Disclosed, Derived or Concluded: Telling the Three Apart
A disclosed figure is one the institution itself put in public. A derived measure is one a reader made by combining disclosed figures, and it always carries a choice about which ones. A conclusion is a statement about what the figures mean, and no arithmetic produces it. Each sits further from anything anybody published, and only the first can be looked up. The Reserve Bank of India decides what gets published.
Any note of analysis of a lender is likely to carry numbers of all three kinds in the same paragraph, in the same typeface, separated by commas. Nothing in a number announces where it came from. A figure lifted straight off a published statement and a figure somebody worked out in a spreadsheet at eleven at night look exactly alike once they are typed, and a verdict with a number stapled to it looks like a third figure of the same sort. The level a number sits at is invisible in the number. Somebody has to write the level down or it is lost.
What has to be true before any of this matters?
Two things, and both are ordinary enough that they slip past. The first is that analysis travels. A note is written, and then somebody who was not in the room reads it, quotes one line of it into a note of their own, and passes that on. Every step of that journey strips context, and the thing most reliably stripped is where each number came from. A figure that left one desk clearly marked as the writer's own division arrives three desks later as a fact about the institution.
The second is that people check things unevenly. Nobody re-derives every number in a paragraph. Readers check the one that looks checkable and take the rest on trust, and a number that looks published gets no scrutiny at all. Put those two together. The least supported statement in a paragraph borrows its appearance from the ones around it, and so gets challenged least of all.
The everyday version happens in every household that has ever argued about a bill. The meter reading is a fact somebody can go and look at. The monthly average is arithmetic somebody did, and if two people in the house divide by a different number of months they will get different averages without either of them being wrong. The bill being too high is a view. All three get said in one breath at the dinner table, and the argument only becomes solvable when somebody separates them.
What makes a figure a disclosed one?
An institution published it, on a stated date, in a statement it was required to publish. The whole of the definition is there, and the important word in it is required. A disclosed figure is not merely a figure that happens to be available; it is one that had to be made available, in a stated form, whether the institution wanted to or not.
Take Suvarna Commercial Bank Limited, an invented lender. For the stated year, Suvarna Commercial Bank published profit after tax of Rs 2,250 crore, total assets of Rs 2,40,000 crore, net worthThe slice of an institution that belongs to its shareholders once every claim against it has been settled. Also called shareholders' funds. of Rs 24,000 crore, deposits of Rs 1,92,000 crore, an operating expense line of Rs 5,040 crore, and interest earned of Rs 18,600 crore. Six figures, six places to go and look them up.
Six figures sit at level one not because they are large or important but because a disagreement about any of them is a disagreement with the institution rather than with whoever quoted it. A reader who believes profit after tax was something other than Rs 2,250 crore has no argument with the person who typed that line. They copied. The argument, if there is one, is with the statement they copied from, and that statement has a date on it and an address to go to. Neither of the other two levels has that property.
Four numbers about the same lender. Which one is sitting at level one?
What is a derived measure, and what is hiding inside every one?
A derived measure is what comes out of combining disclosed figures. Rs 2,250 crore of profit set against Rs 2,40,000 crore of total assets gives 0.9375 per cent. The same Rs 2,40,000 crore divided by Rs 24,000 crore of net worth gives 10.0 times. Rs 5,040 crore of operating expenses put on Rs 9,840 crore of total incomeOther income added to what a lender keeps out of interest. At least three defensible things could sit under an expense figure, and this is one of them. gives 51.22 per cent. All three are arithmetic, all three are correct, and none of them appears anywhere in what Suvarna Commercial Bank Limited published.
Now look at what happened in each of those divisions. The same thing happened three times. Somebody chose the denominator, and the choice is invisible in the answer. The same Rs 5,040 crore of operating expenses, divided instead by the net interest incomeWhat is left once the interest a lender pays away is set beside the interest coming in. Nothing else has been taken out yet. figure of Rs 7,440 crore on its own, comes to 67.74 per cent. Nobody made a mistake. Two people asked slightly different questions of the same expense line and got answers 16.52 percentage points apart.
