Why you cannot value a bank like a technology company
The answer
A bank borrows money as its raw material and lends it out. So a bank's debt is its inventory, not a warning sign, and a debt-to-equity ratio of 8 or 10 is normal rather than dangerous.
Banks are judged on price to book value read against return on equity, and on 5 numbers: net interest margin, cost-to-income, gross and net non-performing assets, provision coverage and capital adequacy. Every industry has its own set. Using one industry's set on another produces a confident wrong answer.
Why this costs you money
Here is the single most common version of this mistake in India.
A reader learns that a debt-to-equity ratio above 2 signals danger. They run a screener across their holdings. A large private bank comes back with a debt-to-equity ratio near 8. They conclude the bank is dangerously leveraged and sell it.
That reader has just misunderstood what a bank is. Deposits and borrowings are how a bank obtains the money it lends. A bank with almost no borrowings would be a bank with almost no business. Applying a manufacturing rule to a lender does not give a slightly wrong answer. It gives an answer with no relationship to the question.
The same error runs in the other direction and costs just as much.
Somebody values a software company on price to book. The company owns almost nothing physical. Its value is in code, contracts, brand and people, and none of those appear on a balance sheet unless they were bought from somebody else. A software company at 15 times book is not expensive. Book value is simply not where its value sits.
Somebody applies EV/EBITDA to a bank. EBITDA means earnings before interest. For a bank, interest is both the main revenue and the main cost, so removing it removes the bank. The number that comes out is not a valuation of anything.
Somebody judges a cyclical steel producer on last year's profit multiple at the top of the cycle, and buys the cheapest-looking multiple in the market weeks before the cycle turns.
The specific loss: a portfolio built from a single screener, run with a single set of thresholds, across 5 different kinds of business. The screener works correctly. The thresholds are meaningless for 4 of them.
How it works
Start with the structural difference, because every metric follows from it.
In most businesses, debt is financing. A company borrows to build a factory. The factory produces goods. Interest is a cost of the money, not the business itself. That is why you can strip interest out (EBITDA) and compare 2 companies regardless of how they funded themselves.
In a lender, debt is the product. A bank takes deposits at, say, 5% and lends at, say, 9%. The 4 percentage point difference, adjusted for the cost of running the bank and the loans that go bad, is the entire business. Interest is not a financing item. It is revenue and cost of goods sold.
That single difference invalidates enterprise value, EBITDA, free cash flow as normally calculated, and every debt ratio, all at once. It also explains why book value works: a bank's assets are financial, they are carried at values close to what they are actually worth, and equity is the buffer that absorbs losses. The balance sheet is the business.
The 3 broad families
Family 1 — lenders and financial companies. Banks, non-banking finance companies, housing finance companies, insurers. Valued on book value. Judged on spread, cost, credit losses and capital.
Family 2 — capital-heavy producers. Steel, cement, chemicals, automobiles, power, mining, shipping. Valued on enterprise value multiples and on assets. Judged on capacity utilisation, asset turnover, cost position and where the cycle is.
Family 3 — asset-light service and technology businesses. Software, IT services, consumer platforms, consulting. Valued on earnings and cash multiples. Judged on growth, gross margin, revenue retention and cash conversion. Book value tells you nothing.
Family 1 in detail: how to judge a bank
Return on equity, and price to book. A bank's fair price to book follows almost mechanically from its return on equity. A bank that earns a 16% return on equity is creating value on every rupee of capital and deserves to trade above book. A bank earning 6% is barely covering the cost of that capital and deserves to trade below it. As a first approximation, justified price to book rises with return on equity and falls with the return investors require. So a bank at 3 times book with a 17% return on equity and a bank at 0.6 times book with a 5% return on equity may both be fairly priced. The multiple alone tells you nothing.
Net interest margin, written NIM. Net interest income divided by average interest-earning assets, as a percentage. It is the spread the bank actually achieves after paying for its money. A higher margin usually means either cheaper deposits or riskier lending. Which of the 2 it is matters enormously, and the next 3 numbers tell you.
