How to Analyse a Bank Stock (NIM, CASA, GNPA, ROA)

Across the calls — from the June 2024 call to the March 2026 one — Axis Bank's management gave analysts the same number for its "through-cycle" net interest margin: 3.80% (Axis Bank Q1 FY25 through Q4 FY26 concalls; Inve data). It was not evasion. It was a structural claim — a statement that, across the rate cycle, the spread the bank earns settles around 3.8%. And here is what makes it worth your attention: over the very same window, the actual NIM walked down from 3.99% in Q1 FY25 (a 19 bps cushion over the 3.80% through-cycle call) to 3.62% by Q4 FY26 (Inve data) as the Reserve Bank cut rates and the bank's repo-linked loans repriced faster than its deposits. The reported margin fell 37 basis points; the guidance never moved. (Illustration of how to read the numbers, not a view on the stock.)

That gap — between a structural through-cycle number and the noisy quarterly print sitting under it — is the whole craft of reading a bank. A bank does not make money the way a factory does. It makes a thin spread on an enormous balance sheet, and then multiplies that spread with leverage. So almost everything you need to know lives in five or six numbers: how wide the spread is, how cheaply it funds itself, whether the loans are being paid back, and how much cushion stands between the depositor and a loss.

This is how to read a bank the way a credit analyst does — the metrics that decide the outcome, where each one hides (some are in the income statement; the ones that matter most are buried in the investor deck and the concall), what "good" looks like, and why you should never, ever value a bank on its P/E. The boundary first: you will not fully underwrite a loan book from the outside. What you can do is read the direction of the spread and the asset quality, and check whether management's account of them survives the Q&A.

Why a bank is leverage on a spread, not a "growth" business

Strip a bank to its engine and it is this: it raises deposits cheaply, lends them out at a higher rate, keeps the difference, and sets aside money for the loans that go bad. The leftover is profit. The trick is that it does this on assets worth eight to twelve times its own equity — a bank is regulated leverage applied to a small, steady spread.

That reframing kills two beginner errors at once. First, loan growth means little on its own. A bank can grow advances 25% by lending to weaker borrowers or by chasing deposits at any price; both load the future with cost that won't show up for six to eight quarters. Second, the income statement is the thin slice, not the business — the balance sheet is the business. Sales and net profit (the lines you'd anchor on for any other company) tell you almost nothing here; SBI earned ₹20,508 crore of net profit in Q4 FY26 (Inve data, Q4 FY26), but that number is meaningless until you know the spread that produced it and the asset quality underneath it.

So the question is never "is it growing?" It is: is the spread real, is the funding cheap and sticky, is the book clean, and is there enough capital to absorb the loans that will inevitably go bad? Five families of metrics answer that. Here they are.

NIM and CASA: is the spread real, and is the funding cheap?

Net interest margin (NIM) is the spread itself, expressed against the assets that earn it.

It is the single most important profitability number for a bank, and it's usually reported in the results and the investor presentation. What's "good" depends entirely on the kind of bank. A large universal bank like Axis runs a NIM in the high-3s — 3.62% in Q4 FY26 (Inve data). A small-finance bank lends to thinner, higher-rate borrowers and runs far wider: AU Small Finance Bank reported a NIM of 5.96% in Q4 FY26 (Inve data). Neither is "better" — they are different books. Read NIM against the bank's own history and against direct peers, and watch which way it's trending as rates move, because in a rate-cut cycle a bank with lots of floating-rate loans sees its yield fall before its deposit costs catch up.

CASA ratio — current-account and savings-account deposits as a share of total deposits — is why the spread is cheap. Current accounts pay no interest; savings accounts pay little. A bank with a high CASA ratio funds itself for less than one chasing bulk fixed deposits, and that funding cost is the denominator of the whole spread. SBI ran a CASA ratio of 39.46% in Q4 FY26 (Inve data) — among the stickiest deposit franchises in the country, which is the real moat a bank can have. AU, still building its liability franchise, sat at 28% (Inve data, Q4 FY26): it earns a wide NIM partly because it pays up for deposits, which is a different, more fragile way to make the same spread. CASA is in the results, but the trend and the cost-of-deposits detail behind it usually live in the investor deck.

