How to Spot Accounting Red Flags in an Annual Report

Looking for red flags is a procedure whose most important step is the one most people skip. Seven steps: establish the basis, compute the divergences, list every flag, generate the ordinary explanation for each before investigating, check which the disclosures already answer, rank what remains, and write down what evidence would settle each. A flag that survives its ordinary explanation is a question. A question is still not a finding.

Underneath that sits one uncomfortable fact about reading. By the time a set of accounts is opened, the reader usually already has an impression of the business, and an impression is extremely good at finding evidence for itself. Almost any pattern in a large document has some supporting detail somewhere: a customer who pays slowly, an estimate that moved, a line that grew faster than the line above it. None of that detail is made up. Supporting detail is simply what a search for supporting evidence returns, and the identical search run over an honest business returns the identical kind of thing. The seven steps below are arranged so that the arithmetic chooses the flags and the innocent explanation is written down before any of it is looked into.

Everything needed to run it is already in place. The divergences used here are all published already: the receivables gap, the stock build, the movement in the provision and the fall in every margin below the gross line were each worked out where they belong, and so was the explanation for each. The order in which those divergences are taken has never been set out. Seven steps run below on a real set of accounts. Step four is the one that gets skipped, and the moment it is skipped everything downstream changes. Five disclosures settle most of what a screen throws up. The output at the end is honest, useful and checkable, and it is not a conclusion.

Where does the procedure start, and why there?

Step one establishes the basis, and it comes first because every number computed later stands on it. Three things get settled before any arithmetic runs. Which set of accounts is in hand, the standalone one or the consolidated one. Next, what changed in the accounting during the year, meaning any change of policy and any change of estimate the company has disclosed. And whether the comparative column was restated. A restated comparative and an original one are not the same base, so a growth rate computed across that join is not a growth rate at all.

Each of the three has a fixed place. The heading printed on the statements themselves settles the first. The notes on accounting policies and on estimates settle the second. A note explaining any restatement settles the third, and where a company has restated, it says so.

Step one produces no flags at all. Step one produces a base, and a divergence computed against the wrong base is not a divergence, it is an artefact of the reading itself. Step one is skipped even more often than step four, and it is skipped for a duller reason: nothing happens during it. There is no output to show anybody. The cost of skipping it is invisible until every later number is wrong by an amount that cannot be seen.

Run step one on Anjani Stationers Private Limited and it takes four lines. The set in hand is the standalone one. No accounting policy changed during the year. No estimate method changed. The comparative column was not restated. One thing did change and it gets recorded rather than judged: the company has bought 70 per cent of Chitra Binding Works at the start of year two, so year two carries a transaction and a set of dealings that year one does not.

Why does computing come before reading?

Step two computes the divergences. Not reads for them, computes them, and it computes all of them whether or not any one of them looks interesting on the way past. A divergenceTwo figures that would normally move together, moving apart. Receivables against revenue, stock against sales, a cost line against the line above it. The word says only that the gap exists, never why. here means two figures that would normally travel together having moved apart, and the step is finished when the whole set has been worked out.

Four things get computed in one pass. Growth in revenue, and then growth in receivables, inventory and payables measured against it. The margin at each rung, from gross down through earnings before interest, tax, depreciation and amortisation (EBITDA) and earnings before interest and tax (EBIT) to net. Cash conversion, meaning operating cash flow set against profit after tax. And the movement in every estimate the company discloses, stated both in rupees and as a multiple. A small rupee figure can be a large multiple, and a large rupee figure can be a small one.

Computing before reading is what stops the narrative choosing the flags. The reading version of this happens outside any set of accounts. Somebody says that the shop at the end of the street is in trouble. The next walk past turns up a shutter half down at four in the afternoon, a window display that has not changed in a month, and a shopkeeper who looked tired. Not one of those three observations is false. All three were collected after the claim arrived, and not one of them would have been collected the week before. Reading a management commentary before computing anything does exactly that to a set of accounts, and it does it invisibly. The numbers that end up being quoted are all real.

Why must the flag list stay unfiltered?