Cost-to-income ratio. Operating expenses divided by total income. It measures how efficiently the bank runs. Lower is better. A bank with an excellent margin and a poor cost ratio is not efficient; it is expensive to run and charging customers for it.
Gross NPA and net NPA. A non-performing asset is a loan where interest or principal has been overdue beyond a set period — in India, generally 90 days . Gross NPA is bad loans as a percentage of total loans. Net NPA is the same figure after subtracting the money the bank has already set aside against those loans. Both matter, and the gap between them is the point. A bank with 4% gross NPA and 0.5% net NPA has recognised its problem and paid for it. A bank with 4% gross and 3% net has recognised the problem and not paid for it, which means future profits must absorb the loss.
Provision coverage ratio. Provisions held divided by gross non-performing assets, as a percentage. It is the same information as the gap between gross and net, stated directly. Higher means more of the expected loss has already been taken through the profit and loss account.
Capital adequacy ratio. Regulatory capital divided by risk-weighted assets. It answers 1 question: how much loss can this bank absorb before depositors are at risk? Under the Basel III framework in India the minimum total capital requirement is 9%, plus a capital conservation buffer of 2.5%, giving an effective requirement of 11.5%, with a minimum common equity tier 1 component . Read it alongside the tier 1 number, because tier 1 is the capital that truly absorbs losses.
CASA ratio. Current account and savings account deposits as a percentage of total deposits. These are the cheapest deposits a bank has, because current accounts usually pay no interest and savings accounts pay little. A high CASA ratio is a durable advantage: it lowers the cost of funds permanently, which supports a higher net interest margin without taking more risk.
Credit cost and slippage. Credit cost is provisions for the year divided by average loans. Slippage is the value of loans that turned bad during the year. These are flow measures. Gross NPA is a stock measure, and a stock can be flattered by writing loans off or selling them. Watch the flow.
Non-banking finance companies are not small banks
An NBFC lends but generally cannot take demand deposits from the public. So it funds itself in the market — bank loans, bonds, commercial paper. That creates the single risk that defines the sector: asset-liability mismatch. If a lender funds 15-year infrastructure loans with 90-day commercial paper, it must refinance every quarter. It is solvent right up until the moment the market declines to roll over its paper.
So for an NBFC add: cost of funds, the spread over cost of funds, the maturity profile of borrowings against the maturity profile of loans, and the share of funding that is short term. Capital adequacy applies too, at different thresholds.
Insurers need a third lens again
A life insurer's profit today tells you little, because it writes policies that generate profit over 20 years and takes the acquisition cost immediately. Life insurers are therefore judged on embedded value — the present value of future profits from existing policies plus net worth — and on value of new business and the new business margin. A general insurer is judged on the combined ratio: claims plus expenses divided by premiums. Below 100% means the underwriting itself makes money; above 100% means the insurer is relying on investment returns. Both are judged on the solvency ratio, which is the regulator's capital test.
Family 2: capital-heavy producers
Here EV/EBITDA and EV/EBIT are the right multiples, because financing choices differ and depreciation policies differ.
Add: capacity utilisation (how much of the plant is running), asset turnover (revenue divided by assets, which measures how hard the assets work), net debt to EBITDA (how many years of cash earnings the debt represents), and the company's position on the industry cost curve. In a commodity, the only durable advantage is being the low-cost producer, because nobody can charge more for an identical product.
And the cycle question: is current profit near the top or the bottom of the range of the last 10 years? For these businesses the multiple is at its lowest when the risk is at its highest.
Family 3: software, platforms and IT services
For a subscription software business, judge: annual recurring revenue and its growth, net revenue retention (what last year's customers spend this year, including upgrades and cancellations — above 100% means the existing base grows by itself), gross margin, and the payback period on customer acquisition cost. Book value is irrelevant. Free cash flow matters, but must be adjusted for stock-based compensation.