A ratio therefore belongs with its base in the same sentence as its value. A derived measure without its base is not exactly wrong. An unbased ratio is unusable, and unusable is worse than wrong. A wrong number can be corrected, and an unusable one gets quoted onward. And there is a further wrinkle worth noticing here: neither Rs 9,840 crore nor Rs 7,440 crore is itself a disclosed figure. Total income is other income of Rs 2,400 crore added to net interest income, and net interest income is interest expended of Rs 11,160 crore taken off interest earned of Rs 18,600 crore. A derived measure can perfectly well have another derived measure sitting in its denominator. Marking the levels is therefore a habit rather than a one-off sort.
Two writers quote the same lender's cost to income ratioOperating expenses set against income, worked through in full elsewhere in this sequence. What counts as income is the part that moves. for the same year, one at 51.22 per cent and one at 67.74 per cent. Which one made a mistake?
What is an inference, and why is it different in kind?
An inference is a statement about what the figures mean. Suvarna Commercial Bank Limited is well run. leverageA count of how much an institution puts to work for each rupee that is genuinely its own. The higher the count, the more of the balance sheet was borrowed. of 10.0 times is prudent, or it is reckless. And 51.22 per cent is high. Each of those is a sentence somebody could write, and none of them is produced by any division of any two published figures.
The whole distinction turns on one point. An inference is not a derived measure with more uncertainty attached. An inference is a different kind of thing. A derived measure has a step in it that anybody can repeat, and an inference has no step at all. Two careful readers looking at exactly the same disclosures can therefore reach opposite inferences without either of them having made a mistake.
The difference changes what checking can mean. A derived measure somebody disagrees with leaves something to do: get the inputs, redo the division, find where the two accounts part company. An inference somebody disagrees with leaves nothing of that sort available. Argument remains, and argument is a perfectly respectable activity, but it is not checking and it does not end the way checking ends. The distinction is not about how confident anybody is. The distinction is about whether a procedure exists.
Three statements about the same lender. Which one is an inference?
Which of the three is it, and how is that told in one second?
Here is the test, and its virtue is that it takes a second and needs no knowledge of banking at all. Ask: who would I go to in order to check this?
If the answer is the institution's own published statement, the figure sits at level one. If the answer is the reader's own arithmetic and that arithmetic can be done again, it sits at level two. If the answer is nobody at all, with no source to go to and no calculation to redo, it sits at level three. The test works because it asks about the procedure for checking rather than about the content of the claim, and the procedure is exactly what the three levels differ in.
The test runs just as well on a sentence the reader did not write, on a subject the reader knows nothing about. Somebody states that a fund house runs assets under managementThe money an institution manages or lends against, measured at a stated date. What counts inside it is set by the kind of institution it is. of a certain size: who would be asked? The fund house. Somebody states that this works out to a fee of so much per rupee: who would be asked? Anybody with a calculator. Somebody states that the fee is reasonable: who would be asked? Nobody. The silence is itself the answer, and a useful answer rather than a failure of the test.
A lender's leverage is stated as 10.0 times. Which level is that, and who checks it?
Predict it before the next block works it through. A summary lists profit, total assets and a cost to income ratio in one column of figures. Which of the three will a reader treat as published?
Where does a derived measure get read as a disclosed one?
In a list. A list is the whole answer, and it is worth being blunt about it because the mechanism is so plain once it has been seen. A summary sets out profit after tax of Rs 2,250 crore, total assets of Rs 2,40,000 crore, deposits of Rs 1,92,000 crore and a cost to income reading of 51.22 per cent, one under the other, in the same column. Three of those four were published by Suvarna Commercial Bank Limited. The fourth is a division somebody performed after choosing between at least three defensible denominators, and it is sitting in the same column looking exactly like its neighbours.
The sentence pattern that does the damage is a ratio quoted with no denominator, in a list beside figures that genuinely were published, and it is recognisable the moment somebody knows to look for it. There is no deception involved and usually no carelessness either. The person compiling the summary knew perfectly well where the ratio came from. The compiler simply did not write it down. At the moment of writing it seemed obvious, and obvious is a property of the writer's desk rather than of the finished note.
Where does an inference get read as a derived measure?
In a sentence like this one: costs are high at 51.22 per cent. Read it once and it feels like a single statement carrying a single number. It is not. Half of it is a division and half of it is a verdict, and the verdict has quietly borrowed the credibility of the division. High compared with what? Compared with which year, which other institution, which base? None of that is present, and the 51.22 per cent is standing in for all of it.
