Analysing Banks & NBFCs

A manufacturer borrows so it can buy machines and raw material. The loan is a means to an end, and the balance sheet says so: borrowings sit on the liabilities side as an obligation, plant and inventory sit on the assets side as the things that earn.

For a lender, money is not the means. It is the raw material and the finished product at the same time. The deposit you place with a bank is the bank's liability, because it owes that money back to you on demand or on maturity. The loan it hands to a borrower is its asset, because the borrower owes that money to the bank and pays interest on it. The institution is a spread business: raise money at one rate, lend it at a higher rate, and carry the risk that some of what was lent does not come back.

That inversion is not a curiosity. It changes what every line means. A jump in liabilities at a manufacturer is a jump in risk; a jump in deposits at a bank is more raw material, and usually a sign the franchise is working. A jump in assets at a manufacturer is usually capex; a jump in advances at a bank is more credit risk taken. Reading the Balance Sheet taught the manufacturer's version of this statement. Everything below re-teaches it for a lender.

Debt-to-equity answers a real question about a manufacturer: how much of the business is funded by lenders who must be paid regardless of how trading goes. For a bank the same arithmetic carries no information. A bank is deliberately, structurally and legally leveraged — its own capital is a single-digit percentage of total assets, and if it were not, the business could not earn a worthwhile return on a spread of two or three percent. A screen that flags debt-to-equity above 2 as risky flags every lender in India, which tells you the screen is wrong, not that the sector is unanalysable.

EV/EBITDA breaks harder. Enterprise value is market capitalisation plus net debt, on the logic that whoever buys the whole business inherits its borrowings and its cash. For a lender, borrowings are inventory. Adding the deposit base to market capitalisation produces a number that describes nothing at all. EBITDA is worse: interest for a bank is not a financing cost sitting below the operating line, it is simultaneously the largest revenue line and the largest cost line. Strip interest out of a bank's earnings and there is almost nothing left to look at.

The same reasoning disposes of several other familiar tools. There is no gross margin, because there is no cost of goods sold. There is no cash conversion cycle in the sense taught in Working Capital & the Cash Conversion Cycle — no inventory days, no payable days. Free cash flow is not informative either, because a growing loan book consumes cash by design. A lender throwing off large free cash flow is usually shrinking.

A lender's top line splits in two. Interest earned comes from advances and from investments. Interest expended is paid out on deposits and on borrowings. The difference between them is net interest income, or NII — the rupee value of the spread. Alongside NII sits other income: fees on cards and accounts, commissions on distributing insurance and mutual funds, and treasury gains on the bond portfolio. Treasury gains in particular are volatile and depend on where bond yields moved during the quarter, so they are read separately from the core spread.

NII on its own says nothing about efficiency, because a larger bank will always produce a larger number. Net interest margin fixes that by expressing NII against the average interest-earning assets that produced it. It is expressed on average interest-earning assets rather than on total assets or on the loan book alone, because a bank's investment portfolio also earns interest but its branches and fixed assets do not.

Three things move NIM. The mix of the loan book, since unsecured personal and microfinance lending price far higher than home loans against property. The cost of funds, which is where CASA comes in. And the position in the rate cycle: a bank whose loans reprice faster than its deposits sees margin expand when rates rise and compress when they fall, while a bank funded with short-dated deposits and lending at long fixed rates experiences the reverse. Which way a particular lender runs is a fact about its own book, not a rule of the sector.

CASA stands for current account and savings account — the share of a bank's deposits sitting in its two cheapest buckets. Current accounts, used by businesses for daily settlement, pay no interest at all. Savings accounts pay a modest regulated-then-deregulated rate. Term deposits pay materially more, because the depositor is explicitly buying yield and is willing to lock the money up to get it.

In Indian banking, CASA is the closest thing to the structural moat described in Economic Moats. It is built out of branch density in the right catchments, salary-account relationships with employers, cash-management mandates from corporates, and the plain inertia of a primary bank account that has every standing instruction attached to it. It is slow to build, which is why it is also slow to lose.

