened and disagree entirely on what caused it. is the harder half: did the gross margin step up once, or is it still climbing? Nobody publishes an estimate of that. No compiled set anywhere carries a line for the durability of a margin gain. So there is nothing to be above or below, and the measuring stick has nothing to rest on.

The fix is conversion. Write both positions as numbers on the same base and the difference becomes measurable immediately. A level shift means gross margin holds where it is, at 46.0 per cent, and earnings grow with revenue at about 9.18 per cent. A durable rate means the margin keeps climbing, and the honest version of climbing further is that the one year gain of 2.0 points, year two to year three, happens again, taking the margin to 48.0 per cent. Published year three revenue of Rs 2,415 crore grown at about 9.18 per cent gives year four revenue of Rs 2,636.7 crore, and putting both positions on that one revenue produces different gross profit.

Rs 52.7 crore is about Rs 2.20/- a share before tax, against a published year three gross profit of Rs 1,111 crore. Rs 52.7 crore is a real number, it lands in a stated period, and a published year four settles it. Keep the period label attached. A dropped period label causes more errors here than anything else. The gain is 2.0 points over one year, year two to year three, and 3.0 points over two years, year one to year three. Only the one year figure is the candidate for repeating, and it is worth seeing where a repeating rate ends up. Five more years of 2.0 points takes gross margin to 56.0 per cent. A destination like that makes a reader ask the durability question without being told to.

One more observation tightens the question without settling it. Across the two years from year one to year three, gross margin gains ran 3.0 points at Sarvani Coatings, 2.4 points at Nandivarman Paints Limited and 3.6 points at Kesaria Surface Solutions Limited. All three rose. Three makers rising together points at conditions across the whole field rather than at anything one maker did. But Sarvani Coatings sits between its two peers, so nothing there separates its own pricing from a shift in what it sold. Evidence that narrows a question and refuses to close it is the normal case, not a disappointing one.

How does anyone actually use this on a working morning?

Three people use the same measurement for three different jobs, and watching them makes the idea concrete.

An analyst uses it as an editing rule before a note goes out. The rule is simple: if the note contains the words differentiated, contrarian or out of consensus, then it must also carry a table with the other side in it and the distance stated in units. If the table cannot be built, the words come out. The claim and its evidence now sit in the same document, so that one rule removes most of the unmeasured claims from a research process.

A portfolio manager uses it for sizing rather than for selection. One idea differs from what the price assumes by 15 points a year and another differs by half a point, so two ideas can both be right and be worth very different amounts. The half point idea can be completely correct and still return nothing. The price already contains it. A measured difference sizes how much of a view is not already paid for, not how right the view is.

A household investor uses it as a filter on what they are being sold. When a note says a company is misunderstood, the question to ask is what number the note thinks other people are working from, and where that number is written down. If the answer is a compiled set of estimates, ask over what period and for which line. If the answer is a price, ask which required return and which exit rating produced it. A reader can change the answer from 37.6 per cent to 16.1 per cent without touching the company at all. A claim of difference that cannot answer either question is a claim about the writer.

Sarvani Coatings Limited, Nandivarman Paints Limited and Kesaria Surface Solutions Limited are invented.
Educational material. Not advice on any investment, tax, budget or market position.