The repair is not to delete the verdict. A view is what analysis is for, and a column of numbers with no view in it has left the reader to do the hardest part alone. The repair is to separate them: operating expenses were 51.22 per cent of total income for the stated year, and then, in its own sentence and marked as a view, whatever the writer wants to say about that. Two sentences instead of one.
The gain from splitting them is precise. A reader can now disagree with the second sentence without doubting the first. Before the split, disagreeing with high meant somehow disagreeing with 51.22 per cent. Nobody can do that, so the reader either swallowed both or distrusted both. Splitting it gives them somewhere to put their disagreement. The split also forces the writer to notice whether they actually have a reason for the verdict, and being forced to notice is useful.
Rewrite this so a reader can check half of it: costs are high at 51.22 per cent. Which rewrite does the job?
Two lenders both post 9.38 per cent on equity. Which level is the equality itself, and which level is anything said about it?
Where does the equal number actually sit?
The equal number is the sharpest case of all three levels sitting together. Suvarna Commercial Bank Limited and Rukmini Finance Limited, both invented, both earn 9.38 per cent on equity in the stated year, and the two figures are equal on purpose rather than by coincidence. They were built that way. Suvarna Commercial Bank gets there from 0.9375 per cent of total assets, taken up by an assets to net worth ratio of 10.0 times. Rukmini Finance Limited gets there from 1.875 per cent of its assets under management, taken up 5.0 times. One of the two makes half as much from every rupee on its books while running twice as many rupees against the money that is genuinely its own, and those two differences cancel out to the last digit.
The rounding is worth stating plainly before it trips anybody. Reported to two places, the limbs on assets read 0.94 per cent and 1.88 per cent. Taking 0.94 up ten times gives 9.40; taking 1.88 up five times gives 9.40 again; and 9.40 is not 9.38. Nothing has gone wrong. Before the rounding the limbs sit at 0.9375 per cent and at 1.875 per cent, each of which carries all the way to 9.375 per cent once it meets its own leverage, and it is 9.375 that becomes 9.38 in print. Somebody working from the two rounded limbs lands on 9.40 and quite reasonably decides the arithmetic cannot add up, so the unrounded limb goes in beside the reported one wherever the pair appears.
Now place all of that. Divide Rs 2,250 crore by Rs 24,000 crore and 9.375 per cent comes out; divide Rs 337.50 crore by Rs 3,600 crore and 9.375 per cent comes out again. Each is a disclosed profit set on a disclosed net worth, so the equality itself is a level two fact and it is checkable in about fifteen seconds. Whether the two institutions are therefore equally good, whether one of them is the better run, and whether the finance company is taking less risk because it carries less leverage are every one of them level three. The record holds one year for each institution, no downturn and no failure, so nothing in it could support any of those statements. The arithmetic is what stands, and what the arithmetic cannot settle is named alongside it.
Seeing the two limbs drawn to scale is what stops the equal number being read as sameness. One institution earns half as much per rupee of assets and carries twice the assets per rupee of its own money. Neither difference is small. Each is a factor of two, the two run in opposite directions, and they cancel exactly. A reader ranking lenders on the equity return alone would put these two in the same place on the list, and by every other measure available the two are nothing like each other.
How somebody writing up a lender for a credit committee actually uses this
Not as a philosophical exercise. A credit note goes into a room where several people who did not do the work have to decide something, and the only defence a writer has against being asked to justify every line is having marked which lines need justifying. So the note is built in three bands, without ever using those words in it. The published figures go in with their statement and its date beside them. The ratios go in with their arithmetic shown and their base named. The view goes at the top or the bottom in its own paragraph, in words, and it does not carry a number.
The reason is entirely practical: a committee that cannot see which parts of a note are checkable will either accept all of it or reject all of it, and both of those outcomes waste the work. A note that marks its levels gets a much better meeting. Somebody disputes the view, and disputing the view is what a committee is for. Nobody disputes the profit figure. There is nothing in it to dispute. And if somebody wants the expense ratio on the other base, the base was written down, so it takes a minute rather than a week.
An analyst covering both Suvarna Commercial Bank Limited and Rukmini Finance Limited runs into the same discipline from the other side. The two returns on equity are identical by construction, so any note treating that equality as a finding about how either does business is making a level three statement out of a level two fact. One stated year each, no downturn in the record and no failure anywhere in it: that is not enough to conclude anything, and the honest note says so in a sentence rather than leaving it out.
