Accounting Red Flags: What Forensic Analysts Look For in Indian Filings

Why the rare event is the wrong thing to look for

Most writing about accounting red flags is organised around catastrophe. It reaches for the famous collapses, works backwards from what was eventually admitted, and presents the warning signs in the order a prosecutor would. That is history, and it is a poor guide to reading a filing you have in front of you now, because it teaches you to look for the shape of a thing that has already been named. The signals that mattered in a case were obvious once the answer was known. Before that, they sat in a document alongside a hundred other numbers that were also unusual and also innocent.

There is a more useful frame, and it starts by accepting that outright fraud is uncommon. What is common, and what actually damages returns, is a company that is steadily borrowing from its own future: recognising revenue earlier, capitalising costs that could have been expensed, letting receivables stretch, holding inventory that is not really moving. None of this is illegal. Much of it is defensible one decision at a time. In aggregate it produces a profit number that is drifting further from cash every year, and the drift is measurable long before anybody uses the word fraud.

This matters for valuation in a very direct way. A multiple is a number divided by an earnings figure, and it inherits every assumption inside that figure. If you would like the mechanics of that, our explainer on what a price to earnings ratio actually measures covers the arithmetic. The point here is narrower: two companies on the same multiple are not on the same multiple if one of them is converting its earnings into cash and the other is converting them into receivables. The denominators are not the same kind of object, and no amount of care in choosing the multiple will fix a denominator that is drifting.

So the question this page answers is not how to spot a fraud. It is how to notice, from a document you can download for nothing, that a company's reported performance and its cash position have started to separate, and then how to work out whether that separation has an innocent explanation. The second half of that sentence is where most of the difficulty lives, and it is the part almost every published checklist omits.

One constraint on what follows. No real company appears anywhere on this page, as an example, a case study, or a historical reference. Every entity is constructed, its financial statements are generated from stated drivers, and the drivers are written out so the arithmetic can be reproduced or disputed. That is partly a matter of caution and partly a matter of teaching: a constructed entity lets us know the answer in advance, which is the only way to find out whether a screen actually works.

The one screen worth running first

Profit is an opinion. It is assembled from judgements about when a sale has been earned, how long an asset will last, what a doubtful debt is worth, and whether a cost belongs in this year or the next. Every one of those judgements is legitimate, and every one of them has a range. Cash from operations is a much harder thing to shape, because at some point a customer either pays or does not. The gap between the two is not a scandal detector. It is a measure of how much of the reported result is currently resting on judgement.

Here is a constructed entity, Entity A, over six years. Revenue compounds at about 18 percent. The EBITDA margin improves gently, from 18.1 percent to 19.2 percent, which is exactly the profile a market rewards. Nothing in the profit and loss account looks wrong at any point.

Reported profit after tax rises from 85.5 crore rupees to 217.5 crore, an increase of 154 percent. Cash generated from operations rises from 119.2 crore to 143.5 crore, an increase of 20 percent. In the first two years the company generates more cash than it reports as profit, which is normal and healthy, because depreciation is a real charge against profit that takes no cash out of the business. Then the lines cross, and they never come back together. Over the six years the entity reports 862 crore rupees of profit and produces 681 crore rupees of operating cash, so 21 percent of the reported profit never arrived. All figures are illustrative and simulated.

Two ratios are worth separating here, because they are routinely confused and they answer different questions. Cash flow divided by profit after tax is the intuitive one, and on this entity it falls from 1.39 to 0.66. Cash flow divided by EBITDA falls from 0.66 to 0.32. The second is the better cross-sectional measure, and the reason is arithmetic rather than accounting theory: profit after tax sits below depreciation and interest, both of which get added back when you build the cash flow statement, so a capital-intensive or highly leveraged company will show a flattering cash-to-profit ratio for reasons that have nothing to do with earnings quality. One of the constructed entities later on this page shows a cash-to-profit ratio above two while being entirely unremarkable. Read cash flow over profit against a company's own history, and use cash flow over EBITDA when comparing one company with another.

The statements themselves are below. The cash flow lines follow the indirect method, which starts from profit before tax, adds back the non-cash and financing charges, and then adjusts for the movement in working capital. If the structure of that statement is unfamiliar, our guide to reading an Indian annual report walks through how the three statements link together and what each one is for. What matters for this page is the single line in the middle.

