How to Build a Financial-Institution Peer Map

The question is fixed in writing before a single institution is chosen. Selection runs on what each one takes in and gives out rather than on what it is called. Every entry sits on one period and one named base. Return on assets and leverage get their own columns before return on equity gets one, and the work finishes by writing down what the table cannot settle.

A table of institutions is a comparison, and a comparison is only ever as good as the thing it holds constant. The rule about holding one thing constant carries almost the whole of the procedure below. Nearly every fault met in one of these tables is something that quietly was not held constant: a different stretch of time in one row, a different denominator in the next, a rule that decided one cell and not the one beside it, or one column standing in for two measures that move in opposite directions. The nine steps below are ordered so that everything which has to be held constant is fixed before a single entry is typed.

Reading a balance sheet and an income statement, and knowing what an asset, a liability and a percentage point are, is assumed here. So is the meaning of each measure. Each measure is treated separately and in full: what return on assets means, why return on equity is return on assets multiplied by leverage, what a cost to income ratio does to its base, and what a capital requirement is measured against. The nine steps below set the order the work is done in, not what the columns mean.

What is a table of financial institutions actually for?

Start with the thing almost nobody does. Skipping it is what makes the other eight steps hard. Write the question down, in a sentence, at the top of the sheet, before a single institution has been chosen to put in it. A table built to answer which lender earns more per rupee of assets holds different institutions from a table built to answer which lender leans hardest on borrowed money. The question decides the membership. The membership does not decide the question.

Here is the everyday version. Two households are moving to a new city and each writes a list of places to live. The first has written at the top of the list that they need to be within twenty minutes of a school. The second has not written anything and is simply collecting flats they liked the look of. Both end up with a list. Only one of those lists can answer a question, and the second household will discover that at the point where it has to choose. A sheet of institutions works exactly the same way. A table assembled first and questioned afterwards will answer whatever it happens to contain. Answering what the person building it wanted to know is a different thing.

The question also does something quieter that matters more later. The written question settles which columns are load bearing and which are decoration. If the written question is about how hard a lender leans on borrowed money, then leverage is the column the whole sheet exists for, and a column of share prices is furniture. Without the sentence at the top, every column looks equally important, so the sheet grows until it is unreadable and then gets sorted on whichever column sorts most cleanly.

Which institutions belong in the same table, and which do not?

Step two, and it is the step where most tables are quietly decided by a label. The test is not what an institution is called. The test is the money an institution takes in and the money it hands out, and one of them can pass on either side while failing on the other.

Run it on the two lenders in this record. On the giving-out side, the two lenders in this record are doing the same thing as each other. Both hand money to borrowers, both earn interest on it, both carry a charge for the part that will not come back. Both belong in one table on that side without any argument. On the taking-in side they are not doing the same thing at all. Deposits of Rs 1,92,000 crore sit on Suvarna Commercial Bank Limited's funding side. On Rukmini Finance Limited's there are none at all, and Rs 14,400 crore of market borrowings in their place.

So does the finance company belong in the table? Yes, if the written question is about lending. A mixed table is allowed. The discipline is to record the mixing at the top of the sheet rather than let somebody discover it halfway down a column. The one thing that cannot be done is to leave it unsaid, because a reader who arrives at the sheet later sees two rows of figures and no note, and reads every difference between them as performance.

The same test throws other institutions out cleanly. Run it before anybody reaches for a label. Chandrika Life Insurance Limited, invented, takes in premiums against a promise to pay somebody later, so the money on its books is mostly a claim the future has on it rather than a book it has lent. Vaidehi Asset Managers Limited takes in nothing onto its own balance sheet at all and earns a fee for running money that belongs to other people. Kaveri Stock Exchange Limited and its clearing corporation take in trades and give out settled ones. Setu Payments Limited takes in instructions and gives out completed payments. None of those four takes in money to lend it out, so none of them belongs in a lenders' table however comfortable the phrase financial institution makes it look.

How is the calendar stopped from doing the comparing?

Step three. Fix the period, and use one period for every entry in the table. A rate without its period is not an entry at all, it is a number, and putting a full year against part of a year in adjacent rows produces a difference that is entirely calendar.

Every rate in this record carries its period, and it is worth noticing how insistent that is. Rukmini Finance Limited earns 14.50 per cent a year on what it lends and pays 8.50 per cent a year on what it borrows. Tax is struck at 25.0 per cent of the year's profit before tax. The Rs 18,600 crore of interest sitting on Suvarna Commercial Bank Limited's account covers twelve months rather than three, and so does the Rs 2,250 crore of profit underneath it.

