How a Bank Makes Money: CASA, NIM and Bad Loans

When I was a kid, the man two doors down ran a small money-lending business out of his front room. He'd explain it to anyone who'd sit still: he borrowed from people wanting a safe place to park their savings, paid them a little, and lent it onward to shopkeepers who needed working capital, charging a lot more. The gap was his living. "But the real skill," he'd say, tapping his ledger, "is not lending to the fellow who'll never pay me back. Anyone can hand out money. Getting it back is the job."

That front room is a bank. Every bank in India — from your neighbourhood branch to the largest private lender on the NSE — runs that exact business, just with more zeros and a regulator watching. See it that way and a bank stops being a mysterious machine and becomes something you can judge. So let's judge one.

A bank buys money cheap and sells it dear

Here is the entire business in one sentence: a bank takes in deposits (money it owes you) and lends them out as loans (money owed to it), and lives on the difference in interest. Its core profit, before costs, is the spread between what it earns on loans and what it pays on deposits. That's it.

The skill lives in three levers: (1) get your funding cheap, (2) lend enough of it, and (3) lend only to people who pay you back. Get all three right for long enough and you have one of the best businesses on earth. Get the third wrong and you have no business at all.

Let me walk through the three with a real bank, so the words attach to numbers. I'll use HDFC Bank — India's largest private-sector lender, one almost every reader already banks with or knows. (This is a worked example of how to read a bank, not a view on whether to own it.) Other large private-sector lenders worth reading the same way include ICICI Bank, Kotak Mahindra Bank and Axis Bank.

Lever one: cheap money, and the word "CASA"

Not all deposits cost the same. (The rates below are illustrative of typical Indian retail levels and move with the rate cycle, not fixed numbers.) A fixed deposit pays you 7%-odd to keep your money locked up. But the cash sitting in your current account (for a business) or your savings account earns next to nothing — a current account pays zero, a savings account around 3%. Banks call these two together CASA — Current Account and Savings Account — and they are the cheapest money a bank can get.

So a bank with lots of CASA buys its raw material — money — at a discount. A bank funded mostly by fixed deposits is like a shop paying full price for stock; a CASA-funded bank buys wholesale. That's why CASA is the first thing analysts check.

It's exactly the lever HDFC Bank's management keeps pointing at. On its Q3 FY26 earnings call (concall), management guided that CASA growth would "beat industry average over time" and borrowing cost would "keep coming down" (Inve data, 2026). Read what they're really saying: we intend to keep our money cheap. For a bank, that's close to the most important sentence there is.

Lever two: the spread, called net interest margin (NIM)

Now the spread. Take what a bank earns in interest on its loans, subtract what it pays on its deposits, divide by the loans it has out. That percentage is its net interest margin, or NIM — the bank's gross profit margin. A NIM of 3.5% means it keeps three and a half rupees of spread for every hundred rupees lent. Cheap CASA and a healthy NIM are two sides of one coin: the cheaper your money, the wider your spread at the same lending rate.

Back in Q4 FY25, HDFC Bank's management told investors to expect "a little bit of a trough in the coming months before it starts to pick up and stabilize" — they warned NIM would dip before recovering, and a year on our records mark that guidance as on track (Inve data, 2026). Notice the honesty of a down-then-up shape rather than a rosy straight line; that's the kind of specific, checkable guidance an owner wants.

Here's the spread at scale. In FY26 HDFC Bank earned about ₹3,48,615 crore of interest income and ₹79,219 crore of net profit (Inve data, 2026). Slow down on that profit: roughly ₹79,000 crore, larger than the entire annual revenue of most companies you can name, squeezed from a spread of a few percentage points — because the balance sheet it's applied to is enormous. That is banking's magic and its menace in one line: a thin margin on a giant, borrowed pile.

Lever three — and the one that kills banks: bad loans

My neighbour's rule comes back here, because it's the one that matters. A bank can have the cheapest CASA in the country and a beautiful NIM and still die — by lending to people who don't pay it back.

A loan that stops being repaid — no interest, no principal, for 90 days — becomes a non-performing asset, or NPA (a "bad loan"). Add up the bad loans, divide by total loans, and you get the gross NPA ratio: the share of the book that has gone sour. This is the number that has buried banks throughout history.