Notice how undramatic every individual year is. The cash conversion ratio never collapses. It steps down by about a quarter in year three, holds, and then stabilises at a lower level. Anybody reading a single annual report would see a company with a slightly weak cash year, which is common and usually temporary. The signal only exists as a sequence, and that is the first practical rule: this diagnostic requires five or six years side by side, and it is invisible in any one of them.

Where the money actually went

The cash did not disappear. It moved onto the balance sheet, and the balance sheet says exactly where. Trade receivables grew from 169.9 crore rupees to 750.0 crore while revenue grew from 1,000 crore to 2,320 crore, so receivables grew more than twice as fast as sales. Inventories did the same thing. Trade payables, the one line that would have offset it, barely moved.

Expressed as absolute rupees, none of this is conclusive, because a growing company obviously needs more working capital. Expressed as days, it is unambiguous.

Receivable days go from 62 to 118. Inventory days go from 74 to 121. Payable days go from 58 to 62, which is effectively flat. The cash conversion cycle, which is receivable days plus inventory days less payable days, more than doubles from 78 days to 177 days. That is the whole mechanism. The company is taking almost a hundred days longer to turn a rupee of cost into a rupee of collected cash than it did six years earlier, and it is funding that extension itself.

The counterfactual makes the cost concrete. Hold the same revenue, the same margins and the same profit, and simply freeze the working-capital days at their year-one levels. Cumulative operating cash flow over the six years would have been 1,210 crore rupees instead of 681 crore. The drift in the days absorbed 529 crore rupees of cash, which is about two and a half times the final year's entire reported profit. Not one rupee of that shows up in the profit and loss account, and no accounting rule was broken to produce it. All figures are illustrative and simulated.

Now run the mechanism forwards rather than backwards, because that is how it happens in practice. Take a separate constructed entity with revenue of 2,000 crore rupees, a gross margin of 38 percent, and profit after tax of 150 crore last year that is on course to grow 8 percent. Suppose it offers longer credit terms in the closing weeks of the year, and receivable days rise by 15. Fifteen days of revenue is 82.2 crore rupees. At a 38 percent contribution margin and a 25 percent tax rate, that adds 23.4 crore rupees of profit after tax. Reported profit becomes 185.4 crore, and growth of 8 percent is reported as 24 percent.

Read that again, because the leverage is the point. A fifteen-day change in a working-capital metric that almost no retail investor tracks tripled the reported growth rate. Ten days would have made it 18 percent, thirty days would have made it 39 percent. The revenue is real in the sense that goods left the building and a contract exists. Whether the cash ever arrives is a question for a later year, and if the customer cannot pay, the correction arrives as a provision for doubtful debts long after the growth was celebrated.

This is why receivable days deserve the second slot in any screen, immediately after cash conversion. Cash conversion tells you that profit and cash have separated. Receivable and inventory days tell you through which door the cash left. The two together are close to a complete diagnosis of the ordinary, non-criminal version of this problem, and both can be computed from four numbers in an annual report.

Building a screen, and watching it get things wrong

A single diagnostic on a single company is a demonstration. A screen is something you run across many companies, and the moment you do that you inherit a different problem: the screen has to be right about companies you know nothing about. To find out how a screen behaves, you need cases where the answer is already known, which is why the six entities below are constructed rather than real. Two were written to be drifting. Four were written to be clean, and each of the four is clean for a different reason, because the innocent explanations are the hard part.

Two versions of the same screen were written before any entity was scored, and neither was adjusted afterwards. The level screen applies thresholds to the latest reported year: cash flow over EBITDA below 0.60, receivable days above 90, inventory days above 90, revenue growth above 25 percent, and an accruals ratio above 0.05. This is how most published checklists are constructed. The change screen applies thresholds to movements across three years instead: cash conversion falling by more than 0.20, mean cash flow over profit below 0.80, receivable days rising by more than 20, inventory days rising by more than 20, and an accruals ratio above 0.05 in two years or more. The accruals ratio here is reported profit less operating cash flow, divided by average total assets.