Who decides what an institution has to publish at all?
Not the institution, and not the reader either. A figure is a disclosed one because a supervised institution was required to publish it, in a stated form, at a stated frequency. Which figures those are, the heads an institution must sort its income and expense into when it reports them, and the capital a bank must report against its risk weighted assetsAn institution's assets restated by how much of each is treated as being at risk. The weights are set by the supervisor, not by the institution. are all decided by the Reserve Bank of India at rbi.org.in, and all three move.
The rule connects straight back to level one, and the connection is easy to miss: the first level of this hierarchy exists because a rule created it, so the boundary of what is checkable moves whenever the rule does. Widen what must be published and figures that used to be level two, worked out by readers from whatever was available, become level one. Narrow it and the traffic runs the other way. The hierarchy is not a fact of nature. The hierarchy is the shape a disclosure requirement leaves behind.
Which authority draws that boundary depends on the kind of institution. For Suvarna Commercial Bank Limited and Rukmini Finance Limited the authority is the Reserve Bank of India. For Chandrika Life Insurance Limited, also invented, an insurer's reporting and its solvency margin sit with the Insurance Regulatory and Development Authority of India (IRDAI) at irdai.gov.in. Vaidehi Asset Managers Limited, together with Kaveri Stock Exchange Limited and the clearing corporation attached to it, answers to the Securities and Exchange Board of India (SEBI) at sebi.gov.in on registration and reporting. For Setu Payments Limited the route runs back to the Reserve Bank of India. And deposit cover, where a deposit is involved at all, is a row of the Deposit Insurance and Credit Guarantee Corporation's own, and dicgc.org.in is where somebody reads it. Six institutions, four authorities, and the boundary of level one drawn in a different place by each.
Three rows drawn here and answered somewhere else
| The row | Where it is read from |
|---|---|
| Which figures a supervised institution has to publish, in what shape and how often, which is what makes a figure a disclosed one rather than an estimate | Reserve Bank of India, rbi.org.in |
| The heads a supervised institution has to sort its income and its expense into when it reports them | Reserve Bank of India, rbi.org.in |
| The capital a bank has to report and carry against its risk weighted assets | Reserve Bank of India, rbi.org.in |
Each row is settled at the authority named beside it, and the three levels behave the same way whatever the answer turns out to be.
What makes profit after tax a disclosed figure rather than something a reader worked out?
What does the marking look like once somebody else picks it up?
Three rules, and they are deliberately mechanical. A habit that needs judgement at the moment of writing is a habit that fails under deadline. Write disclosed figures with their source and their date. Write derived measures with the arithmetic that produced them and the base they were struck on. Write inferences as sentences that carry no number at all.
The third rule is the one doing the real work, and it is the one that feels strangest until the reason is clear: an inference with a number in it borrows the number's authority, and an inference written without one has to stand on its own two feet. Strip the figure out of costs are high at 51.22 per cent and what remains is costs are high, inviting the question compared with what. Compared with what was always the right question. The number was hiding it.
There is a second benefit that only shows up later. A note written this way survives being cut up. Somebody lifts one line for a different purpose, and the line arrives carrying its own provenance: the source if it was disclosed, the base if it was derived, and no number at all if it was a view. Nothing has to be reconstructed from memory by somebody who was not there, and reconstructing from memory is where most of the damage gets done.
What happens when one note of analysis is marked line by line?
Take a note of analysis of Suvarna Commercial Bank Limited and go through it. Every figure lands in one of the three levels, and the sorting takes a couple of minutes.