Think of CASA as the bank's tap water versus bottled water. A high-CASA bank drinks from a cheap, always-on tap; a low-CASA bank buys its funding by the bottle, at a price that spikes exactly when it's thirstiest. When liquidity tightens, the bottled-water bank's margin is the first to crack.

GNPA, NNPA, PCR, credit cost and slippage: is the book clean?

This cluster decides whether the spread you admired is real or borrowed from the future. Five numbers, and they only mean anything read against each other.

Gross NPA (GNPA) is the stock of bad loans on the book — the share of advances where borrowers have stopped paying. SBI reported GNPA of 1.49% in Q4 FY26; Axis 1.23%; ICICI's was lower still (Inve data, Q4 FY26 / Q3 FY26). For a large bank, GNPA under ~2% is healthy; the level matters less than the direction.

Net NPA (NNPA) is GNPA after subtracting the provisions already set aside against it — the bad loans the bank has not yet absorbed into its accounts. This is the one that can hurt you, because it's the unprovided hole. ICICI's NNPA was just 0.37% in Q3 FY26 and SBI's 0.39% in Q4 FY26 (Inve data) — meaning almost every rupee of their bad loans is already provided for.

Provisioning coverage ratio (PCR) is the bridge between the two: the share of GNPA already provided against. SBI's PCR was 74.36% and ICICI's 75.4% (Inve data, Q4 FY26 / Q3 FY26). A PCR above ~70% means the bank has been honest about the losses already on its book; a low PCR with rising GNPA means the pain is being deferred.

Credit cost is the flow — the provisions charged to this quarter's profit to cover expected losses, as a percentage of the book. It's the price of bad lending recognised now. SBI has guided to "50 basis points… through the cycles" for several quarters running (SBI Q4 FY26 concall); Axis printed a net credit cost of just 0.37% in Q4 FY26 (Inve data). Credit cost is partly in the results, but the forward guidance — the number you can hold management to — lives in the concall.

Slippage ratio is the leading indicator: the rate at which performing loans turn into NPAs, annualised. It's where next quarter's GNPA is already forming. SBI guided to "contain slippages below 0.6%" (SBI Q1 FY26 concall); HDFC Bank reported slippages of about 24 basis points ex-agri in Q2 FY26 (HDFC Bank Q2 FY26 concall). Slippage is the number most often buried in the deck, not the income statement — and it's the one that turns first.

Here is the relationship that matters. GNPA tells you about the past; slippage tells you about the next two quarters; credit cost tells you what's being recognised right now; and PCR tells you whether what's already gone bad has been honestly absorbed. The dangerous pattern is a flat GNPA sitting on top of rising slippage and falling credit cost — losses forming faster than they're being recognised, with the catch-up landing in one brutal quarter. A clean book is one where these four move together and management can quote you each of them without hedging.

CRAR, Tier-1 and the credit-deposit ratio: is there a cushion, and is it stretched?

Capital adequacy (CRAR) is the equity cushion that absorbs losses before depositors are touched. The regulatory floor for most banks is around 11.5% including buffers, but the floor isn't the interesting part — the trend and the quality are. SBI's CRAR strengthened 115 bps year-on-year to 15.4% in Q4 FY26 (SBI Q4 FY26 concall), and Axis carried a CET-1 (the purest, equity-only tier) of 14.43% in Q1 FY26 (Inve data). Tier-1 / CET-1 is the number that truly absorbs losses — a bank leaning on Tier-2 subordinated debt to flatter its headline CRAR is thinner than it looks. A bank growing its book 20% a year on a 12% CRAR is heading for a dilutive capital raise; the timing of that raise (opportunistic versus forced) tells you a great deal about management.

The credit-deposit ratio (CD ratio, or loan-to-deposit ratio, LDR) is how much of its deposits the bank has lent out — a gauge of how stretched the funding engine is. HDFC Bank's LDR ballooned past 100% after it absorbed HDFC Ltd's mortgage book, and management has spent two years guiding it back down: "we will try and be in a range of somewhere between 90% to 96% in FY26… by FY27, with natural growth, we should" get below 90, toward the pre-merger "87%, 88%" (HDFC Bank Q2 FY26 concall). A high CD ratio means future loan growth must be funded by chasing deposits — which lifts cost of funds and squeezes NIM. This number is in the deck, and the guided glide-path is in the concall.