The consequence runs through everything else. A bank funding 40% of its deposits through CASA and one funding a quarter of them that way are paying different prices for the same rupee. Across an entire book, that gap is the difference between a lender that can lend to the safest borrowers and still earn a spread, and a lender that has to reach further up the risk curve to earn the same margin. This is why funding cost and asset quality are connected rather than independent: an expensive liability base pushes a lender towards riskier lending, and the bill for that arrives years later as credit cost.

A loan stops being a performing asset when the borrower stops servicing it. Ninety days of overdue interest or principal is the standard classification trigger in Indian banking, at which point the loan becomes a non-performing asset and the bank must set aside a provision against it. Provisioning rises the longer an account stays bad and depends on whether it is secured.

The headline number is the gross NPA ratio: bad loans as a percentage of the total loan book. The net NPA ratio strips out the provisions already made, so it answers a different question — how much of the damage is still uncovered by money the bank has already set aside. Provision coverage ratio expresses the same relationship from the other end: what share of the bad book is already provided for. A lender with high gross NPAs and high coverage has recognised its problem; a lender with low gross NPAs and thin coverage has less cushion for the next one.

The stock numbers, though, are the least informative part of the disclosure. Watch the flows: slippages are fresh loans that turned bad in the quarter, upgrades and recoveries are accounts that came back, and write-offs are loans removed from the books entirely. Gross NPAs can fall for three completely different reasons — because slippages genuinely slowed, because bad loans were written off, or because the loan book grew fast enough to shrink the ratio while the rupee amount rose. Restructured or resolution-framework accounts are a fourth category: still performing on paper, on renegotiated terms, and worth tracking separately because a portion of them typically slips later.

Gross NPA is a photograph of the balance sheet on one date. Credit cost is the video, and it runs through the profit and loss account. It is the provisioning charge plus write-offs taken during the period, expressed against average advances — in other words, what asset quality actually cost the shareholder this year.

This is the number that connects the credit book to earnings. A lender's pre-provision operating profit can be perfectly stable while reported profit collapses, because everything happened in the provisioning line. Conversely, a quarter in which provisions are released or written back shows a profit jump that has nothing to do with the lending franchise getting better.

Because credit cost is a flow and gross NPA is a stock, the two routinely move in different directions in the same quarter, and reading only one of them produces the wrong conclusion.

Capital adequacy ratio measures a lender's own capital against its risk-weighted assets. Risk weighting matters: a home loan secured against property and an unsecured personal loan of the same size do not consume the same amount of capital, because they do not carry the same probability of loss. Capital itself is split into tiers — Tier 1 is equity and reserves, the capital that absorbs losses while the bank keeps operating, and Tier 2 is subordinated instruments that absorb losses later in the queue.

A regulator sets a floor because the people funding a bank are depositors, and a depositor is in no position to assess a loan book. A bank failure also does not stay contained; it moves through the payment system and through other lenders' exposures. So the RBI prescribes a minimum, expressed as a percentage of risk-weighted assets and set in the low double digits for commercial banks, with the exact floor differing by the type of institution and changing as regulation evolves. Read the current norm from the RBI rather than a figure quoted in an article, including this one.

The practical consequence for an analyst is that capital is a growth constraint, not just a solvency test. Every additional rupee lent consumes capital. A bank operating close to the floor has only two options: slow lending, or raise equity. Raising equity dilutes existing shareholders, and if it is raised below book value it reduces book value per share. That is why a capital-raise announcement is a fundamental event for a lender and not merely a treasury one.

Profitability & Return Ratios and DuPont Analysis established that ROE can be split into margin, asset turnover and an equity multiplier. For a lender, that third term does most of the work. A bank funds a very large asset base with a thin sliver of its own capital, so even a modest return on assets is multiplied into a respectable-looking return on equity.

Take the illustrative lender used through this topic. Profit after tax of ₹1,800 cr on total assets of ₹1,50,000 cr is a return on assets of 1.2%. The same profit on equity of ₹12,000 cr is a return on equity of 15%. The entire gap between 1.2% and 15% is the balance sheet's gearing of roughly 12.5 times — not a single rupee of it is operating skill.