Step three writes down every flag the divergences produced, in the order the arithmetic returned them, with nothing taken out. Step three is the shortest step in the procedure and the easiest one to spoil.

The temptation is to tidy. A divergence that looks small in rupees gets dropped. One that seems to have an obvious explanation gets dropped because the explanation seems obvious. One that does not fit the shape of what the reader already thinks gets dropped without any decision being made about it at all. The third drop is the most common of the three and the only one that leaves no trace.

A filtered list is a conclusion that has already started to form, and it has formed before a single explanation has been written down. Two habits keep the step mechanical. The divergences that did not cross go down as well as the ones that did. A list holding only the alarming half is not the arithmetic's output any more, it is the reader's. And not one word goes down about why any item is on the list. The reason belongs to step four, and writing it here means step four has been started on one flag and skipped on the rest.

Six divergences were computed on Anjani Stationers and four of them crossed. The two that did not cross go on the list in exactly the same handwriting as the four that did.

What is step four, and why is it mandatory?

Step four generates the ordinary explanationA specific, checkable account of how an honest business could produce the exact number in front of the reader. Specific is the operative word: a vague guess such as could be seasonal is not one. for every flag on the list, and it generates all of them before a single flag is investigated. Step four is the one step written in capitals in every note anybody has ever kept about this procedure.

Step four is not politeness, and it is not giving anybody the benefit of the doubt. Step four is the only step that changes what can be seen at every later step. Here is why the ordering carries all the weight. Investigation is a search, and a search returns what it is pointed at. Point it at a suspicion and it comes back with confirming detail. A set of accounts and its notes run to tens of thousands of numbers, and somewhere among them sits a customer who pays slowly, an estimate that moved, a cost line that grew faster than the one above it, a party that is related. None of that is fabricated. Confirming detail is what a search for supporting evidence returns, and the same search run on a business doing nothing whatsoever wrong returns the same catch.

Writing the innocent explanation down first does something that a resolution to be fair cannot do. Writing it down puts a second claim on paper, equally specific and equally checkable, before the search begins, so the search now has two things to be evidence for rather than one. Step four is confirmation biasThe tendency to notice, remember and go looking for whatever supports what is already believed, while the evidence that would count against it never gets collected in the first place. handled as a procedure rather than as a good intention, and that is exactly why it is written as something performed rather than something held.

The mechanics are small. For each flag, write one sentence describing a way an honest business could produce that exact number. Then check the sentence against a single rule: the ordinary explanation has to be written at the same length and the same specificity as the flag it answers. A flag written as a full sentence with numbers in it, sitting beside an explanation that says the words could be seasonal, has not had step four run on it. The flag has had step four gestured at rather than run, and a gesture is worse than skipping the step outright. A gesture leaves a tick in the box.

Most people have done this correctly at least once without calling it anything. A parent looks at a phone bill that has doubled and the suspicious explanation writes itself in about a second. The ordinary explanations are just as specific once they are actually written out: the plan changed part way through the cycle, the month carried an extra billing day, somebody else on the same account was travelling. A parent who writes those three down before opening the itemised statement reads that statement looking for four things. A parent who does not, reads it looking for one, and will find something. An itemised statement always contains something.

Which disclosures already answer the flags?

Step five takes each flag with its written explanation and asks one narrow question of the document: does something already printed in here settle which of the two accounts is right? For most flags the answer is yes, and that is not luck. The disclosures exist because somebody decided, years before any reader opened the report, that a reader would need exactly these facts.

Five places carry most of the answers. Each of them has a name, and knowing the name is most of the skill.

The document that produced the flag is usually the same document that answers it. A reader who skips the notes has generated a question that was already answered further on in the same report. Skipping the notes is the single largest source of wasted work in this whole subject, and it is entirely avoidable. Step five is also where the procedure becomes cheap: four flags took about twenty minutes of arithmetic to raise and about ten minutes of note reading to settle.

How is what is left ranked?

Step six ranks whatever survived step five, and the ranking rule is narrow. How much would this change the reading of the business if the unfavourable explanation turned out to be the right one? Consequence is the only sort order. How alarming the words are is not a sort order, it is a description of the reader's own reaction.