ROA, ROE and cost-to-income: does the machine actually compound?

Two summary numbers tie the whole thing together. Return on assets (ROA) is profit as a percentage of the total balance sheet — the cleanest measure of how well a bank runs its core engine, because it's before the leverage. For a large Indian bank, an ROA around 1% is respectable and above 1.6–1.8% is excellent: SBI runs "ROA consistently greater than 1%" (SBI Q4 FY26 concall), Axis posted a consolidated ROA of 1.64% in Q4 FY26 (Inve data), and AU targets 1.8% by FY27 (AU Small Finance Bank Q4 FY26 concall).

Return on equity (ROE) is ROA multiplied by leverage — the return to the shareholder. SBI guides to "15% return on equity to our investors, minimum… through the cycles," and reported 18.5% exiting FY26 (SBI Q4 FY26 concall). The relationship is the lesson: ROE = ROA × (assets/equity). A bank can manufacture a high ROE with thin capital and high leverage, which is exactly why ROA — the un-levered number — is the more honest one, and why CRAR sits right next to it in your reading.

Cost-to-income is operating efficiency: how much of each rupee of income is eaten by running the bank. SBI aims "to keep cost-to-income below 50" (SBI Q4 FY26 concall) — a large, scaled bank's advantage. A small-finance bank carries a far heavier cost-to-income — AU targets "below 60%" (AU Q4 FY26 concall) — because it's still building branches and systems. The number is derivable from the results; the target is in the commentary.

How to value a bank: P/B against ROE, never P/E

Now the part most retail investors get backwards. You do not value a bank on P/E. You value it on price-to-book against ROE.

Here's why. For an industrial company, earnings are the thing and book value is an accounting artefact. For a bank, book value (equity) is the engine — every rupee of capital is what the bank levers up to lend, and the regulator caps how far it can stretch that capital. So the right anchor is: what return does the bank earn on its book, and what should you pay for that book? The cleanest framing is a Gordon-growth lens — the justified price-to-book rises with the spread between ROE and the cost of equity. A bank earning a sustainable 18% ROE deserves to trade at a meaningful premium to book; a bank earning 10% deserves roughly book value or less. Two banks on the same P/E can be worlds apart if one earns 18% on equity and the other 11% — the P/E hides exactly the variable that matters.

This is why a bank's quality shows up as a P/B multiple, and why ROA is the number to defend in your model. SBI's "≥1% ROA, ≥15% ROE through the cycle" framing (SBI Q4 FY26 concall) is, in effect, management telling you the book should compound at 15% and pricing the stock is a question of what multiple of that compounding you'll pay. When you see a bank trading at a low P/B, the only useful next question is: is the low multiple because the market doubts the ROE is sustainable (a real warning), or because the book is genuinely cheap relative to a durable return (an opportunity)? The P/E tells you nothing about either.

A worked case: the through-cycle number that didn't flinch

Go back to where we started. Across the calls in that window — Q1 FY25 (June 2024), Q3 FY25, Q4 FY25, Q1 FY26, Q3 FY26, and Q4 FY26 — Axis Bank's management held its through-cycle NIM guidance at exactly 3.80% (Inve data; Axis Bank concalls). The wording barely changed: "it is rate cycle agnostic, which is why we say it's a through cycle NIM guidance. We are not walking away from that even today" (Axis Bank Q3 FY26 concall), and at the FY25 call, "FY25 NIM at 3.98%, 18 bps cushion over our through-cycle NIM call-out of 3.80%" (Axis Bank Q3 FY25 concall). (Illustration, not a view on the stock; figures as reported in the company's concalls and the KPI record.)