The household version is the one everybody has already lived. A cousin says they saved eighty thousand rupees. The figure impresses until it turns out to have been over three years, against somebody else's sixty thousand in one. The two numbers were never comparable and nobody was lying. Between two entries struck on different stretches of time, the gap belongs to the stretches and not to the institutions. The correction is not a footnote after the fact, because a footnote under a column that has already been read is a record of the fault rather than a fix for it. The period is decided at step three, before anything is typed.

Where does the base of a ratio have to be written down?

Step four, and the answer is: in the column heading, before anything goes underneath it. Not a column headed margin, but a column headed margin on earning assets, or a column headed margin on total assets, and never both readings in one column. A heading that names its base makes the commonest fault in these tables impossible, and a heading that does not makes the same fault invisible.

Watch one unchanged figure produce two different entries. Suvarna Commercial Bank Limited earned net interest income of Rs 7,440 crore in the stated year. Divide it by earning assetsThe assets that actually pay a lender interest, being its advances and its investments. Earning assets are a narrower base than total assets. of Rs 2,04,000 crore and the entry reads 3.65 per cent. Put the same Rs 7,440 crore over total assets of Rs 2,40,000 crore and the entry reads 3.10 per cent. The two sit about half a percentage point apart. Nothing about the bank changed between those two readings. The only thing that changed was the denominator somebody chose, and if the heading does not say which one they chose, a reader cannot tell that anything was chosen at all.

The unnamed base is why a column of net interest marginA lender's net interest income turned into a rate, against whichever asset base the heading names, over a stated stretch of time. Net interest margin is taken apart separately. figures collected from several institutions is one of the least trustworthy things that can be built. Each institution wrote a correct figure. The institutions did not all write it on the same base. Collecting correct figures is not the same activity as building a comparable column, and step four is where the difference gets settled.

Which columns have to sit in the table before return on equity does?

Step five is the step this whole procedure exists for. Return on equityProfit for the period divided by net worth. Return on equity is taken apart separately and in full. is return on assetsProfit for the period divided by the asset base, with the base named. Return on assets is covered separately. multiplied by leverageAssets divided by net worth. Leverage says how many rupees of book are being run per rupee of the institution's own money., so a table carrying only the product has carried one column where three belong. Build the two first. Put the product beside them and never instead of them.

Now the part that has to be said out loud rather than left for a reader to trip over. Both lenders in this record come out at 9.38 per cent on equity for the stated year. The two figures are equal for a reason rather than by coincidence. This record was built that way deliberately, so that a reader who ranks institutions on the one column that sorts finds out here, on two invented lenders, that the ranking cannot tell them apart. A figure that arrives identical in two rows with nothing said about it looks like somebody mistyped a cell, and a reader who settles on that explanation walks away having learned none of this.

Take the two routes apart. Suvarna Commercial Bank Limited turns Rs 2,250 crore of profit on Rs 2,40,000 crore of assets. The result is 0.9375 per cent, reported as 0.94. Its assets are 10.0 times its net worth of Rs 24,000 crore. Rukmini Finance Limited turns Rs 337.50 crore of profit on Rs 18,000 crore of book. The result is 1.875 per cent, reported as 1.88. Its book is 5.0 times its net worth of Rs 3,600 crore. Half the return per rupee of book against twice the leverage, and the two roads arrive at one figure.

The rounding will bite anybody who checks the arithmetic with the printed figures, so here it is printed rather than hidden. 0.94 multiplied by 10.0 is 9.40, and 1.88 multiplied by 5.0 is 9.40, and neither of those is 9.38. The reported returns on assets are roundings. The exact figures are 0.9375 per cent and 1.875 per cent. The first multiplied by 10.0 and the second by 5.0 both land on 9.375 per cent, and 9.375 is the figure that rounds to 9.38. Neither rounding is an error on its own. A table that prints 0.94, prints 10.0 times, prints 9.38 and leaves the reader to reconcile them misleads, so every entry of the decomposition below carries the exact limb beside the reported one.

Read the three columns across and the point lands without anybody having to argue it. The last column ties and the two before it do not, so a reader sees immediately that the agreement lives in the product rather than in the business. That is the whole reason the ordering rule exists: the product column is never built before the two columns it is a product of.

What goes into the columns a regulator decides?

Nothing at all, and that is step six. A table of lenders wants a column for what each one must hold as capital and a column for how far each one is allowed to lever. Both of those are set by the Reserve Bank of India at rbi.org.in, both differ between a bank and a finance company, and both move. The column is drawn, the authority goes inside the cell, and it is filled from the source on the day the table is actually used; a value is never typed into a table that will still be in use next year.

The empty cell protects against something other than staleness. A figure that is merely old announces itself: the sheet shows it was built two years ago, and it gets checked. A typed requirement does not announce anything. The typed requirement sits in the cell looking exactly as authoritative as it did on the day it was true, and on the day the rule moves it stops being old and starts being incorrect, with nothing on the screen to say so. A labelled empty row never goes out of date, and a filled one has no way of signalling when it did.