The two come apart constantly. A tripled provision sounds severe and moves Rs 6,00,000. A payable stretch of twelve and a half days sounds procedural and moves a considerably larger number through the cash cycle. Ranking by the sound gives one order; ranking by consequence gives another, and the second one is the one worth anybody's time.

Ranking by consequence makes the sort order stop matching the order of the reader's own alarm almost immediately. The mismatch is the clearest sign the step is working. There is a second filter running alongside it: whether the item is materialLarge enough that knowing it could reasonably change what somebody reading the accounts decides to do. It is a judgement about the reader, not about the size of the number in isolation. at all. An item that could not change anybody's reading, whichever explanation is right, comes off the ranking and goes into a line saying it was checked and did not matter.

Whatever survives step six is usually a great deal less than what was listed at step three, and on many sets of accounts nothing survives at all. An empty result is not a failed procedure. A procedure that only produces output when it finds something is a procedure that will always find something. Another name for that arrangement is a machine for manufacturing suspicion.

What is written down before the reading stops?

Step seven writes three things for each surviving flag and then stops. The question, in words somebody else could act on. The evidence that would settle it. And where that evidence would come from.

Naming the evidence is the whole difference between a question and a suspicion. The receivables look stretched is a suspicion: it cannot be settled by anybody, it cannot be closed by anybody, and so it sits in a note forever and gets copied into next year's note. Whether the largest customer group's payment pattern holds next year, settled by that group's closing balance and its overdue bands in next year's ageing profile, is a question. The question has a date on which it becomes answerable and a place where the answer will appear.

A question with named evidence has a way of being closed, and a suspicion without one never does. Suspicions accumulate over the years, and questions do not. Write the source alongside the evidence, and prefer the audited statements and their notes over anything said about them. A statement somebody made about a number and the number itself are two different kinds of evidence carrying two different weights.

What happens when all seven steps run on Anjani Stationers?

Run the whole procedure on Anjani Stationers Private Limited, a business making school notebooks and exercise books, and this is what falls out. Step one settles the basis: standalone accounts, no policy change, no estimate change, no restatement, and a 70 per cent holding in Chitra Binding Works bought at the start of the year. Step two computes six divergences. Step three lists all six, four of which crossed: receivables growth of 21.8 per cent against revenue growth of 12.5 per cent, inventory growth of 47.4 per cent from Rs 19,00,000 to Rs 28,00,000, the provision for doubtful debts tripling from Rs 3,00,000 to Rs 9,00,000, and every margin below the gross line falling while gross margin held at 45.0.

Step four writes the four ordinary explanations. Step five checks them against the notes and finds all four already answered. Step six finds nothing material left standing. Step seven records two questions with their evidence and stops.

Anjani Stationers raised four classic red flags and came through all four, and that is an ordinary outcome for an honest business rather than a lucky one. Four flags is a perfectly normal number for a small manufacturer that had a busy year: it bought a business, stocked ahead of its season through a period of rising paper prices, and carries a large school group that pays on the terms it agreed. Every one of those facts produces a divergence, and none of them is anybody doing anything wrong. Treating four flags as unusual means treating most honest businesses as unusual, and a test that most honest businesses fail says nothing about the ones that are not honest.

The arithmetic also said something in the reassuring direction, and step three recorded it rather than dropping it: operating cash flow of Rs 36,30,000 against profit after tax of Rs 30,00,000, a cash conversionCash generated from operations set against reported profit for the same period. Above one means the period turned more cash into the business than it reported as profit; it is one reading among several and settles nothing on its own. of 1.21 times. The cash conversion figure was on the unfiltered list from the start, and a reader who had tidied the list down to the alarming four would never have had it in front of them.