At a threshold of two tests out of five, the level screen flagged three entities. Only one of them was drifting. It flagged Entity C, which is growing revenue at 44 percent a year with its receivable days sitting at 67 and 68 across the whole period, and whose cash conversion is low for the entirely ordinary reason that a business growing that fast has to fund a proportionally larger working-capital base every year. It flagged Entity D, whose receivable days are 146 and whose inventory days are 95, both comfortably over the thresholds, and both exactly where they have always been, because it is a long-cycle project business where retention money is a contractual norm. The level screen fired on the level, and in both cases the level was a fact about the industry rather than a fact about the accounting.

The change screen flagged one entity, Entity A, correctly, and cleared C and D. That looks like a decisive win for measuring movement rather than position, and up to a point it is. But both screens missed Entity F completely. Entity F was written to be drifting: its cash conversion falls from 0.73 to 0.55, its receivable days rise from 71 to 89, and its inventory days rise from 70 to 83. It scored zero out of five on the change screen. Its receivable days rose by 18, and the test fires above 20. Its inventory days rose by 13. Every one of its movements sat just inside a threshold that had been chosen to avoid false alarms.

That is the honest result and it deserves to be stated plainly rather than buried. The change screen did not look precise because it was well designed. It looked precise partly because it was insensitive, and the two are extremely difficult to tell apart from the outside. A screen tuned until it stops producing false alarms has usually been tuned until it stops producing findings, and because the misses are invisible by construction, nothing in the output warns you.

There is one further result, and it is the inconvenient one. A single ratio, net working capital as a percentage of revenue, separated all six entities correctly. Entity A rises by 14.2 percentage points and Entity F by 7.0, while the four clean entities move by 0.7 points or less, including the one growing at 44 percent a year. Dividing by revenue removes the scale effect that fooled the level screen, and reading the change removes the industry effect. A threshold of four percentage points would have flagged both drifting entities and cleared all four clean ones, which is a better result than either five-test screen achieved.

That threshold was chosen after seeing the six answers, on a panel of entities written by hand, and it should be treated accordingly. A perfect score on six cases you constructed yourself is not evidence that a rule works. It is a demonstration of where the information sits in this particular set of statements, which is a much weaker claim and the only one the exercise supports. What survives is the design principle rather than the number: divide out scale, measure the movement, and prefer one ratio you understand to five you have averaged together.

The false alarm problem, stated honestly

Everything so far has been about whether the screen can tell a drifting company from a clean one. There is a second problem, and it defeats good screens rather than bad ones. The thing being searched for is rare, and rarity does something to test results that intuition consistently gets wrong.

The arithmetic below rests on an assumption that could not be verified from any primary source and should be read as a stated assumption rather than a statistic: suppose one listed company in a hundred is materially misstating its position in a given year. Suppose further that the screen is good, catching eight of every ten such companies, and that it wrongly flags one clean company in ten.

Out of a thousand companies, ten are drifting and 990 are clean. The screen catches 8 of the 10 and misses 2. It also flags 99 of the 990 clean companies. The list handed back has 107 names on it, of which 8 are the ones you were looking for, so about 7 percent of the flagged names are genuine. Halve the assumed prevalence and precision falls to about 4 percent. Double it and precision rises to about 14 percent. Every one of those numbers is dominated by the assumption, not by the quality of the screen, and that is exactly the point.

Two consequences follow, and they should change how the output is used. The first is that a screen is a queue, not a verdict. Its job is to reduce a thousand companies to a hundred worth an afternoon each, and it has done that job well even when ninety-three of the hundred turn out to be fine. The second is that improving the screen's sensitivity is usually the wrong optimisation. Going from catching eight in ten to catching nine in ten adds one genuine name. Halving the false-positive rate removes roughly fifty innocent ones. When the base rate is low, specificity is worth far more than sensitivity, which is the opposite of what most screening effort goes into.

It also explains why the innocent explanations deserve as much space as the flags. A screen that flags everything is not conservative, it is useless, because a list nobody can work through is the same as no list. Each of the four clean constructed entities on this page had a specific, checkable, ordinary reason for looking unusual: a genuine growth phase, a sector norm, a one-off order that reversed the following year, and in one case nothing unusual at all. Establishing which of those applies takes perhaps twenty minutes with the annual report, and that twenty minutes is the actual work.