Indian IT services companies are a different business from subscription software, and the metrics differ. They sell people's time. Judge them on revenue growth in constant currency, employee utilisation, attrition, revenue per employee, the mix between fixed-price and time-and-materials contracts, the share of revenue that is renewable, and client concentration. Margins are driven by wage inflation, the rupee-dollar rate and the proportion of work done offshore. A subscription metric like net revenue retention does not apply.
What it tells you, and what it does not
Matching the metric to the business tells you whether a comparison is legitimate. It stops the largest category of analytical error, which is comparing 2 things that are not comparable.
It does not tell you the answer within an industry. Two banks with identical metrics can still have very different loan books, and the composition of the loan book — unsecured lending against home loans against corporate credit — may matter more than any ratio.
It also does not protect you from the deeper problem with lender analysis. A bank's reported profit depends on how much it set aside for loans that may not be repaid, and that provision is a management judgement. Gross NPA is a reported number, but the decision to classify a loan as non-performing involves discretion. Every metric in the lender family is downstream of that judgement. This is why the gap between a bank's own reported bad loans and a regulator's assessment is one of the most important disclosures in the sector.
The decision rule
Before you look at any ratio, name the family. Then use only that family's numbers, and compare only against companies in the same family.
If it is a lender:
- Never use debt-to-equity, enterprise value, EV/EBITDA or free cash flow.
- Start with return on equity, then price to book. Ask whether the multiple matches the return.
- Then the 5: net interest margin, cost-to-income, gross NPA, net NPA and provision coverage, capital adequacy.
- Then the flow: slippage and credit cost this year against the last 5 years.
- Then the funding: CASA ratio for a bank, maturity mismatch for an NBFC.
If it is a capital-heavy producer:
- EV/EBITDA and EV/EBIT, on mid-cycle earnings, not last year's.
- Net debt to EBITDA, capacity utilisation, asset turnover, cost position.
If it is asset-light:
- Growth, gross margin, retention, cash conversion.
- Ignore book value. Adjust free cash flow for shares issued to employees.
The single conditional that matters most for lenders. If net interest margin is rising and gross NPA is stable, that usually means the bank has improved its funding or its pricing, and it is a good sign — unless the loan book has shifted toward unsecured or higher-yield lending, in which case the margin rose because the risk rose, and the bad loans have simply not arrived yet. New loans do not default in year 1. A fast-growing loan book always looks clean.
Try this now
Five minutes. Do this on a bank you already hold, or on any bank in your index if you hold none.
- Open the bank's latest quarterly results or investor presentation on the NSE, BSE or SEC website. Banks publish these with the numbers already collected on 1 or 2 slides.
- Write down 5 figures: net interest margin, gross NPA %, net NPA %, provision coverage ratio, and capital adequacy ratio (with the tier 1 figure if shown).
- Add 2 more: cost-to-income ratio and return on equity.
- Now do the same for 2 competitors. Pick banks of a similar type — 2 large private banks, or 2 public sector banks. Do not mix the 2 groups.
- Look up the current price to book for all 3, and place it next to each bank's return on equity.
What you should see. Four things.
The bank with the highest price to book will usually be the bank with the highest return on equity. If one of the 3 has a high multiple and a low return, you have found something to explain — either the market expects the return to rise, or the multiple is not justified.
The gap between gross NPA and net NPA will differ across the 3. Convert it into the provision coverage ratio and rank them. The bank with the lowest coverage has the most loss still to take through future profits.
The bank with the best net interest margin is not automatically the best bank. Check whether its cost-to-income ratio is also good, and check what it lends against. A high margin from unsecured lending is a different animal from a high margin from cheap deposits.
And you will notice something absent: you never once used debt-to-equity, EV/EBITDA, or free cash flow. That is the lesson.