That is the problem with reading ROE first for a lender. ROE can be lifted simply by running thinner capital, which is another way of saying it can be lifted by taking more risk, until the regulator or a credit cycle interrupts. ROA cannot be manufactured that way. It asks the only question about operating quality that survives leverage: for every ₹100 of assets carried, how much does the lender actually keep? Note that ROA is read on a completely different scale for a lender than for a manufacturer — a figure that would look feeble for a factory is a normal outcome for a bank, so compare it with the lender's own history and its peer set rather than against a cross-sector benchmark.

Valuation Ratios introduced P/E and P/B as general tools. For lenders the ranking between those two reverses, for two reasons.

The first is that book value means more for a bank than for a manufacturer. A factory's assets are machines carried at historical cost less depreciation, a number that can drift far from what the machines are worth. A lender's assets are financial contracts, classified, provisioned and disclosed under a regulatory framework. Net worth is therefore a closer approximation of economic reality, and the price the market pays relative to it is a meaningful ratio.

The second is that a lender's earnings are provisioning-dependent. How much to provide against a stressed account involves judgement about eventual recovery. A year of light provisioning inflates profit; a year of clean-up crushes it. A P/E computed on either year describes the provisioning cycle rather than the franchise. Book value absorbs the same events, but gradually.

P/B is not read alone, though. What justifies one lender trading at a higher multiple of book than another is the return it sustainably earns on that book — which loops straight back to ROA, leverage and the cost of funds. Two banks with identical book value do not have identical value if one funds itself more cheaply and loses less on its lending. For NBFCs the same framework applies, with an additional layer of scepticism about how conservatively stress has been recognised, since the book is only as good as the provisioning policy behind it.

A non-banking financial company does the lending half of a bank's job without the deposit half. It has no savings-account franchise and therefore no CASA. It funds itself in wholesale markets instead: term loans from banks, non-convertible debentures, commercial paper, securitisation of its own loan pools, and external borrowings.

That funding model has two consequences, and the second one is the dangerous one. First, the cost of funds is set by the market's current view of the NBFC and its credit rating rather than by a sticky retail base — when credit spreads widen, an NBFC's funding cost rises quickly and its margin compresses, often before anything has gone wrong with its borrowers. Second is asset-liability mismatch. Writing five-year vehicle loans funded partly with one-year borrowings means the lender must keep refinancing. In normal conditions this is routine and profitable, because short money is cheaper than long money. In a liquidity squeeze the refinancing window shuts, and the lender faces a payment it cannot make even while every one of its own borrowers is paying on time. That is how a lender can fail while remaining solvent on paper.

Banks have two buffers an NBFC does not: a deposit base that does not run in normal conditions, and access to the central bank's liquidity windows as a regulated bank. This is why a system-wide liquidity event reaches NBFCs first and why NBFC valuations react to credit-market stress before bank valuations do.

So an NBFC is read with everything above plus four additional disclosures: the asset-liability maturity statement and where the negative gaps sit, the funding mix by instrument and how concentrated it is, the share of short-term borrowings such as commercial paper, and the liquidity buffer — cash, liquid investments and undrawn bank lines — against the next few months of repayments. How to Read an Annual Report covers where those disclosures sit in the document.

The table below is the practical summary of everything above: which measures transfer across the three business types, and which have to be replaced. The point of the middle two columns is not that lenders are harder to analyse — it is that a metric applied to the wrong business model produces a confident, precise, wrong answer.

Loan growth can look steady for years while a lender's earnings swing violently, because income and loss are separated in time. A loan written today produces interest income immediately. If it is going to go bad, it usually does so two to four years later. The cost of a lending mistake is booked long after the growth it produced has been celebrated.

That lag is the whole shape of the credit cycle. In the easy phase, competition for lending is intense, underwriting standards loosen at the margin, credit costs are low because the loans that will fail have not aged yet, and reported returns look excellent. In the tightening phase, slippages rise, provisioning jumps, and profit falls far faster than the loan book does — often while the loan book is still growing.

The reading discipline that follows is uncomfortable: the best-looking quarter in a credit cycle is the one with the lowest credit cost, and that is also the quarter in which newly written risk is hardest to see. Growth Analysis & CAGR made the general point that the path matters more than the endpoints. For a lender the path is the credit cycle, and a five-year CAGR that happens to span only the easy phase describes a phase, not a business.