Who India's banks lend to, and how safely.

When the blue line runs above the gold one, banks are lending faster than they are taking in deposits and must find the money elsewhere: by selling government bonds, borrowing, or competing harder for deposits with higher rates. When it runs below, money piles up in banks that cannot find borrowers.

Credit growth swings with the economy and with the banks' own health. It peaked in the 2004 to 2008 boom, fell after the 2008 crisis, and sank to its lowest after 2016, when banks laden with bad loans stopped lending to industry. The merger of HDFC Ltd into HDFC Bank in July 2023 added a year of artificial growth; the pale line shows growth without it.

Every deposit goes one of three ways: lent out, put into government bonds, or kept as cash and balances with the RBI. Banks must hold a minimum share in government bonds (the statutory liquidity ratio) and in cash with the RBI (the cash reserve ratio), so the credit-deposit ratio can never reach 100% for the system as a whole. A rising ratio means banks are turning deposits into loans rather than bonds.

This is how deep banking runs in the economy. The climb after bank nationalisation in 1969 and the branch expansion that followed brought deposits from villages into the system; credit caught up after the 1990s reforms and the 2000s boom. India's ratio is still low next to China's or most rich countries', which is why economists argue both that Indian firms are short of credit and, at times, that households are borrowing too fast.

Each bar is one kind of borrower's share of the money lent to the four big sectors. Personal loans are loans to individuals for their own use: homes, vehicles, consumer goods, education, credit cards, gold loans. Loans to a proprietor for her shop count under services or industry, not here. NBFCs are finance companies that borrow from banks to lend on, so bank credit to them ends up with households and firms one step removed.

The lines are shares, so one rising means another falling. Industry's share fell through the 2010s as big projects soured and banks pulled back; personal loans, which rarely go bad in large lumps, took its place. The two classifications are drawn apart because some loans moved between sectors when the RBI changed the codes in 2019; where they overlap, read each against its own history.

Growth from a small base can look dramatic: a kind of loan that doubles from a low level moves the total less than a big one growing slowly. Read the bars with the shares above. Since May 2024 one bank has reclassified a category of farm loans as gold loans, which inflates gold-loan growth and depresses agriculture's for a year.

Home and vehicle loans are secured: the bank can take the house or car. Credit cards and "other personal loans" are unsecured, and are the ones regulators watch for strain. In November 2023 the RBI raised the capital banks must set aside against unsecured personal loans and cards, and their growth slowed afterwards. Gold loans are secured by jewellery and are often taken by families who cannot borrow any other way; a surge in them can mean either a high gold price making jewellery worth more, or households under pressure, and the data cannot say which.

These are the rates borrowers are actually paying on loans already on the books, not the rates banks advertise for new ones. Old loans keep old rates until they are repriced, so this average moves slowly after the RBI changes the repo rate (see the interest rates guide). Credit-card rates are the interest charged on balances carried over, which most cardholders who pay in full never see.

A circle far to the right of its bar means many borrowers holding little money: the crop loans, microloans and small personal loans most Indians have. A bar far to the right of its circle means a few borrowers holding a lot: the companies and projects whose fortunes decide whether banks stay healthy. The bad-loan crisis of 2015 to 2020 lived in the bottom rows.

Big companies can raise money from bond markets and abroad; small ones depend on banks. So large-industry credit can be weak even when big firms are investing, while small-firm credit tracks the health of the many workshops and units that employ most industrial workers. Part of the jump in medium-firm credit after 2020 is definitional: when the government raised the MSME limits, firms that had counted as large were reclassified as medium.

Infrastructure covers power, roads, telecom, ports, airports and railways other than Indian Railways. These are long, large loans, and many of the bad loans of 2015 to 2020 began here, especially in power. A big share is not by itself a risk; what matters is whether the projects earn enough to repay.

NBFCs are a second channel of credit: they borrow from banks, bond markets and the public, and lend to people and businesses banks find too small, too risky or too remote. Bank credit to NBFCs (the services row above) and NBFCs' own lending (this plate) overlap: part of the same money is counted twice, once as a bank loan to the NBFC and once as the NBFC's loan to you. Never add them.