The signals that are not in the ratios

Everything so far can be computed. The signals in this section cannot, and forensic analysts weight them heavily precisely because they resist quantification: they are behaviour rather than arithmetic. Each one below comes with where it lives in an Indian filing, the mechanism that makes it informative, and the innocent explanation that has to be ruled out before it means anything at all.

A modified audit opinion. The Standards on Auditing recognise three kinds of modification, set out in SA 705(Revised): a qualified opinion, an adverse opinion, and a disclaimer of opinion. They differ in severity and in kind. A qualification says the accounts are fine except for a specified matter. An adverse opinion says they are not fine. A disclaimer says the auditor could not obtain enough evidence to form an opinion at all, which is the most serious of the three and the one most often misread as merely procedural. Separately, SA 701 requires the auditor to communicate key audit matters, being those that required the most judgement, and SA 706(Revised) provides for an emphasis of matter paragraph that draws attention to something already disclosed without modifying the opinion. All three standards are effective for audits of financial statements for periods beginning on or after 1 April 2018. The practical instruction is simple: read the opinion paragraph and the key audit matters before you read the numbers, because they tell you which numbers the auditor found hardest to verify.

An auditor resigning, and how often auditors change. This is one of the few events where the regulation itself assumes the reason matters. Under Schedule III Part A Para A(7A) of the SEBI Listing Obligations and Disclosure Requirements Regulations, 2015, a listed entity must disclose the detailed reasons for an auditor's resignation, as given by the auditor, within twenty four hours of receiving them. SEBI circular CIR/CFD/CMD1/114/2019 dated 18 October 2019 goes further: it prescribes a format in which the auditor must state whether information was withheld by management and whether alternative procedures were performed, requires the auditor to raise concerns with the audit committee chairman without waiting for a scheduled meeting, and requires the audit committee's own views to be disclosed within twenty four hours of its meeting. The circular also constrains the timing, so that an auditor resigning cannot simply leave a reporting period unaudited. The innocent explanations are real and common: mandatory rotation under section 139(2) of the Companies Act, 2013, which caps an individual auditor at one term of five consecutive years and an audit firm at two such terms, a merger between audit firms, or a straightforward disagreement over fees. Read the stated reason first. A resignation whose stated reason is rotation is an administrative event. A resignation with no stated reason, or one describing an inability to obtain information, is a different document entirely.

Frequent changes of chief financial officer. A change of chief financial officer is a deemed material event under Schedule III Part A Para A(7) of the Listing Regulations, which names the role expressly alongside the managing director, chief executive officer and company secretary, and it must be disclosed under Regulation 30 within twelve hours, or within thirty minutes of the close of the board meeting at which it was decided. No single departure means anything. What is informative is the run rate, because the chief financial officer is the person who has to sign off on the judgements the screen has been measuring. Three in four years is worth a note; one in four years is a career. The innocent explanations here are the strongest of any signal on this page, which is why it belongs in a supporting role and never as a primary flag.

Related party transactions. These are dealings between the company and parties connected to it, and they matter because they are the ordinary route by which value moves out of a listed entity and into somewhere else, at prices nobody negotiated at arm's length. The disclosure architecture in India is unusually strong. Under Regulation 23(2) of the Listing Regulations, all related party transactions require prior approval of the audit committee, and only the independent directors on that committee may approve them. Under Regulation 23(4), a material related party transaction requires prior shareholder approval, and no related party may vote on the resolution whether or not it is a party to that specific transaction. The materiality thresholds themselves are set out in Schedule XII to the Regulations, which was substituted with effect from 19 December 2025 and now scales with the entity's consolidated turnover rather than applying a single flat figure. Regulation 23(9) requires a listed entity to file related party transaction disclosures with the exchanges every six months, on the date it publishes its results, and to put them on its own website. The Companies Act, 2013 adds its own layer through section 188, which requires a board resolution for specified categories of transaction, and section 177(4)(iv), which places approval of related party transactions within the audit committee's mandate. What to rule out: intra-group transactions in a legitimately structured business are normal and often unavoidable, and a company with a manufacturing subsidiary and a distribution subsidiary will show large related party numbers forever. What is informative is a related party number growing faster than revenue, or a party appearing for the first time in the year the receivables balloon.

