GNPA, NNPA, provision coverage and credit cost: the loans that don't come back

The cost that arrives late

A manufacturer knows its cost of goods when it ships. A bank finds out its real cost of goods over the next several years, as it learns which loans won’t be repaid.

This is the single most important thing about reading a bank, and the reason the first post said a bank’s profit is partly a decision. Interest income arrives now. Losses arrive later, in lumps, and the bank has to estimate them in advance. The estimate is called a provision, and the loans it is estimating about are non-performing assets (NPAs).

From a missed EMI (equated monthly instalment) to a loss

When a borrower misses an EMI — the fixed monthly loan repayment — the clock starts. In India the classification is rule-based (the Reserve Bank of India’s, or RBI’s, income recognition and asset classification norms), which is what makes it comparable across banks:

Stage	Definition (simplified)	What the bank does
Standard	Paying on time	Small general provision
SMA-0 / 1 / 2	Special Mention Account: 1–30, 31–60, 61–90 days overdue	Watch list; for exposures of ₹5 crore and above, reported to the RBI’s CRILC (Central Repository of Information on Large Credits)
Sub-standard	Over 90 days overdue — the loan is now an NPA	Stop recognising interest income; provide a percentage of the balance
Doubtful	NPA for more than 12 months	Higher provision, rising with age and depending on security
Loss	Identified as unrecoverable	Provide 100%
Written off	Removed from the books	Any later recovery is income

Two things happen at the 90-day line. The bank stops counting interest on that loan as income (so NII falls), and it must set money aside. Both hit the P&L, which is why the moment of classification matters and why banks watch the SMA-2 bucket — the loans one month from tipping over.

GNPA vs NNPA: gross, net, and the number in between

Gross NPA (GNPA)  =  Total loans classified as non-performing
Net NPA (NNPA)    =  Gross NPA  −  Provisions held against those NPAs

GNPA %  =  Gross NPA  /  Gross advances
NNPA %  =  Net NPA    /  Net advances

Provision coverage ratio (PCR)  =  Provisions held  /  Gross NPA

Gross NPA is what has gone wrong. Net NPA is what has gone wrong that the bank has not yet set money aside for — the part still exposed to shareholders. Provision coverage is the bridge between them.

Item	Gross NPA (GNPA)	Net NPA (NNPA)
What it counts	Every loan classified as non-performing	The part of those loans not yet covered by provisions
The % is taken on	Gross advances	Net advances (after provisions)
What moves it	Loans turning bad, recoveries, upgrades, write-offs	All of those, plus how much the bank chooses to provide
What it answers	How much of the loan book has gone wrong?	How much of that is still exposed to shareholders?

Two banks with the same GNPA can report very different NNPA, because provisioning is partly a choice (more on that below). Read the two together, with the coverage ratio, rather than either one alone.

Worked example: HDFC Bank, FY25

From HDFC Bank’s results for the year ended 31 March 2025 (standalone, audited). ₹ crore, historical, for illustration only.

Item	31 Mar 2024	31 Mar 2025
Gross NPA	31,173.32	35,222.64
GNPA % of gross advances (reported)	1.24%	1.33%
Net NPA	8,091.74	11,320.43
NNPA % of net advances (reported)	0.33%	0.43%
Provisions held against NPAs (GNPA − NNPA)	23,081.58	23,902.21
Provision coverage ratio	74.0%	67.9%

So on 31 March 2025, of every ₹100 of loans about ₹1.33 was non-performing; the bank had set aside about ₹0.9 of that, leaving ₹0.43 still exposed. (The filing also notes GNPA excluding agricultural loans was 1.13% — farm lending has a seasonal NPA pattern that banks often show separately.)

Notice that coverage computed this way is specific provisions against NPAs. Banks also hold general and floating provisions not tied to any particular loan; some report a “PCR including write-offs” or including those buffers, which comes out higher. Same warning as with NIM: check which definition a number uses before comparing it.

Credit cost: the P&L view

GNPA and NNPA are balance-sheet snapshots — the stock of bad loans. Credit cost is the flow: how much the year’s provisions ate, relative to the loan book.

