A bank reports gross bad loans of 3.0% and net bad loans of 0.92%, with provision coverage of 70%. Those three numbers are arithmetically consistent and they can still describe a bank whose loan book is deteriorating rapidly.
Reading asset quality properly means reading a flow rather than a stock, and reading four disclosures together rather than the one that gets quoted.
The four numbers, defined
Gross non-performing assets. Loans on which the borrower has stopped paying, as a percentage of the total loan book, before deducting any provision. A bank with ₹1,260 crore of bad loans on a ₹42,000 crore book reports 3.0%. Below 3% is generally considered healthy for an Indian bank.
Net non-performing assets. The same bad loans after subtracting the provisions already made against them, as a percentage of net advances. With ₹882 crore of provisions, ₹378 crore remains uncovered, which is 0.92% of net advances. Below 1% is comfortable. This is what still threatens the capital, because the provided portion has already been charged to profit.
Provision coverage ratio. Provisions divided by gross bad loans: ₹882 crore over ₹1,260 crore is 70%. Above 70% is generally adequate.
Slippage ratio. Loans that turned bad during the period, as a percentage of the opening standard book. ₹1,050 crore of fresh slippages on an opening standard book of ₹40,000 crore gives 2.62%.
The first three describe a position. The fourth describes a movement, and it is the one that leads.
How a loan becomes non-performing in India
The classification follows a defined progression and knowing it explains why the reported numbers behave the way they do.
An account overdue by more than 90 days becomes non-performing. Before that, it moves through special-mention buckets: overdue by 1 to 30 days, by 31 to 60 days, and by 61 to 90 days. Those buckets are reported, and they are the earliest published warning available.
Once an account is non-performing it moves through sub-standard, doubtful and loss categories as time passes, with higher provisioning required at each step. That progression matters for profit: a bank that is not recovering or writing off its old bad loans faces a rising provisioning charge from accounts that turned bad years ago, purely because they have aged.
So the ageing of the bad-loan book matters as much as its size, and a bank with a stable gross figure can still see credit costs climb.
Why the stock lags and the flow leads
The gross figure is a balance. It rises with new slippages and falls with recoveries, upgrades and write-offs.
That means a bank can report an improving gross number while its underwriting deteriorates, simply by writing off faster. Write-offs remove the loan from the books entirely; the loss was already provided for, so profit is unaffected, and the ratio improves. Nothing about the borrower has changed.
The disclosure that defeats this is the movement schedule, which every Indian bank publishes: opening balance, additions during the period, reductions split into recoveries, upgrades and write-offs, and closing balance. Read the additions line and the write-off line separately. A bank with ₹1,050 crore of additions and ₹900 crore of write-offs has a flat gross figure and a book that is deteriorating.
Coverage falling while bad loans rise
This is the single most informative combination in bank analysis, and it takes two numbers to see.
If gross bad loans rise from 2.4% to 3.0% while provision coverage falls from 78% to 70%, the bank is recognising new problems faster than it is providing for them. The profit it reported in that period was supported by under-provisioning, and the charge has been deferred into the next few years rather than avoided.
The reverse combination, gross bad loans rising while coverage also rises, is a bank being conservative about a genuine problem. It hurts current profit and it is the better position to be in.
One refinement matters here. Coverage is quoted in two forms and the difference is large. Excluding technical write-offs, it measures provisions against loans still on the books. Including them, it counts accounts already fully written off, which inflates the figure without any additional provisioning. Indian banks disclose both. Read the excluding figure.
Credit cost: what the risk actually costs
Provisions as a percentage of the loan book. A bank charging ₹294 crore against a ₹42,000 crore book has a credit cost of 0.7%, which is normal for an Indian bank in an ordinary year, where 0.5% to 0.8% is the usual band.
This is the number that makes lending yields comparable. A lender earning 18% on unsecured personal loans with a 4% credit cost is earning a thinner risk-adjusted spread than one earning 9% on secured loans with a 0.4% cost, and the headline yield says the opposite.
Credit cost is also deeply cyclical and averages badly. A five-year average that excludes a downturn tells you what the book costs in good conditions, which is not the question. Look at the worst year in the series, and if there has not been one, treat the average as provisional.
The restructured book, which sits outside the headline
Loans whose terms were changed because the borrower could not meet the originals are disclosed separately from non-performing assets.
Restructuring is legitimate for a borrower with a temporary problem and a viable business. It is also the standard mechanism for postponing recognition of a loss, and the history of Indian banking includes long periods in which restructured books grew while reported bad loans stayed flat.
The arithmetic is straightforward. A bank with ₹1,260 crore of gross bad loans and ₹1,680 crore of restructured loans on a ₹42,000 crore book has ₹2,940 crore, or 7.0% of its book, in some form of difficulty. Quoting the 3.0% figure alone describes less than half of it.
Add the two before judging asset quality. Then ask what proportion of restructured accounts have historically slipped, which banks disclose during stress periods and rarely otherwise.
Unseasoned growth, and why fast lenders look good for three years
A loan book that grew 40% in a year has not had time to default.