Level one, the six figures the institution published
| The figure | Value | Where the argument would go |
|---|---|---|
| Profit after tax | Rs 2,250 crore | Its own published statement |
| Total assets | Rs 2,40,000 crore | Its own published statement |
| Net worth | Rs 24,000 crore | Its own published statement |
| Deposits | Rs 1,92,000 crore | Its own published statement |
| Operating expenses | Rs 5,040 crore | Its own published statement |
| Interest earned | Rs 18,600 crore | Its own published statement |
Level two, six measures, each carrying the choice that made it
| The measure | Value | The base it was struck on |
|---|---|---|
| Return on assets | 0.9375 per cent | Profit after tax over TOTAL assets |
| Leverage | 10.0 times | Total assets over NET WORTH |
| Return on equityProfit set against the institution's own money. Its two limbs are taken apart separately. | 9.375 per cent | Profit after tax over NET WORTH, and also the first two multiplied together |
| Cost to income | 51.22 per cent | Operating expenses on TOTAL INCOME, which is Rs 9,840 crore |
| The same expenses, other base | 67.74 per cent | Operating expenses on NET INTEREST INCOME alone, which is Rs 7,440 crore |
| Deposit share of the funding | 80.0 per cent | Deposits over TOTAL assets |
Look at the fourth and fifth rows together. The pair repays the trouble. Same numerator, same year, same institution, and a gap of 16.52 percentage points between the two readings. Notice also the complement nobody prints: if operating expenses were 51.22 per cent of total income, then 48.78 per cent of total income was everything else, and on the other base the split is 67.74 and 32.26. Two pairs of readings, four numbers, and each pair adds to a hundred per cent of its own base and to nothing at all across bases.
Level three, four statements the figures above do not settle
Suvarna Commercial Bank is well run. 10.0 times is prudent. A cost to income reading of 51.22 per cent is high. Suvarna Commercial Bank Limited and Rukmini Finance Limited are equally good because both return 9.38 per cent on equity. Each of those is a sentence somebody has written about figures like these. None is produced by any arithmetic above, and each one is named so that a reader can see exactly where the arithmetic stops.
The sentence that carries four statements and marks none of the joins
An analyst writes: the bank is efficiently run, with costs at 51.22 per cent and a return on equity of 9.38 per cent, in line with its finance company peer. One sentence, and four separate things are happening inside it.
Efficiently run is an inference with no source. The 51.22 per cent is a derived measure whose denominator is not stated, and the same expenses read 67.74 per cent on the other defensible base. The 9.38 per cent is a derived measure that is the product of two limbs, neither of which appears: 0.9375 per cent of assets at 10.0 times at the bank, and 1.875 per cent of assets at 5.0 times at the finance company. And in line with its peer is a second inference resting on an equality that was constructed to be equal in the first place.
The specific cost is that a reader who tries to check the sentence can verify none of it, and will therefore either accept all of it or reject all of it, and both of those responses are wrong. Worse, it gets repeated into the next note. The sentence read as a summary of facts, and summaries of facts get quoted. The analyst most likely to write it is not a careless one. The writer at risk is the one summarising under deadline, for whom marking the levels feels like padding at exactly the moment there is no room for padding.
The repair costs three sentences instead of one. The disclosed figures with the statement they came from. The derived measures with their bases. The view on its own, in words, with no number attached to it.
One last check, and it is the marking rule that people resist most. Why should an inference carry no number at all?
Where each blank above gets filled in
| Confirmed on | Address | Authority | Where the answer is to be read |
|---|---|---|---|
| 23 August 2026 | rbi.org.in | Reserve Bank of India | Three rows: which figures a supervised institution has to publish and in what shape and how often, the heads it has to sort income and expense into when reporting them, and the capital a bank has to report against its risk weighted assets. The first of those is what makes a figure a disclosed one at all, so the whole hierarchy moves when it moves. |
| 23 August 2026 | sebi.gov.in | SEBI | Registration and reporting conditions reaching an exchange, its clearing corporation, a broker and an asset manager. The boundary of level one is drawn by a different authority for those institutions. |
| 23 August 2026 | irdai.gov.in | IRDAI | What an insurer reports and the solvency margin it works to, set by a different authority again. |
| 23 August 2026 | dicgc.org.in | Deposit Insurance and Credit Guarantee Corporation | Whatever stands behind a deposit. Deposits of Rs 1,92,000 crore are one of the six figures marked at level one. |
| 23 August 2026 | ideas.repec.org | Research Papers in Economics | The route a writer takes before putting a name to an idea. The decomposition used above is algebra rather than a borrowed frame. |
Suvarna Commercial Bank Limited, Rukmini Finance Limited, Chandrika Life Insurance Limited, Vaidehi Asset Managers Limited, Kaveri Stock Exchange Limited and Setu Payments Limited are invented.
Educational material. Not advice on any investment, tax, budget or market position.