Three real cases
1. Yes Bank, March 2020 (India) — the reported book that was not the real book Yes Bank grew its loan book quickly through the 2010s and reported healthy capital and manageable bad loans for much of that period. From 2017 onward the Reserve Bank of India's asset quality assessments identified divergences between the bank's reported non-performing assets and the regulator's assessment . Bad loans rose sharply, capital fell, and on 5 March 2020 the RBI placed the bank under a moratorium and superseded its board. Under the reconstruction scheme that followed, additional tier 1 bonds with a face value of approximately ₹8,415 crore were written down entirely. Price to book is the right multiple for a bank. It is only as good as the book, and the book is a management estimate of how many loans will be repaid.
2. Infrastructure Leasing and Financial Services, 2018 (India) — the mismatch that no leverage ratio showed IL&FS was a large infrastructure financing group. It began defaulting on debt obligations from around June 2018, with a series of defaults through September 2018, and the government superseded its board on 1 October 2018. Group debt was reported at roughly ₹91,000 crore. The group had held high credit ratings shortly before the defaults. The core problem was not the size of the borrowing, which was normal for a lender. It was structure: long-dated infrastructure assets funded substantially with short-dated market borrowing, inside a group with a very large number of subsidiaries. A debt-to-equity ratio would not have shown this. A maturity profile would have.
3. Silicon Valley Bank, 10 March 2023 (United States) — good capital ratios, wrong risk Silicon Valley Bank was closed by regulators on 10 March 2023 in what was at the time the second-largest bank failure in US history. It had not made a large volume of bad loans, and its headline capital ratios did not signal distress. The problems were elsewhere: a very large portfolio of long-dated securities classified as held-to-maturity, which are not marked to market in reported capital, and whose value had fallen sharply as interest rates rose; and an unusually concentrated deposit base of technology companies, much of it above the insured limit, which could leave quickly and did. The lesson is not that bank ratios are useless. It is that the standard ratios are the beginning, and you must also read the deposit concentration and the securities portfolio.
The question that resolves it
A novice looks at a company's ratios and asks: is this number good or bad?
An expert looks at the same company and asks first: what kind of business is this, and therefore which numbers are even meaningful here?
Almost nobody gets the second question wrong once they have been told it exists. Almost everybody gets it wrong before that.
What would make this wrong
If matching metrics to industries reliably produced good investments, then anybody applying the correct bank framework would have avoided every bank failure. They did not. Analysts using exactly the right metrics have been wrong about banks repeatedly, because the inputs are estimates and the estimates are made by the people being analysed.
The honest limits: every lender metric depends on the provision, and the provision is a judgement. A bank can show excellent numbers until a regulator disagrees with its classifications. Capital adequacy is calculated on risk-weighted assets, and the risk weights are set by rules that may not reflect actual risk — Silicon Valley Bank held government-backed securities carrying low risk weights that nevertheless caused the failure. Price to book assumes book value is close to real value, which is true for a healthy bank and badly untrue for a failing one, exactly when you need it.
Also, the 3 families are a simplification. Real companies straddle them. A car manufacturer with a large captive finance arm is 2 businesses in 1 set of accounts, and consolidating them produces ratios that describe neither. When that happens, separate the segments and value each one on its own family's terms.
In India
The disclosure format is unusually good for lenders. The Reserve Bank of India requires banks and non-banking finance companies to report gross NPA, net NPA, provision coverage ratio and capital adequacy in a standard format, and a loan is generally classified as non-performing after 90 days overdue. Assets are further classified into substandard, doubtful and loss categories with prescribed minimum provisions. This means you can put 5 Indian banks side by side and compare bad loans directly.
Divergence disclosure. Where the RBI's assessment of a bank's bad loans exceeds the bank's own reported figure by more than a prescribed threshold, the bank must disclose the divergence. That single disclosure has preceded several of the sector's worst outcomes. It is one of the highest-value pages in an Indian bank's annual report.
