Evidence in Finance Work: What Counts and What Does Not

Evidence in finance work is anything that could be handed to a stranger to show why a claim is true. Evidence has four grades. A fact is filed or observed. An assumption is one somebody chose. An inference is one the analyst drew. A scenario is one the analyst constructed. A claim that does not name its grade is not evidence yet, whatever it is built from.

Every number in finance work arrives by one of four routes, and the four routes carry completely different weight under challenge. Once they are mixed into one paragraph the reader cannot tell which sentence would survive being questioned, so the whole paragraph quietly inherits the weakest grade inside it. Grading is not bureaucracy; it is what stops one guess from contaminating four facts.

The Kavery research desk, four people inside the invented lender Kavery Capital Services Private Limited, was asked a question about Meenakshi Tubes Private Limited, a borrower the desk already lends to. Sharada Iyer produced the work, Prakash Nadar reviewed it, and Latha Menon commissioned it and made the decision at the end. The file was opened on 3 March, delivered on 11 March and decided on 12 March.

The craft is assembling and recording evidence, not the thing the evidence happens to be about. Substitute a bond, a fund, a warehouse or a single machine for Meenakshi Tubes Private Limited and not one sentence below needs rewriting. Four grades, one answerable question, a ranking that settles a clash between two documents, and a judgment written down where somebody can audit it six months later: that is the whole of the craft.

What counts as evidence in finance work?

EvidenceAnything a stranger could check without asking its author for help. is anything that could be handed to a stranger to show why a claim is true, without having to hand over its author as well. Not having to hand over the author is the whole test. If the only way somebody can verify a sentence is by phoning its author and asking what was meant, the sentence is not evidence. A sentence like that is a claim with a person attached to it, and people leave, forget and change jobs.

Start with something small enough to picture. A household argues every winter about whether the electricity bill has gone up. One person says it feels higher. The other pulls out a folder with twelve months of bills in it. The feeling is not evidence, not because it is wrong but because nobody else can pick it up and reach the same sentence from it. The folder is evidence. Anybody in the household can now argue about what the bills say rather than about what the winter felt like.

Evidence is not a kind of document; it is a property a document has when somebody else can get from it to the sentence it supports. Three things make that property hold. The item exists outside the analyst's head, so it can be handed over. The item is identified precisely enough to be found again. Precisely means a name, a numbered sheet and a position rather than a gesture at a folder. And the item is dated. Anybody reading it then knows which period it describes. The third is the one that fails, and it fails often enough that most disputes about evidence turn out to be disputes about which period a number describes.

How to Separate Fact, Assumption, Inference and Scenario in Finance Writing: What Are the Four Tests?

Four grades cover everything that will ever appear in finance work, and each grade is settled by one question that takes about two seconds to ask. Was it filed, observed or measured by somebody other than the writer? Then the sentence is a fact. Did a person choose it? Then the sentence is an assumption. Did the writer derive it from two or more facts? Then the sentence is an inference. Was it built on purpose to test a boundary? Then the sentence is a scenario. The grade is decided by where the sentence came from, never by how confident the writer sounds.

Grade one, fact: what has been filed, observed or measured?

A fact is reproducible by somebody who has never spoken to the writer. Revenue of Rs 21,20,00,000 and operating profit of Rs 2,30,00,000 for Meenakshi Tubes Private Limited are facts because they sit in the filed accounts for the year just reported, and any person with the filing in front of them lands on the same two numbers. The operating margin of 10.8 per cent is also a fact, in exactly the same way. Dividing one filed number by another filed number adds no choice of the writer's to the result. Arithmetic on facts stays factual as long as nothing was selected on the way through.

Grade two, assumption: what did somebody choose?

An assumption is the point where a person entered the work. Nobody files a split between fixed and variable cost, so the Kavery research desk decided that sixty per cent of the Rs 18,90,00,000 total cost moves with revenue and forty per cent does not. The split puts Rs 11,34,00,000 in one column and Rs 7,56,00,000 in the other. The split may be sensible, defensible and consistent with everything else in the file, and it is still a choice. The thing to record about an assumption is not that it is reasonable but that it was chosen, and by whom. A careful person choosing carefully does not upgrade an assumption into a fact, and the assumption register exists precisely so that the choosing stays visible after the person has moved on.