When the government buys wheat after the rabi harvest (April to June) and rice after the kharif harvest (October onwards) at the minimum support price, a consortium of banks lends the Food Corporation of India the money. The loan is repaid as the grain is sold through ration shops and the government reimburses the FCI. So food credit spikes in procurement season and falls back, and its level over years tracks how much the government relies on bank loans rather than the budget to carry its grain stocks. It is a tiny share of bank credit, which is why the RBI reports non-food credit separately: food credit's swings would otherwise blur the signal.

The old rule of thumb was that money growing faster than the economy's output pushes up prices. In India the rule has held loosely over decades but poorly over months: food prices, oil and the rains move inflation far more than money does, and much new money is held as savings rather than spent. Since 2016 the RBI has targeted inflation directly through interest rates, and treats money growth as one signal among many. The two M3 lines are different rulers: the pale one excludes HDFC Ltd's deposits that entered the banking system in July 2023, and stops in June 2025.

On 8 November 2016 the government withdrew ₹500 and ₹1,000 notes, most of the cash in circulation. People deposited the old notes in banks, so cash fell and deposits rose, and the share of money held as cash dropped. As new notes were printed the share climbed back. A second climb came in 2020, when people held more cash during the pandemic. The rise of UPI payments since then has not, so far, pushed cash's share much below where it was before 2016.

Velocity falls as people keep more of their wealth in bank deposits instead of land, gold or cash under the mattress: money stops being only a means of payment and becomes a store of savings. That is financial deepening, not a slowdown. The multiplier rises as banks lend out more of each rupee the RBI creates; it fell back when the RBI raised the cash banks must keep with it, and after demonetisation.

A ratio below 100% means the state's depositors are funding loans somewhere else. Some of that is real: savings from Bihar or Uttar Pradesh pay for loans in Maharashtra or Tamil Nadu. Some is bookkeeping: a company's loan is booked where its head office banks, not where its factory stands. The circles correct for that by counting credit where it is used. A big gap between bar and circle marks a state that books loans for the rest of the country (Maharashtra, Delhi) or one whose projects are financed from elsewhere.

Credit is counted where the branch is. Big-city branches book loans for companies operating everywhere, so the metropolitan share overstates how much of the money is used in cities. Even so, a rural depositor's rupee is more likely to be lent in a city than a city depositor's is to be lent in a village. RRBs, which serve villages, are left out of this quarterly table from March 2023; they would raise the rural share a little.

The table counts loans by the borrower named on the account. Joint home loans, loans to women-led self-help groups and loans to firms women own are not all captured here, and a loan in a woman's name is not always hers to decide on. The southern states' high shares owe much to self-help group lending and to gold loans, often taken against women's jewellery.

A loan becomes a non-performing asset (NPA) when interest or principal is unpaid for 90 days. In the boom of the late 2000s banks, public sector banks above all, lent heavily to power, steel, roads and telecom projects. Many stalled. For years the losses were hidden by restructuring loans on easy terms. In 2015 the RBI's asset quality review made banks recognise them, and the NPA ratio jumped. The Insolvency and Bankruptcy Code of 2016, write-offs and fresh capital from the government brought it down. Every rupee of bad loans is a rupee of depositors' money the bank must cover from its profits or capital.

A write-off removes a bad loan from the books after the bank has set aside money to cover it. It does not forgive the borrower: the bank can still pursue recovery. But money recovered later from written-off loans is a small fraction of what was written off, so most of it is a loss, borne by the banks' shareholders, which for public sector banks means the taxpayer.

A fraud is counted in the year a bank reports it, which can be many years after it happened: the peaks of 2019-20 and 2020-21 are mostly loans from the boom years declared fraudulent once they had gone bad. So a peak is a year of discovery, not of crime. The rise in 2024-25, the RBI says, came mostly from old loan frauds that banks re-examined and reported afresh after a Supreme Court judgment of March 2023. Card and internet frauds are many and small; they are the ones ordinary customers meet.