Promoter shares that are pledged. A pledge converts a promoter's shareholding into collateral, which means a fall in the share price can force a sale of the shares that were securing the loan, which can push the price down further. It is a mechanism that turns a market move into a control event. Under Regulation 31 of the SEBI Substantial Acquisition of Shares and Takeovers Regulations, 2011, a promoter must disclose the creation, invocation and release of an encumbrance within seven working days, both to every exchange where the shares are listed and to the company itself, and must file an annual declaration to the exchanges and the audit committee that no undisclosed encumbrance was created. One detail is easy to miss: a proviso inserted with effect from 1 April 2022 disapplies that disclosure requirement where the encumbrance is undertaken in a depository. Separately, SEBI circular SEBI/HO/CFD/DCR1/CIR/P/2019/90 dated 7 August 2019 requires promoters to disclose detailed reasons for the encumbrance, in a prescribed annexure and within two working days, once combined encumbrance reaches 50 percent of their shareholding or 20 percent of the company's total share capital. The running position appears in the quarterly shareholding pattern filed under Regulation 31 of the Listing Regulations, within twenty one days of each quarter end. Our page on building a fundamental scorecard treats pledge as one of its governance inputs and sets thresholds for it, so this page will not duplicate that. What to rule out here is the reason: pledging to finance an acquisition, disclosed and time-bound, is a different object from pledging that rises quarter after quarter with no stated purpose. The trend matters more than the level, which is the same lesson the working-capital screen taught.

Contingent liabilities. These are obligations that may or may not crystallise, and they are the one place where a genuinely large number can sit outside the balance sheet entirely. Under Schedule III to the Companies Act, 2013, in the division applying to companies following the Indian Accounting Standards, contingent liabilities are disclosed in the notes rather than on the face of the balance sheet, under the heading contingent liabilities and commitments to the extent not provided for, and are split into claims against the company not acknowledged as debts, guarantees, and other money for which the company is contingently liable. The accounting standard on provisions and contingent liabilities defines the concept as a possible obligation confirmed only by an uncertain future event, or a present obligation not recognised because payment is not probable or cannot be reliably measured. We could not read that standard's exact wording from the ministry's own site during preparation of this page, so the definition here is described rather than quoted, and the precise text should be checked at source before it is relied on. The signal is not the existence of contingent liabilities, because every company of size has tax disputes. The signal is a contingent liability that is large relative to net worth, and one that grows every year without ever being either settled or provided for.

What the disclosure regime actually gives you

An argument that runs through this page is that the raw material is already public. That claim is worth pinning to specific instruments, because the strength of the Indian disclosure regime is the reason a retail investor can do any of this at all.

Regulation 33 of the Listing Regulations governs financial results. Quarterly standalone results are due within forty five days of the quarter end under Regulation 33(3)(a), and annual audited results within sixty days under Regulation 33(3)(d). The provision that matters most for the diagnostic on this page is Regulation 33(3)(g), which requires a listed entity to submit a statement of cash flows for the half year as a note to its half-yearly results, both standalone and consolidated. That sub-regulation was inserted with effect from 1 April 2019. Note the frequency carefully: the cash flow statement is a half-yearly requirement under Regulation 33, not a quarterly one, and the full annual cash flow statement comes through the Companies Act as part of the annual financial statements. In practice this means the cash-versus-profit check can be refreshed twice a year rather than four times, which is a real constraint on how quickly the divergence on this page would have become visible.

Regulation 30, read with Schedule III Part A, governs event disclosure. Some events are deemed material and require no judgement by the company at all, including changes in directors, key managerial personnel, auditor and compliance officer under Para A(7), and the auditor resignation reasons under Para A(7A). Regulation 30(6) sets the clock: twelve hours for most events, and thirty minutes from the close of a board meeting where the decision was taken there.

Regulation 31 requires the shareholding pattern one day before listing, quarterly within twenty one days of each quarter end, and within ten days of any capital restructuring that changes paid-up capital by more than two percent. The prescribed format is set out in the Listing Regulations master circular of 11 November 2024, as partially modified by a SEBI circular of 20 March 2025 which expanded the pledge tables to capture non-disposal undertakings and other encumbrances alongside conventional pledges. If you are comparing pledge disclosures across periods, that expansion matters, because a rise in the reported figure may reflect the wider definition rather than new borrowing.

