The private-bank margin advantage is half pricing, half funding

The answer

Private banks out-earn public banks by 114 basis points of net interest margin (FY2026: 3.52% versus 2.37%), and the gap has been persistent, running 116bps as far back as FY2023. Decomposed, that advantage splits almost evenly between what private banks earn and what they pay.

Three supporting arguments

1. The gap decomposes cleanly into pricing and funding. In FY2026, Neither side dominates. Private banks earned 7.38% on assets against 6.79% for public banks: a 59bps asset-side advantage, while paying 3.86% for funding against 4.42%: a 56bps liability-side advantage. The two halves are within 3% of each other, so a public bank cannot close the gap by repricing loans alone; it has to win the deposit too. The two components sum to 115bps against a measured NIM gap of 114bps, so the decomposition is complete.

2. The gap is stable across a rate cycle, which makes it structural. Across FY2023 to FY2026 the cohort gap moved from 116bps to 114bps. That is the weaker half of the claim; the stronger half is that it held while the price of money did not. Over the same window India’s call money rate ranged from 4.25% to 6.75% (250bps of travel, 4.25% at the start against 5.50% at the end). A 250bps swing in funding costs moved the cohort gap by only 2bps, so the advantage is not a rate artefact. It is a franchise.

3. The likely mechanism is deposit mix, and this report does not measure it. Low-cost current and savings balances are won through primary-relationship behaviour: salary credit, bill payment, and everyday transactions, which is precisely what UPI intermediates. That is the standard explanation and it is stated here as a hypothesis, not a finding: no CASA or deposit-composition series is computed anywhere in this pipeline. What is measured is the consequence, the 56bps funding half of the gap above. Rule 1 applies to causal claims as much as to figures, so the joint between the payments sections and this one is marked as untested rather than asserted.

So what

• For a bank client: closing a 114bps margin gap needs both levers. The 56bps funding half is won through primary-account status, which is a payments and behaviour problem, not a treasury one.
• For an investor: treat the cohort gap as a franchise moat with a measurable width. Convergence would show up first in deposit mix, which this report does not measure, so treat that as the next thing to check rather than as a finding here.
• The link to payments: whoever owns the transaction owns the relationship that produces the cheap deposit. That is the strategic reason banks tolerate zero MDR.

Two further readings

Inclusion: the rails are deep, not just wide. Account ownership reached 89.0% of adults in 2024 (973.6mn banked adults). Against UPI volume that same year, each banked adult now runs 14.9 transactions a month, up from 4.0 in 2021. Access stopped being the constraint some years ago; usage intensity is the story now, and it is what makes the zero-MDR cost base grow. Ownership among women reached 89.2% in 2024, from 26.5% in 2011: on this measure the gender gap has closed, which means the remaining inclusion question is about use rather than about access for either group.

The market’s verdict. Over 2021-11-08 to 2026-11-05, the median private bank returned 11% on price against 197% for the median public bank. The market has NOT paid for the margin franchise: public banks outperformed despite the thinner spread, which says the gap was already in the price. Price return only, dividends excluded, so this understates total return for the higher- yielding public cohort.

What the margin series cannot explain. A cohort earning 114bps less on every rupee of assets should not re-rate harder, and the NIM series is silent on why it did. Asset quality is not: India’s non-performing loans fell from 9.98% of gross loans in 2017 to 2.06% in 2025, a 7.9 point repair concentrated in the cohort that was carrying the bad book. The public-bank re-rating reads as balance-sheet repair rather than margin expansion, and those are different things to underwrite: repair is finite and largely spent, margin is not.

Method and its limits

NIM here is a proxy: net interest income divided by average total assets, computed from filed annual income statements and balance sheets (10 NSE-listed banks). Reported NIM uses average earning assets, a smaller denominator, so these levels read low by roughly 20-40bps. The comparison is sound because the bias applies equally to both cohorts; the levels should not be quoted against a bank’s disclosed NIM.