Why does a small-sounding percentage matter so much? Leverage. A bank lends out far more than its shareholders' capital — it lends mostly your deposits. Warren Buffett, explaining why he'd buy a bank only when he trusted the managers completely, put it exactly right in his 1990 letter: "Because leverage of 20:1 magnifies the effects of managerial strengths and weaknesses, we have no interest in purchasing shares of a poorly-managed bank at a 'cheap' price." (Berkshire Hathaway, 1990 letter). At 20-to-1, a 5% hole in the loan book can wipe out the entire equity. The thin spread that makes a bank rich in good times is the same thin cushion that makes it fragile in bad ones.

The good news for the sector now: Indian banks are unusually clean. The RBI's Financial Stability Report found the gross NPA ratio of scheduled commercial banks fell to 2.6% — a 12-year low — with net NPAs at about 0.6% (RBI Financial Stability Report, via Outlook Business, Dec 2024). After the bad-loan crisis of the late 2010s, the system has healed. But read that the other way too: a 12-year-low NPA print is as likely to be a cycle peak — the cleanest the books will look before the next downturn surfaces loans made in good times — as it is a permanent new normal. And "the system is clean today" is exactly the moment to remember it wasn't always — and won't always be. Yes Bank had to be rescued in 2020 when hidden bad loans finally surfaced and its capital evaporated almost overnight; the public-sector banks spent the late 2010s drowning in NPAs that earlier reports had not shown. A low ratio is a snapshot, not a guarantee.

Putting it together: how to read a bank in five numbers

Look at a bank as a part-owner, not a depositor, and here's the cheat sheet:

• CASA ratio — is the funding cheap?
• Net interest margin (NIM) — is the spread healthy and stable?
• Loan growth — is the book growing sensibly, not recklessly?
• Gross NPA ratio — how much of the book has gone bad? (The one that kills.)
• Return on equity (ROE) — does it all add up to a good return on shareholders' capital?

On loan growth, watch the pace — fast growth and bad loans are old friends. A bank growing its book at 30% a year is either brilliant or about to discover it lent to the wrong people. HDFC Bank's management gave an unusually specific roadmap: loan growth "lesser than the system" in FY25, "in line with the system" in FY26, "faster than the system" in FY27 — a staged re-acceleration after digesting its mega-merger, which our records mark as on track (Inve data, 2026). For a lender, that restraint is usually a feature, not a flaw.

One word on price. As of late April 2026, HDFC Bank trades at about 15 times FY26 earnings and roughly 2.1 times its book value (Inve data, 2026; multiples move daily). For banks, price-to-book — market value over shareholders' equity — is often a more natural yardstick than P/E, because a bank essentially is a pile of financial assets and liabilities. Whether 2.1x is cheap or dear is a separate question — one for how to value a bank — and not one I'm answering here. And cheap can be a trap: a 2.1x bank that hides bad loans is dearer than it looks, while a low multiple often prices in a credit cycle about to turn.

The one thing the levers can't show you: trust

One honest caveat before this becomes a checklist. I picked HDFC Bank precisely because it survived and compounded — it's the kind of bank these five numbers flatter. But survivorship cuts both ways: the same five numbers looked perfectly healthy, right up until they didn't, at banks that later had to be rescued. A clean scorecard is necessary, not sufficient.

Here's where my neighbour's ledger ends and judgement begins. Every number above is reported by the bank itself. The NPA ratio is the bank's own assessment of which loans have gone bad — and the history of banking blow-ups is the history of bad loans hidden, evergreened, or wished away on the call long before they showed up in the ratio. The number tells you what management chose to recognise, not what they hope you won't ask.

Which is why, with a bank more than almost any business, you read the commentary, not just the statements. Does guidance only ever move in the flattering direction? Do they answer the hard question on slippages, or deflect it for a third straight quarter? In HDFC Bank's own record, alongside the commitments it's tracking on CASA and growth, a few have quietly gone silent — guidance on its cost-to-income trajectory, its borrowing mix, and its capital runway each surfaced on a call and then dropped out of the conversation (Inve data, 2026). None is damning alone. But a bank is a 20-to-1 leveraged commitment to behave well, and the only way to judge whether management keeps it is to watch what they said, quarter after quarter, against what they did — the unglamorous, portfolio-wide tracking that Inve's Promise Tracker exists to do, because no one holding ten stocks can keep that ledger by hand.

If you've followed this series, the frame is familiar: a bank is a business you own a slice of, and its moat — a sticky, low-cost deposit base built over decades — is what protects the spread from competition. The levers are just how that story shows up in the accounts.

Frequently asked questions

Related reading

• Order Book: How a Capital-Goods Business Earns
• How a Cement Company Makes Money in India
• How a Pharma Business Earns Money: Brands vs Generics