Fact vs Opinion in Research: Where the Line Actually Is

A fact is a statement a document can settle. An opinion is a judgement evidence can push either way and never settle. Sitting between them is the category that causes almost all the trouble. An inference is arithmetic or reasoning run on facts, and an inference quietly carries every assumption the reasoning needed. Research goes wrong when an inference is written in the grammar of a fact.

The verification pass a document takes before it goes out is covered separately: what gets verified, against which source, and in what order. The verification pass works beautifully on one kind of sentence and does almost nothing to another kind. A reader can check that Rs 1,111 crore appears in the statements. Whether management was too cautious cannot be checked in the same sense. Both sentences may sit in the same paragraph of the same note, set in the same typeface, ending in the same full stop.

Sorting has to happen before any checking is worth doing. Nine statements about Sarvani Coatings Limited, an invented maker of paints, get sorted into three categories, and two of them refuse to sit still. The interesting finding is not that some statements are opinions; it is that a statement's category is often decided by something outside the sentence, and the sentence never says so.

What makes a statement a fact here?

Not confidence. Not that it comes from an accountant. The test is narrow and mechanical: a statement is a fact when there is a document that settles it, and settling means that two people who disagree can both go and look and one of them has to change their mind. Gross profit was Rs 1,111 crore in year three. Open the statements, read the line, done. Nobody has to be persuaded.

Here is the part people get wrong. Doing arithmetic does not demote a statement out of this category. Gross marginWhat is left of every rupee of sales once the cost of the goods actually sold is taken out, before salaries, freight, advertising and everything else. Built and read in the accounting layer, not here. was 46.0 per cent in year three. The margin is Rs 1,111 crore over Rs 2,415 crore, both of them published on the face of the statements, and the computation is one division with nothing optional in it. There is no assumption to disagree with, so there is nothing for the reader to be handed unseen. The statement is a fact, and calling it an interpretation because a calculator was involved would leave an analyst unable to state almost anything.

A fact carries its own precision. The division actually gives 46.0041 per cent. Printed to one place the ratio reads 46.0 per cent, and that reading is true. Printed as forty six per cent it is still true, and coarser. Written as exactly forty six per cent it becomes false. The ratio is not exactly forty six. A fact is only a fact to the number of places it is stated to, and the word exactly is doing real work whenever it appears.

What makes a statement an opinion?

An opinion is a judgement that evidence can make more or less reasonable and that no document brings to an end. Management was too cautious in its guidanceWhat management tells the market in advance about what it expects, usually as a range or a direction rather than a promise. How a release is read against it belongs to the earnings sequence.. The charge of caution can be argued hard in both directions. Volume came in at 6.0 per cent against a guide of high single digit growth, so on the measure management chose to emphasise the year was at or just below the bottom of the range. Gross margin was guided to hold and rose 1.9947 points, printing as 2.0. Was that caution, or was it a company being sensible about a number it could not control? Nothing settles it.

The everyday version. A neighbour says her son is a careful driver. There is no document. Evidence can be gathered for years, insurance records, what the car looks like, what he does at a blind turn, and the evidence will move a listener without ever finishing the question. An opinion is not a weaker sort of fact and it is not a failure of rigour: most of what research exists to produce is opinion, and the professional obligation is to label it rather than to avoid holding one. A note made entirely of facts would be a photocopy of the filings and would tell a reader nothing they could not have read themselves.

What is an inference, and why does it cause the trouble?

Between the two sits the category almost nobody names. An inference is arithmetic or reasoning performed on facts, and it inherits every assumption the reasoning required. Management presents an adjusted EBITDAEarnings counted before interest, tax and the writing down of assets, so two businesses carrying different borrowings can be lined up against each other. Built in the accounting layer., or earnings before interest, tax, depreciation and amortisation, of Rs 452 crore for year three by adding back a Rs 6 crore restructuring charge. A note replies that the figure is Rs 4 crore too high. Every number in that reply is correct. The reply is still an inference. The test behind it removes the favourable Rs 4 crore write-backMoney set aside earlier for a cost that then did not arrive, released back into the accounts, so an expense line ends up smaller than it otherwise would have been. as well as adding back the unfavourable charge, and choosing that test was a decision somebody made.

Here is the property that makes inference dangerous: it arrives in exactly the grammar of a fact, a number and a full stop, so nothing in the sentence warns a reader that a choice went into it. Set the two sentences side by side and there is nothing to see. Same length, same shape, same confidence. One of them can be checked against a line in a filing. The other cannot be checked at all until the reader knows which test was run, and the sentence does not say.

How should each of the three be written?

The sorting stops being philosophy here and becomes a writing rule that can be applied on a Tuesday afternoon. Each category needs a different sentence shape, and the shape is not decoration. A reader who wants to settle a fact needs the document it came from and the period it covers, so a fact is written with both. A footnote is read by nobody and a sentence is read by everybody, so an inference is written with the assumption it required, in the same sentence rather than in a note underneath. An opinion is written with whose it is and what would change it.

The third shape is the whole point: a reader can only disagree with what has been labelled, so labelling is not a courtesy to the reader but the condition that makes disagreement possible at all. Taking the third shape seriously does something useful to the analyst's own work. Writing down what would change the view, before anybody argues against it, is the hardest sentence in a note to write and the one that shows whether what is held is a view or a preference.

A statement about somebody else's statement

Now the case most treatments skip entirely. Management said it expected gross margin to hold around the prior year level. Which category does that sentence belong to? The sentence is a fact, and a checkable one: the words were said, they sit in the transcriptThe word by word written record of a company call, published afterwards, so what was said can be checked rather than remembered or paraphrased. or in the release, and anybody who disputes it can be shown the line. The sentence records a fact about a statement.