The two paths part at step four, and every reading after it differs. On the full procedure, the flag count reaches four at step three and stays there, four explanations appear at step four, all four are answered at step five, nothing material survives step six, and step seven prints a question list carrying two questions and no findings. On the skip path, the flag count still reaches four at step three, the explanation column stays permanently empty, and step five fills instead with four pieces of supporting detail, each of which is simply the flag restated as its own confirmation. At step seven the skip path prints a conclusion that the accounts are aggressive, in confident capitals, on a business that did nothing wrong, and every one of its four supporting points is answered in the notes of the same annual report it was written from.

When does the reading stop, and what is the output?

The reading stops when three conditions hold at once. Every flag on the list has an ordinary explanation written against it. Each of those pairs has been checked against the disclosures. And whatever remains is written as a question with named evidence and a named source.

The output is a question listA short written set of open questions, each paired with the specific evidence that would settle it and the place that evidence would come from. It is a finished piece of work, not a draft of something else., and a question list is the finished output rather than an unfinished one. The question list is the part that feels wrong the first several times. An afternoon has gone into a set of accounts and the thing to show for it is two questions. Two questions do not feel like a result. Both can be checked by somebody else, closed by somebody else and dated, and a paragraph of impressions offers none of that.

The urge to convert the question list into a conclusion is the single failure this whole procedure exists to prevent. A reader who converts it has thrown away the only thing the procedure produced, a small set of claims that can be tested, and swapped it for one that cannot.

What must never be a step?

Three things are never steps, and each of them arrives disguised as efficiency.

Skipping step four is the first and by a distance the worst. Everything downstream still runs and produces output that looks identical to the honest version. Ranking by alarm is the second: it puts the loudest item at the top of a list that was supposed to be sorted by consequence, and the person reading the work has no way of telling which sort order was used. The third is publishing a flag as a finding. Publishing an unresolved flag as a finding about a named business is wrong on the facts and, in most places, actionableCapable of being taken to court by the person it was written about. Whether any particular statement is actionable depends on where it was published and the law that applies there, which is a legal question rather than an accounting one..

Who runs this procedure, and what do they do with the output?

Leave the steps for a moment. Four different people run this on the same set of accounts in the same fortnight, and none of them wants the same thing out of it.

A lender turns the question list into conditions, an equity analyst turns it into questions for the results call, a supplier turns it into credit terms, and a finance controller runs it on her own accounts before anybody outside sees them. Watch each one work. The output being a question list rather than a verdict is what makes all four uses possible.

A lender sizing a working capital limit for Anjani Stationers cares about one of the two surviving questions far more than the other. If the largest customer group's payment pattern lengthens again, the cycle stretches and the limit has to carry it. So the lender writes the question into the review file with a date on it, and asks for the ageing profile at the next review. A finding would have been useless here: the lender cannot lend against an accusation and would have had to go and generate the question anyway.

An equity analyst takes the same two questions to the results call, and they are answerable questions rather than rhetorical ones. The gap between an answerable question and a rhetorical one is the gap between getting an answer and getting a polite non-answer. Whether the school group's payment pattern held, and whether the session stock cleared, are both things a management team can answer with a number in one sentence.

A supplier deciding whether to extend 60 day terms to a new customer runs a much shorter version of the same thing, and the shape is identical to a household deciding whether to lend money to a neighbour's shop. The supplier looks at what does not fit, writes down the boring explanation before asking around, and then checks the boring explanation before acting on the interesting one. The reason that habit is worth having is not fairness. The reason is that acting on the interesting explanation when the boring one was true costs a customer who was never going to be a problem.

And Vaidehi Rao, as finance controller of Anjani Stationers, has the use that surprises people most: she runs the whole procedure on her own accounts before they are published. Steps two and three tell her which four flags a reader will raise. Step four tells her the explanation in each case. Step five tells her whether the disclosure as currently drafted actually answers it, and where the drafting is thin, that is what gets fixed before publication. A set of accounts where every predictable flag is answered by a disclosure a few notes later is not a set of accounts that is hiding less. Such a set of accounts has simply had the reader's procedure run against it in advance.

Anjani Stationers Private Limited, Chitra Binding Works, the Sunrise Public School group and Vaidehi Rao are invented.
Educational material. Not advice on any investment, tax, budget or market position.