Grade three, inference: what did the writer draw from the facts?

An inferenceA conclusion drawn by the writer from two or more facts. is a sentence the writer manufactured out of facts the writer did not manufacture. Volume at Meenakshi Tubes Private Limited rose 21 per cent while revenue rose 15.2 per cent, and both of those come from items in the source log. More units sold for proportionally less money leaves no other arithmetic available. Put the two facts together and realisationWhat each unit actually fetched, after discounts and mix. per unit must have fallen about 4.8 per cent. Nobody filed that 4.8 per cent. Sharada Iyer produced it. The 4.8 per cent is a legitimate and useful sentence, and it is only ever as strong as the weaker of the two facts holding it up. An inference is written with its two parents named beside it for exactly that reason.

Grade four, scenario: what was built on purpose?

A scenarioA case built on purpose to test a boundary, not a prediction. is the only grade that is not trying to describe anything. One order in December produced 14 per cent of the revenue of the year at Meenakshi Tubes Private Limited. Ask what the figures look like if that order simply does not repeat, and the operating margin computes at 5.0 per cent. The 5.0 per cent is not a forecast, not an expectation and not a probability. The number is a machine built to find out how far a conclusion can be pushed before it breaks, and grading it as a scenario is what stops anybody downstream reading it as a view about next year.

Inference vs Scenario: Telling the Two Apart

Fact and assumption almost never get confused. One has a document behind it and the other does not. Inference and scenario get confused constantly. Both are sentences the writer produced, and both contain arithmetic the writer did. The line between them is drawn by direction of travel: an inference runs backwards from what happened, and a scenario runs forwards from something chosen.

One question settles any sentence the writer produced. Was it drawn from things that actually occurred, or built to find out what might? Realisation per unit at Meenakshi Tubes Private Limited fell about 4.8 per cent because volume and revenue both moved in the year just reported, and both movements are in the record. Nothing about that sentence is chosen. The margin of 5.0 per cent without the December order is the opposite: nothing in the record says the order will not repeat, and the whole point of the sentence is to remove something that did happen and look at the hole it leaves.

The distinction is not vocabulary policing. The two grades behave differently under attack. An inference can be strengthened by finding more facts, and can be destroyed by discrediting one of its parents. A scenario can be neither strengthened nor destroyed by evidence, only by showing the arithmetic is wrong or the construction was pointless. A reader who cannot tell which of the two is in front of them cannot tell which of those two challenges to mount. Put less politely, that reader cannot use the work at all.

How to Define a Finance Research Question, and What Are Its Three Parts?

The assignment arrived on 3 March in the form most assignments arrive in. Should the desk be worried about Meenakshi Tubes. The question cannot be answered, cannot be finished and cannot be reviewed. There is no state of the world in which it comes back yes and no state in which it comes back no. Work aimed at a question like that runs until somebody gets tired, and then produces a paragraph nobody can check.

A question becomes answerable when it names three things: a measure, a period and a threshold. The measure says which number the answer depends on. The period says which stretch of time the number describes. The threshold says what the number has to clear. Rewritten, the assignment reads: has the operating margin of Meenakshi Tubes Private Limited stayed above the 11 per cent floor set by the credit policy of the desk, in the year just reported, and what single thing could move it most. The 11 per cent floor is an internal policy invented for this illustration and is not a rule of Indian lending.

Notice what the rewrite bought. Sharada Iyer now knows which document to open first, knows when the work is finished, and knows in advance that the answer has exactly two possible shapes. Prakash Nadar can review it. He can check the measure, check the period and check the threshold separately. Latha Menon can act on it. A yes and a no lead somewhere different. The everyday version is a household asking whether it can afford a wedding: hopeless as asked, tractable the moment it becomes whether the savings balance on the last day of the month clears the quoted cost of the hall.

How Source Hierarchy Improves Finance Research: What Settles a Clash Between Two Documents?