On the company law side, section 143(12) of the Companies Act, 2013 requires an auditor who has reason to believe an offence of fraud is being committed to report it, to the Central Government where the amount involved is above a prescribed threshold and to the audit committee or the board below it. The Act itself specifies no rupee figure, using the phrase such amount as may be prescribed, and the threshold of one crore rupees sits in the rules made under it rather than in the statute. We were unable to read the gazetted rule directly from the ministry's own site while preparing this page, and confirmed the figure only from a professional guidance note, so it is reported here with that caveat and should be verified at source before being relied on. What is not in doubt is the structure: frauds below the threshold are reported internally and must then be disclosed in the board's report, which means the board's report is a place worth reading rather than skipping.

The general shape of all this is that India requires a great deal to be disclosed and requires much of it quickly. What it does not do, and cannot do, is read any of it for you. The regime produces documents. Whether anyone opens them is a separate question, and the answer, for most listed companies most of the time, is that almost nobody does. That asymmetry is the entire opportunity described on this page. The disclosure is mandatory, the format is standardised, the filings are free, and the analysis is arithmetic that fits on one screen. What is scarce is not information and not capability. It is the patience to read six years of one company instead of one year of six.

Running this yourself

The arithmetic on this page needs six years of five numbers, all of which appear in the annual report: revenue, profit after tax, operating cash flow, trade receivables and inventories. That is thirty cells in a spreadsheet. The barrier has never been the data or the mathematics, and pretending otherwise is how this subject acquired its reputation for difficulty.

Step nine deserves its place at the end. By the time anybody has spent an afternoon on a company's filings, they have formed a view, and every subsequent page of the annual report gets read in the light of it. The discipline that protects against this is the same one that protects a strategy backtest from its author: decide in advance what would count as a satisfactory answer, write it down while you have no attachment to the outcome, and then hold yourself to it. A forensic screen run by somebody who has already decided is not a screen, it is a search for supporting quotations.

It is also worth being clear about what this exercise cannot do. It cannot detect a fabrication that is internally consistent, because a fabricated cash balance produces a perfectly normal cash conversion ratio. It cannot see anything that happens between reporting dates. It cannot distinguish deliberate aggression from ordinary commercial pressure, and in most cases that distinction does not exist even inside the company. What it does is narrow a thousand names to a hundred, and then tell you what to read in each one, which is a modest claim and an achievable one.

None of it requires special access, and none of it requires advanced mathematics. It requires the willingness to open a document most people never open, to read a sequence of years rather than a headline, and to accept that the honest output of a good screen is mostly a list of companies that turn out to be fine. Those are habits rather than techniques, which is both why they work and why they are rare. They are also a fair description of what serious analytical training consists of, and if the arithmetic on this page felt like the interesting part rather than the tedious part, that is the method we teach.

Every entity on this page is constructed. None corresponds to any real company, in India or elsewhere, and no real company is referred to anywhere in the text. Each entity is generated from a written set of drivers, being a prior-year revenue that fixes the opening balance sheet, a revenue path, an EBITDA margin path, and paths for receivable, inventory and payable days. From those drivers the profit and loss account, the working-capital balances and the indirect-method cash flow statement are derived arithmetically. Depreciation, finance cost, capital expenditure and tax are set as fixed proportions, which keeps the six entities comparable and is a simplification rather than a model of any real capital structure.

Both screens were specified in full before any entity was scored, and the thresholds were not adjusted afterwards. The one exception is stated in the text where it occurs: the four percentage point threshold on working capital to revenue was chosen after the six results were known, and is therefore presented as a demonstration of where the information sits rather than as a validated rule. The base-rate arithmetic rests on an assumed prevalence that could not be verified from any primary source and is labelled as an assumption throughout.

Every regulatory provision cited was checked against the consolidated text of the instrument or the original circular, with two exceptions noted in the body where the primary document could not be reached. All results are illustrative and simulated. They are not a track record, not a forecast, and not an assessment of any company. Nothing on this page is investment advice, and the presence or absence of any signal described here does not indicate that any company has done anything improper.

• What the market regulator actually does →
• Why small listed companies carry different risks →
• The eight ways a backtest can lie →
• How Indian and American market structure differ →