Read the columns against each other. The reported NIM did exactly what a rate-cut cycle does to a bank with repo-linked loans — it fell, from 3.99% to 3.62%, a 37 bps slide. The guidance did not move, because it was never a forecast of next quarter; it was a claim about where the spread settles across the cycle, once deposits reprice down to meet the loans. The discipline this teaches: distinguish a structural through-cycle number from a quarterly print, and watch whether the gap between them is closing for cyclical reasons (deposits lagging, as here) or for ominous ones (the bank reaching for risk to defend a headline). Axis's own forward markers — credit growth "300 bps above market," a standard-asset provision write-back glide-path, no equity raise needed — were largely marked achieved or on-track in Inve's Promise Tracker (Inve data), which is the corroboration that the margin slide was the rate cycle, not a deteriorating franchise. The number to trust was the direction of the spread and the asset quality under it, not the unchanging headline — and not the quarterly NIM in isolation either.

Red flags specific to banks

• A flat GNPA sitting on rising slippage and falling credit cost. The cosmetic clean book: bad loans held down by write-offs and restructuring while the leading indicator (slippage) and the provisioning honesty (credit cost, PCR) quietly deteriorate. Losses forming faster than they're recognised — the reckoning arrives in one quarter.
• NIM held suspiciously steady while system rates fall and the loan book is floating-rate. Either genuine pricing power, or the bank is reaching for higher-yield, higher-risk borrowers to defend the headline. The mix tells you which.
• A credit-deposit ratio climbing toward or past 100%. Future growth must be bought with expensive deposits; NIM is about to be squeezed, or the bank slows down.
• ROE flattered by thin Tier-1 capital. A high ROE on low CET-1 is leverage doing the work, not the franchise — and it's one bad cycle from a dilutive raise.
• CASA quietly eroding while management talks up loan growth. The funding base is getting more expensive at exactly the moment the book is expanding — the spread is being borrowed from the future.

The same instinct that distrusts a suppressed credit cost is what separates reported profit from real cash earnings in any business — and the lending discipline overlaps heavily with how to analyse an NBFC, where the funding mismatch, not the bad loan, is what usually kills the lender.

Where this lens can be wrong

The strongest case against everything above is that a clean concall and a tidy set of ratios told you nothing in 2018, when IL&FS and DHFL took down genuinely profitable, well-provisioned lenders in a system-wide liquidity freeze. A bank can read perfectly on NIM, CASA, GNPA, PCR and CRAR and still be carrying a concentration risk, a treasury bet, or a contingent exposure that no quarterly metric surfaces — and the very metrics this guide leans on are the ones a determined management can dress for a quarter or two. Reading the spread against the asset quality lowers your odds of owning the worst house on the street; it does not tell you when the whole street floods. And there is a subtler trap: a bank that always reiterates the same through-cycle number sounds disciplined, but if you only ever reward the steady guide you'll mistake a bank quietly losing its franchise for one holding its nerve — the only defence is to check the guidance against what the book actually did, quarter after quarter. Tracking that by hand across a portfolio of ten banks, every results season, is precisely the work nobody has time for; Inve's Promise Tracker pins each forward commitment — a NIM guide, a credit-cost target, an ROA goal — to the quarter and quote it was made in, then marks it as later calls confirm or contradict it, and the KPI Screener lines up NIM, CASA, GNPA and slippage across banks with the trend and a data-confidence flag per number.

Frequently asked questions

The discipline comes down to refusing to be impressed by the headline profit. The balance sheet is the business, and it speaks through the spread (NIM, CASA), the asset quality (GNPA, NNPA, PCR, credit cost, slippage), the cushion (CRAR, Tier-1) and the efficiency (ROA, ROE, cost-to-income). So invert the question you bring to a bank's results. Don't ask "was this a good quarter?" Ask: if this management were quietly funding tomorrow's losses to flatter today's profit, what would the numbers look like — and does this book rule that out? A flat GNPA over rising slippage and a falling credit cost does not rule it out; it is the pattern itself.

And the owner's question, the one to sit with before you buy a share of any bank: what must I believe about the next downturn — not this quarter's NIM — for this bank to still be earning its cost of capital, still funded by cheap sticky deposits, and still compounding its book on the other side of it? If the honest answer leans on an unchanging through-cycle guidance rather than on what the book has actually done, you've read the headline, not the bank.

Related reading

• How to Analyse a Building Products Company (Pipes, Tiles)
• How to Analyse a Cables and Wires Stock
• How to Analyse a CDMO / CRAMS Pharma Company