Two items will eventually disagree, and when they do the worst possible way to settle it is a conversation about which one feels more reliable. A source hierarchyA fixed ranking that decides which item wins when two disagree. is a ranking fixed in advance, before anybody knows which document is about to be inconvenient. Ranking sources in advance is what converts a disagreement from an argument into a lookup.

The ladder runs from what a regulator or a filing system holds down to what somebody forwarded to the desk. A filed and audited statutory record sits at the top because a third party had a duty attached to it. A regulatory register entry comes next. Then a published document with a named publisher, then a published work with an identified author, then a dated and attributed conversation, and at the bottom an item with no traceable origin at all. Nothing on that ladder is worthless, and nothing on it outranks the rung above by argument.

In the working file of the Kavery research desk this stopped being theoretical. Four of the twelve public documents arrived as forwarded attachments with no traceable origin, two of those four were an extract of an extract, and one of them carried revenue of Rs 20,90,00,000 against the filed Rs 21,20,00,000. The two figures differ by Rs 30,00,000, or 1.4 per cent, small enough that a reader skimming would never notice it and large enough to change a margin. The filing won. Filed records sit at the top of the ladder and untraceable forwards sit at the bottom. The forward was not deleted. Marking it superseded and leaving it in the file lets anybody rereading see that the clash was noticed rather than missed.

Ranking sources answers which item wins. Ranking does not answer whether the item should have been believed at all, and that is the job of corroborationA second, independent item saying the same thing as the first.. Eleven of the fourteen items in the source log of this assignment were confirmed against a second independent item. Three could not be, and each of those three carries a visible mark in the log saying so. The mark is the point. An uncorroborated item is perfectly usable. An uncorroborated item that looks exactly like a corroborated one has cost the reader the ability to weigh it.

What Is Professional Skepticism, and What Is It Not?

Professional skepticismThe habit of asking what would have to be true for this to hold. is the habit of asking what would have to be true for a claim to hold, and then going to look. Skepticism is a question aimed at the claim, never a verdict passed on the person making it. The distinction sounds like manners and is actually mechanical. Doubting a person leads nowhere. Doubting a claim leads to a document that can be opened.

The best illustration in this whole assignment is a call that went perfectly well. On 10 March the finance head of Meenakshi Tubes Private Limited told Sharada Iyer that the margin had held around 12. Nobody was lying, nobody was managing anybody, and the sentence was true. Twelve was last year. The filed accounts for the year just reported give 10.8, and the filed accounts for the prior year give 12.0, so two people were each correct about a different period and neither of them noticed.

The skeptical move here was a narrow one. The move was not deciding the finance head was unreliable, and it was not averaging the two numbers into something diplomatic. Sharada Iyer asked one question: which year does that number describe? A question about the year leads to a document rather than to a judgement about somebody. The clash resolved in under a minute and cost nobody any dignity, and that is the entire content of professional skepticism as a working habit rather than a personality trait. A household version: when a relative says the fixed deposit paid nine, the useful reply is not that they must be misremembering, it is to ask which year that rate was booked in.

What Did the Four Grades Look Like on One Sheet of Working?

The whole assignment, compressed onto one sheet and graded, runs as follows. The measure is the operating margin, the period is the year just reported, and the threshold is the 11 per cent floor.

Revenue grew Rs 2,80,00,000 or 15.2 per cent, and operating profit grew too, from Rs 2,21,00,000 to Rs 2,30,00,000. Both are facts and both are cheerful. Profit grew more slowly than the revenue it came out of, so the margin still fell. A grading discipline exists for exactly that: three facts in a row can point in one direction while the measure the question actually named points in the other.

Four sentences, four grades, and only the first of them is a fact. Read them without the grade column and they sound like four equally solid statements from a competent person. Read them with it and the shape of the work becomes visible: one thing is filed, one thing was chosen, one thing was derived, and one thing was built. A reader who wants to attack the work now knows exactly where to push. A well-made file exists for exactly that, and the attack is not something to be feared.

The scenario build runs like this. Revenue falls 14 per cent to Rs 18,23,20,000. Variable cost falls with it, from Rs 11,34,00,000 to Rs 9,75,24,000. Fixed cost does not move at all and stays at Rs 7,56,00,000. Total cost is therefore Rs 17,31,24,000, operating profit is Rs 91,96,000, and the operating margin computes at 5.0 per cent. Set the slider to zero instead and the same machine returns the reported year: operating profit Rs 2,30,00,000 on revenue of Rs 21,20,00,000, a margin of 10.8 per cent.