But look at what it is evidence about. The sentence is evidence about management's expectation. The sentence is not evidence about margin. Margin went on to rise from 44.0094 per cent to 46.0041 per cent, a one year gain of 1.9947 points, and the guidance sentence stayed exactly as true as it had been on the day it was spoken. The two get merged constantly, and a note that reports a company's characterisation of its performance as though it were an observation about the business has changed the category without changing a single word.

The household version is immediate. An uncle says the wedding will cost about four lakh. The fact here is that he said it, and a recording would prove it. The cost of the wedding is a separate question, and his sentence settles almost nothing about it beyond what he currently believes. Reporting the first as though it settled the second is the oldest move in the book, and in a research note it looks completely respectable.

Conditional facts, where the arithmetic is exact and the event never happened

Suppose Sarvani Coatings Limited ran a buybackA company using its own cash to purchase and cancel some of its shares, so fewer of them are left outstanding afterwards. of Rs 240 crore at Rs 600/- a share. The buyback takes out 0.40 crore shares from 24.00 crore, leaving 23.60 crore. Profit after tax of Rs 278 crore spread over the smaller count gives earnings a share of Rs 11.7797/- against Rs 11.5833/- before, so the printed figure moves from Rs 11.58/- to Rs 11.78/-. Take that same Rs 240 crore out of net worth of Rs 1,486 crore, leaving Rs 1,246 crore, and book value a shareThe accounting worth of the business after everything it owes has been taken out, divided by the number of shares. Built in the accounting layer. falls from Rs 61.9167/- to Rs 52.7966/-, printing as a fall from Rs 61.92/- to Rs 52.80/-.

Every one of those divisions is exact. There is no rounding trick, no assumed growth rate, no view about anything. And none of it has happened. A conditional fact is a fact about the arithmetic and a hypothesis about the world. The condition is the only thing standing between a worked illustration and a claim about a company, and the condition has to travel with the sentence every single time the sentence appears.

Notice what else is riding along, unmentioned. The earnings a share figure holds profit after tax at Rs 278 crore. Holding it there quietly assumes the Rs 240 crore of cash was earning nothing before it left. The record does not break other income down by which asset produced it, and that assumption cannot even be tested here, let alone corrected. A second condition rides in the balance sheet: cash and investments of Rs 312 crore less Rs 240 crore leaves Rs 72 crore against borrowings of Rs 240 crore, so a net cash position of Rs 72 crore would become net debt of Rs 168 crore. State the buyback arithmetic without the stamp and a reader inherits all three of those at once.

Can a fact stop being a fact?

The line is not fixed anywhere. Statement three of the nine says that materials cost, measured against each unit produced, rose about 3.6 per cent in year three. Work it. The cost of materials went from Rs 1,187 crore to Rs 1,304 crore, a rise of 9.8568 per cent. Revenue went from Rs 2,120 crore to Rs 2,415 crore, a rise of 13.9151 per cent. Volume rose 6.0 per cent, and that figure is published. Strip volume out of each and input cost per unit rose 3.6385 per cent while realisation per unit rose 7.4671 per cent. Nothing was assumed. Every input was disclosed. The statement is a fact.

Take away one line of disclosure, the volume growth, and the identical sentence becomes an inference resting on an assumed volume, and the size of the assumption is enormous: on a different volume figure the per unit answer moves to wherever that figure puts it. The category therefore belongs to the record rather than to the phrasing. A writer has to know where each number came from, and a reader who cannot see the source cannot do the sorting at all.

One caution, and it is an easy trap to walk into. The cost share of revenue fell from 55.9906 per cent to 53.9959 per cent. Pushing that year two share along on the two per unit growth rates arrives at the year three share exactly, to every place. Feeding the same identity the printed 56.0 instead of the true 55.9906 lands it on 54.0050, a residual of 0.0091 of a point that still prints 54.0 and no longer ties. And the tie itself is not a check of anything: volume cancels out of the ratio, so the identity is forced arithmetic and would hold whatever volume did. Two lessons in one line. A printed display is never an input, and an identity that cannot fail is not evidence.

Sorting the nine

Here is the whole set, sorted, with the reason each one lands where it does. The last column carries more than the middle one: the reason is what would have to be argued with, and the label is only a summary of the reason.

Why does the Rs 4 crore claim feel exactly like a fact?

Statement eight is the one that fools trained people. Reported EBITDA for year three is Rs 446 crore, a margin of 18.4679 per cent. Put the Rs 6 crore restructuring charge back on, as management does, and the presented figure is Rs 452 crore, 18.7164 per cent. The Rs 4 crore write-back that flattered the very same line in the opposite direction stays where it is. Run the adjustment symmetrically instead, charge back on and write-back taken off, and the result is Rs 448 crore, 18.5507 per cent. So the claim that management is Rs 4 crore too high is correct, and it is correct on a test the sentence never names, worth 0.1656 of a margin point, printing as 0.17, and 0.89 per cent of the two way figure.

Three figures are all defensible and the sentence quietly picks one. Picking one unannounced is what makes the claim an inference wearing the grammar of an observation. Somebody applying the adjustment one way, adding back the charge and leaving the write-back alone, gets Rs 452 crore and is not making an arithmetic mistake. Such a person is running a different test. Until the test is in the sentence, there is no way for a reader to know that a choice was made, let alone which choice.