Retail and unsecured loans typically show their losses in the second and third years after origination. So a lender growing quickly has a denominator inflated with young loans and a numerator that reflects an older, smaller book, and every asset quality ratio it reports is flattered by the arithmetic.
The correction is to look at asset quality on a lagged book: bad loans as a percentage of the loan book two years ago rather than today. For a lender growing 40% a year, the two versions differ by roughly half.
This is why the sentence "margin propped up by aggressive or unseasoned loan growth" appears on every experienced analyst's red-flag list. The margin is real, the growth is real, and the provisioning arrives after both have stopped.
The sector-level checks worth running
Divergence. Where the regulator's assessment of a bank's bad loans differs materially from the bank's own, the bank must disclose the divergence. A history of divergences is a history of a bank marking its own homework generously.
Sectoral concentration. The disclosure of exposure by industry shows where the risk sits. A bank with a large book in one stressed sector has a correlated problem rather than a diversified one, and the aggregate ratio hides it.
Large borrower concentration. The top twenty borrowers as a share of the book. Corporate lending fails in lumps, and one large account can move the entire ratio.
Security cover. Secured and unsecured exposure, disclosed separately. Loss given default is very different between the two, so the same gross ratio implies different eventual losses.
Putting the four together
The reading that works takes eight quarters of data and asks four questions in order.
Is the flow rising? Slippages first, and the 61-to-90-day bucket before that.
Is the stock being managed or improved? Additions against write-offs in the movement schedule.
Is the provisioning keeping pace? Coverage, excluding technical write-offs, against the direction of the gross figure.
And what is outside the headline? Restructured loans, and the share of the book written in the last two years.
A bank that passes all four is genuinely improving. A bank that reports a falling gross figure and fails the other three is reporting an outcome rather than describing one.
Why this matters more than the valuation
For a bank, asset quality is not one input among several. It sits upstream of everything.
Under-provisioning overstates profit, which overstates return on equity, which overstates book value, which makes the price-to-book multiple look cheaper than it is. All four errors point the same way, and they compound.
So the sequence is not negotiable. Establish that the book is what the bank says it is, then value it. Doing those in the other order produces a precise valuation of a number that was wrong.
Where each disclosure lives
All of it is in the quarterly results and the annual report, in predictable places.
The asset quality table in the results release gives gross and net bad loans in rupees and as percentages, along with provision coverage. Most banks also give the movement schedule here: opening balance, additions, reductions split into recoveries, upgrades and write-offs, and closing balance.
The special-mention buckets appear in the investor presentation for most banks and in the annual report's risk management section for the rest. Coverage varies, and a bank that stops disclosing them has told you something.
The restructured book is disclosed separately, with the amounts and, during stress periods, the proportion that has subsequently slipped.
The sectoral exposure table in the annual report shows the book by industry, which is where concentration becomes visible.
The divergence disclosure, where the regulator's assessment of bad loans differs materially from the bank's own, appears in the annual report and is one of the highest-information paragraphs a bank publishes.
Five places. About fifteen minutes once you know where they are.
Reading a full cycle rather than a quarter
Asset quality analysis done on one quarter is close to useless, because the numbers are seasonal, lumpy and managed.
Build an eight-quarter table with six columns: gross bad loans in rupees, slippages, recoveries and upgrades, write-offs, provision coverage excluding technical write-offs, and credit cost annualised. Then read across rather than down.
What you are looking for is divergence between the columns. A gross figure improving while slippages rise means write-offs are doing the work. Coverage falling while gross rises means provisioning is lagging recognition. Credit cost falling while the book grows quickly means the new lending has not aged.
Any one of those is a question. Two of them together is a finding, and it usually appears two to four quarters before the market reprices the stock.
What good looks like
A bank with a genuinely improving book shows a specific pattern.
Slippages falling before gross bad loans fall. Recoveries and upgrades contributing more of the reduction than write-offs. Coverage rising while gross falls. Credit cost coming down on a book that has been growing at a moderate rate for several years, so the loans have had time to season. And the special-mention buckets thinning ahead of all of it.
That combination is rare and it is what a turnaround actually looks like from the outside.
The opposite pattern is more common and easier to spot once you know the shape: a headline that improves while everything underneath it deteriorates.
Why this sits upstream of everything else
For most companies, accounting quality is one input among several and a mistake in it produces an estimate that is somewhat wrong. For a bank it is the whole analysis, because provisioning is simultaneously the largest discretionary item in the income statement and the thing that determines whether the balance sheet means anything. A bank that under-provides reports higher profit, higher return on equity, higher book value and a lower price-to-book multiple than the truth would support, and all four of those errors point in the same direction and reinforce each other. There is no other business where a single judgement moves that many numbers at once. Which is why experienced financial-sector analysts spend most of their time on four disclosures that a generalist would treat as footnotes, and why the ones who do not tend to be surprised in the same way at roughly the same point in every credit cycle.
Slippages lead. Gross bad loans lag.
Coverage tells you whether the cushion is being maintained.
Restructured loans sit outside all three.