Set the slider to 14 with the desk split selected and the panel returns exactly the worked instance above: Rs 18,23,20,000 of revenue, Rs 91,96,000 of operating profit, a margin of 5.0 per cent. Now press the second button. A bigger variable share means more cost walks out of the door alongside the revenue. Same slider, same revenue, same total cost, and the margin reads 6.5 per cent instead. At zero on the slider the two assumptions give an identical answer, and they only separate once the scenario starts running. An untested assumption can sit invisibly in a file for months for exactly that reason. Nothing was filed about that split. Somebody chose it, and the choice is worth 1.5 points of margin in the case the work was actually built to examine.

How Professional Judgment Works in Finance Practice, and Where Does the Evidence Run Out?

Evidence runs out earlier than most people expect, and it runs out in a specific place: at the moment somebody has to decide what a number means for what happens next. No filing, no log item and no corroborating document says whether a fall of 1.2 points below an internal floor is something to monitor or something to escalate. The choice between monitoring and escalating is judgment, and pretending otherwise by dressing it in one more citation is worse than admitting it.

Two axes locate the problem. One is how easily a stranger could check a claim. The other is how heavily the conclusion leans on it. A claim that is easy to check and carries no weight is harmless. A claim that is hard to check and carries no weight is also harmless. The dangerous corner is the one where a claim cannot be checked and the conclusion cannot do without it, and in this assignment there were exactly three claims sitting in it. The three claims are load-bearingRemove it and the conclusion falls over.: that the filed accounts are final, that the December order repeats, and that input cost stays inside the range of the two reported years. Drop any one and the answer changes.

The move at that point is not to gather more. The move is to write the judgment down in a form somebody can audit later, and three lines do it. Line one states, plainly, the question no source answers. Line two is the options that were weighed, all of them, including the ones rejected. Line three is what was chosen and why. Almost everybody writes only line three. A future reader is then unable to tell whether the alternatives were considered or never occurred to anybody. The decision record for this assignment ran to one sheet, six fields, dated 12 March and decided by Latha Menon, resting on four of the fourteen log items and naming three assumptions by row number.

One test tells whether the record worked, and it is brutally simple. Six months later a different person on the desk rebuilt the 10.8 per cent figure from the workpaper without asking anybody a single question. The rebuild took twenty-five minutes. A file that only its author can navigate stops existing the day the author leaves, so a rebuild by somebody else is the only real proof a record works.

How Does a Lender, an Analyst or Somebody Working Alone Use the Four Grades?

A lender reading a credit file does not read it front to back. Facts are cheap to verify and assumptions are where a file has been steered, so a lender goes looking for the assumptions. The first useful question against any body of work is which sentences were chosen rather than found, and a file that answers it on one sheet has saved the reader an afternoon. The second question is which of those chosen sentences the conclusion cannot survive without.

An analyst uses the grades in the other direction. Grading happens while writing rather than while reading. The source is still open and the analyst still remembers whether something was looked up or worked out, so grading as the work is produced is faster than grading afterwards. Grading at the end invites misremembering, always in the flattering direction. An inference from a Tuesday afternoon becomes a fact by Thursday.

Somebody working alone, with no reviewer and no desk, gets the most out of this rather than the least. There is no colleague to catch a transcription, so the grade column is the only reviewer available. A solitary worker never otherwise asks where exactly a sentence came from. Writing the grade beside each sentence forces the question. A household deciding whether it can carry a loan can run the same four columns on one sheet: the salary slip is a fact, the assumption is that both incomes continue, the inference is that spending has been rising faster than income, and the scenario is what the month looks like if one income stops. Four grades, one sheet, and it is the same discipline a desk of four people runs.

The Kavery research desk, Kavery Capital Services Private Limited, Meenakshi Tubes Private Limited, Sharada Iyer, Prakash Nadar and Latha Menon are invented.
Educational material. Not advice on any investment, tax, budget or market position.