Credit cost  =  Provisions for bad loans in the year  /  Average advances

Item	₹ crore
Provision for non-performing assets, FY25	12,715.31
Average advances (FY24–FY25)	25,52,235.06
Credit cost	0.5%

Set this against the spread of about 3.41 percentage points and the arithmetic of lending becomes visible: the bank earns roughly three-and-a-half rupees per hundred lent, and loses about half a rupee of it to loans that fail. In a bad year for credit, credit cost can run several times higher — and at a few percent it consumes most of the spread. That is what a “credit cycle” means in rupees.

Why net NPA is partly a choice

Look again at the total provisions line from the first post:

Item	FY24	FY25
Provisions and contingencies (total)	23,492.14	11,649.42
of which floating provision	10,900	Nil

In FY24 the bank chose to make a ₹10,900 crore floating provision — money set aside against no specific loan, as a buffer for the future. RBI rules allow it; nothing forced it. In FY25 it made none. Strip that choice out and the two years’ provisioning look far more similar than the headline halving suggests.

This is the honest complication with every NPA number. Gross NPA is largely rule-driven — 90 days is 90 days. But how much to provide against a doubtful loan (within RBI minimums), whether to write a loan off (which removes it from GNPA entirely), whether to build floating buffers, how aggressively to restructure — all of these are management decisions that change net NPA, coverage and reported profit without changing a single borrower’s behaviour.

So two banks with identical loan books can report different net NPAs and different profits. The conservative one shows lower profit now and a bigger cushion later. Which is “better” depends on what happens next, which nobody knows. What you can do is read the provisioning policy in the notes and check whether coverage is rising or falling over time.

🧒 Explain it like I'm 10 (optional — skip if this is already clear)

You lend ₹100 each to ten friends. Two of them have stopped replying to your messages.

Gross NPA is the ₹200 those two owe you. You reckon you’ll probably get ₹130 back eventually, so you tell yourself you’ve “lost” ₹70 already — that ₹70 is your provision. Net NPA is the ₹130 you’re still hoping for.

Now, your friend Priya also lent ₹100 to the same ten people. She decides she’s lost ₹150, not ₹70. Same two non-payers, same ₹200 — but Priya reports a smaller net NPA and a worse year. Who’s right? Nobody knows yet. That’s why you look at both the gross number and how much each lender chose to set aside.

Reading the trend, not the level

A GNPA of 1.3% means little on its own. What tells you something:

• Direction of GNPA and the SMA buckets. Rising early-overdue balances precede rising NPAs by a few quarters.
• Slippages vs. recoveries. Filings usually disclose fresh slippages (loans newly turning NPA) and recoveries/upgrades. Gross NPA can fall because loans were written off, not because borrowers paid.
• Coverage trend. Coverage falling while GNPA rises means the buffer is being run down.
• Mix. Unsecured retail and microfinance carry higher NPA rates than home loans. A bank shifting mix toward higher-yield lending should be expected to show higher credit cost — that’s the trade, not a surprise.

Common mistakes

• Comparing GNPA across banks with different loan mixes. A home-loan lender and a credit-card lender at the same GNPA are in very different positions relative to their own norms.
• Reading a falling GNPA as recovery. Write-offs lower GNPA too. Check slippages and recoveries in the notes.
• Treating net NPA as “the real number.” It depends on provisioning policy. Gross NPA plus coverage tells you more than net NPA alone.
• Ignoring the timing. NPAs are recognised at 90 days; the loan may have been in trouble for a year. Provisions are the estimate of a loss that has already happened economically. The P&L catches up late by construction.
• Forgetting the interest effect. An NPA doesn’t just need a provision — it stops earning. A rising NPA book squeezes NIM and raises provisions at the same time.

Takeaway: Gross NPA is the loans that have stopped paying; net NPA is the part the bank hasn’t yet set money aside for; the provision coverage ratio is the bridge between them and credit cost is the year’s bill for all of it. HDFC Bank’s ₹10,900 crore floating provision, made in FY24 and not in FY25, is the cleanest illustration you’ll find that a bank’s reported profit is partly a decision.