Public sector and private sector banks are not comparable to each other. Ownership, cost structures, lending mandates, deposit franchises and capital-raising ability all differ. Compare within the group.
Non-banking finance companies are a very large part of the Indian market, and their funding structure is the risk. Read the borrowing profile, the share of short-term funding, and the maturity buckets disclosed in the annual report.
Indian IT services, which form a large share of the index, are judged on their own metrics — constant-currency revenue growth, utilisation, attrition, revenue per employee, deal wins and total contract value. Applying subscription software metrics to them is another version of the same error this article is about.
In the United States
The metric names differ, the ideas do not. The US equivalent of cost-to-income is the efficiency ratio, defined as non-interest expense divided by revenue. Lower is better, and the arithmetic is essentially the same .
Bad loans are reported differently. US banks report non-accrual loans, loans 90 days or more past due and still accruing, net charge-offs, and the allowance for credit losses. There is no single standard "gross NPA" and "net NPA" pair presented the way Indian banks present them. To compare a US bank with an Indian bank you must reconstruct the equivalent yourself, and the definitions will not match exactly.
Loan loss accounting changed. The current expected credit loss standard, known as CECL, requires banks to provide for expected losses over the life of a loan at the time it is made, rather than only when a loss becomes probable . That makes provisions more forward-looking and more sensitive to economic forecasts. It also means provisions can rise sharply on a change in outlook, with no loan having actually gone bad.
Capital is reported under Basel III with the common equity tier 1 ratio as the headline, alongside supplementary leverage ratios, and large banks publish the results of annual supervisory stress tests. Those stress test disclosures are genuinely useful and have no direct Indian equivalent.
Securities portfolios matter more. US banks classify securities as available-for-sale or held-to-maturity. Losses on held-to-maturity securities do not flow through reported capital for most banks. The unrealised loss figure is disclosed in the notes, and after 2023 it is a number no reader of a US bank should skip.
Where they differ, and what that tells you
Indian banks are easier to compare with each other than US banks are. The RBI's standard format for gross NPA, net NPA and provision coverage means a reader can rank 8 Indian banks on asset quality in 10 minutes. A US reader comparing 8 banks must reconcile non-accrual loans, past-due-and-accruing, charge-offs and allowance balances, which are related but not identical concepts. The practical instruction: in India, use the standard ratios and compare directly. In the United States, build a like-for-like measure yourself before comparing, and expect it to be approximate.
The main risk sits in a different place in each market. For Indian lenders, the most frequent source of severe loss has been credit — loans that were not repaid, sometimes recognised late. So the Indian reader's priority is asset quality, the divergence disclosure and the flow of new bad loans. For US banks, credit matters too, but the 2023 failures came from interest rate risk and deposit concentration. So the US reader must also read the securities portfolio and the proportion of uninsured deposits. Neither market's checklist is complete in the other.
Non-bank lending is structured differently. India's NBFC sector is large, regulated by the RBI, and funds itself substantially through bank borrowing and market instruments — which is why funding stress transmits quickly. The nearest US equivalents are more fragmented across specialty finance, mortgage originators and private credit funds, with less uniform disclosure. An Indian reader can pull a comparable capital adequacy and funding profile for most NBFCs. A US reader often cannot.
Segment separation is easier in the United States. Where a company combines a manufacturing business and a lending business, US segment reporting usually lets you separate them and value each on its own family's terms. Indian segment disclosure is thinner, so the reader more often has to work with a consolidated set of ratios that describes a hybrid. When that happens, weight the balance sheet analysis toward whichever segment carries the borrowing.
Carry this
- Name the family before you use a ratio. Lender, capital-heavy producer, or asset-light.
- For a lender, a bank's debt is its raw material. Use price to book against return on equity, then net interest margin, cost-to-income, gross NPA, net NPA, provision coverage and capital adequacy.
- Never use EV/EBITDA, debt-to-equity or free cash flow on a bank. Never use price to book on a software company.