The same treatment covers every column a rule decides. The capital each institution type must hold, how far each may lever, what a supervised institution has to publish about itself and how often, and which periods it has to report on are all drawn as rows with the authority named inside and the value left blank. A reader gets something that can be filled from the source. A reader handed a stated minimum gets something that will quietly go wrong.

How is a published entry told apart from a worked-out one?

Step seven, and it takes about a minute to do and saves an argument that would otherwise arrive later. Mark what was published apart from what was worked out, in the table itself rather than in anybody's memory of building it.

Look at what actually sits in the built table. Rs 2,250 crore of profit after tax and Rs 2,40,000 crore of total assets are entries Suvarna Commercial Bank Limited published. So are its Rs 24,000 crore of net worth, and Rukmini Finance Limited's Rs 337.50 crore of profit, Rs 18,000 crore of book and Rs 3,600 crore of net worth. Six cells came off an account. Then there is 0.9375 per cent. A reader made that entry by dividing one of those cells by another and choosing which denominator to use. There are six of those too, one return on assets, one leverage and one return on equity for each institution. Twelve cells in the built table, and exactly half of them were published.

The marking matters because tables travel. A finished table will be forwarded, pasted into something else, and read by somebody who was not in the room when the denominator was chosen. A table that travels without the marking comes back with every entry treated as published. The assumed ones come back that way too. That is not a small problem, because the published cells can be checked against a source and the worked-out ones can only be checked against the reasoning that produced them, and a reader who cannot tell them apart will not check either.

Which direction is the finished table read in first?

Step eight, and it is a habit rather than a calculation. Read across a row before reading down a column. Reading down a column ranks institutions against each other. Reading across a row describes one institution and shows whether its own figures agree with each other. A row that does not hold together internally makes every column it appears in worse, and only reading across shows it. So the second has to happen first.

The two rows already built show it. Across Suvarna Commercial Bank Limited the row reads 0.9375 per cent of assets at 10.0 times, producing 9.375 per cent. Across Rukmini Finance Limited it reads 1.875 per cent at 5.0 times, producing the same 9.375 per cent. Both rows hold together. Down the last column the two are identical. Reading across has already shown that identity to be an agreement in the product and nowhere else. Reading across first is what turns an identical pair of numbers from a puzzle into a finding.

The everyday version is a set of shop accounts. Reading down gives which of four shops took the most money last month. Reading across one shop gives whether its takings, its stock and its wages are consistent with each other, and if they are not, that shop's number was never fit to be ranked against the other three in the first place.

What does a finished table have to say about itself?

Step nine, the last one, and it is an output rather than a caution at the foot. Write down what the table cannot settle, on the table. For the table built here, that line reads: one stated year, two institutions, both invented, no downturn and no second period anywhere in the record behind them.

Think about why that is a product of the work rather than a hedge against it. A table that names its own limits can be handed to somebody else. The person receiving it knows the shape of what they are holding. A table that looks complete cannot be audited by anybody who did not build it, so it gets used for whatever its columns seem to support. The limits line is the part of the table that survives being forwarded.

And it forces the honest sentence about this particular table. The arithmetic settles what each institution's return on assets, leverage and return on equity are for the stated year. The arithmetic settles nothing about which way of running a lender is better. A lender running at higher leverage is not thereby reckless, and a lender earning a wider spread is not thereby skilful. Each of these two is one invented institution in one year, and there is no failure, no cycle and no second period anywhere in the record, so neither can be evidence about what its way of doing business achieves. The table sets down the arithmetic, names the questions that arithmetic leaves open, and goes no further.

What does the whole table look like once it is built?

Here it is, with all nine steps applied. The written question at the top, both institutions selected on what they take in and give out, one stated year throughout, every base named in its heading, the three return columns in their proper order, the two requirement columns drawn empty, and every cell marked as published or worked out.

The written question, at the top of the table: for the stated year, what did each lender earn per rupee of its own book, how many rupees of book did it run per rupee of its own money, and what did those two produce together? Membership follows from that sentence and not the other way round. The mixing, recorded at the top: one of these two takes deposits and the other takes none, so the taking-in sides are not alike.

The limits of this finished table: one stated year, two institutions, both invented, no downturn and no second period in the record behind them, so it settles what each arithmetic is and nothing whatever about which way of running a lender is better.

Read down the last return column and the two institutions are identical. Read down the two before it and they disagree by a factor of two in both directions. Split the one column into the three it is made of and the agreement disappears. Step five puts the parts before the product for exactly that reason.

How does a table like this go wrong in practice?

Who actually builds one of these, and what do they do with it?

What is not covered here?