Credit and deposits

Source: Reserve Bank of India. Separate datasets: Scheduled Banks’ Statement of Position (HTML); Sectoral Deployment of Bank Credit (Excel statements 1 and 2); Sectoral Deployment of Credit by NBFC (Excel); and one-time RBI DBIE historical bank-sector and industry exports (Old Format and New Format).

Matched credit and deposit totals

The default universe is scheduled commercial banks, including RRBs, SFBs and payments banks. All scheduled banks, including co-operative banks, are a separate comparison. Deposits exclude deposits from banks; credit excludes inter-bank advances. Each record retains the current, previous and corresponding prior-year reporting dates printed in one release. Demand plus time deposits must reconcile to deposits. Annual growth is current / printed prior-year stock minus one.

Credit/deposit ratio = credit / deposits. Growth gap = credit annual growth minus deposit annual growth, in percentage points. Annual additions = current stock minus corresponding prior-year stock. Incremental credit/deposit ratio divides these additions and is absent when deposit additions are not positive. Stock changes are not disbursements. Demand deposits are not CASA.

Sector allocation and contributions

Amounts are INR crore. Source-reported annual growth is reconciled with the printed amounts and comparison dates. Full bank credit, food credit and non-food credit cover all SCBs. Detailed bank sectors cover selected SIBC banks, with separate year-ago dates. The four broad-sector denominator is their own combined stock. Category share = stock / labelled parent stock. Contribution to parent growth = (current category stock − corresponding prior-year category stock) / prior-year parent stock × 100. Industry top-level categories use the industry total; nested categories use their immediate parent. Priority-sector categories and “of which” rows overlap; do not sum every row.

NBFC/HFC lending has its own selected-lender universe and five broad categories, including other loans. It is separate from bank lending to NBFCs. Adding these lender totals double counts funding links and does not measure unique borrower financing. Source coverage is provisional and benchmark-relative.

History and comparability

The newest release containing an annual comparison wins per date, statement and lender universe. A later financial-year-anchor column that contains only a stock cannot erase an existing annual comparison. The entire statement within a lender universe uses one release; individual category gaps are not filled from older releases. Historical values can be revised. This is descriptive latest-vintage analysis, not an as-of backtest. Reporting-fortnight dates changed from December 2025. Retain the actual earlier dates, rather than imposing an exact calendar-year match. Values are not adjusted for mergers, bank sample changes or classification changes. No inference of missing observations or cross-universe reconciliation is made. Initial archive backfills are bounded and coverage is visible in each chart’s table and download.

Raw artifacts are immutable and catalogued in the Data bank. Schema, key, date, amount and independent SQL checks precede warehouse publication. Workbook access can be intermittent; HTML challenge responses fail raw file validation.

The workbook parser supports both the five-endpoint monthly layout and the three-endpoint March layout. March comparisons use adjacent annual endpoints; monthly comparisons use the two annual pairs around the financial-year columns. Both source percentage columns are checked against the underlying stocks. A repeated fetch of one workbook selects its latest validated artifact; a duplicate fact inside a workbook fails validation. Raw category labels remain in the warehouse; display labels remove source footnote markers.

Chart histories preserve missing months as null markers, with no interpolation or line bridging. In particular, year-ago stocks printed in three recent NBFC releases do not establish a continuous older monthly history. Chart null markers are not new warehouse observations and do not change latest-value tables.

Downloaded historical exports

The supplied major-sector and industry exports add 20,008 source facts. Stocks begin in April 2007 and the new-format exports end in June 2026; newer releases extend supported categories beyond that endpoint. Annual growth starts in April 2008 where both observations exist. Coverage differs by category.

The raw downloads are content-addressed and catalogued as local imports, with original filenames, checksums and import dates. An import date is not a release date. Source footnotes are retained in the warehouse. Originals are unchanged.

Calculate historical annual growth using the last actual endpoint in the same calendar month one year earlier, within the same source format and category. Never use a twelve-row lag, since missing months would change its meaning. Duplicate months keep their last dated observation. In recent overlapping releases, choose the last dated annual-comparison endpoint within each month and universe, then the newest complete vintage of that endpoint. A newer release of an earlier fortnight does not move the historical month backwards. Source dates remain actual reporting dates, including the December 2025 transition to month-end reporting.

Old and new formats are kept separate when forming annual pairs. Overlapping old-format observations remain usable after January 2019 when they contain a valid annual comparison that the revised-format sheet lacks. Thus the absence of revised-format 2018 stocks does not itself justify blanking all 2019 growth. Native release workbooks also print paired comparisons. Each selected record keeps its current amount, prior amount, source date and growth together; an old-format rate is never attached to a revised-format stock.

Categories are matched by label and hierarchy, normalising spacing, punctuation, ampersands, footnotes and reviewed naming variants (Adv./Advances, Dep./Deposits, and securities-advance descriptions). Numeric codes alone are unsafe: old 4.8 is Other Personal Loans, while new 4.8 is gold loans; services categories and infrastructure subcategories were also renumbered. The changed wholesale-trade definition and railway qualifier are not assumed equivalent. Unmatched categories remain in the source warehouse but are not attached to the current category by guessing. Chart bands mark the reporting revision period. Rate comparisons across that period remain sensitive to definitions and sample changes.

In overlapping recent months, a complete release with printed annual comparisons outranks the export. An amount-only release endpoint cannot overwrite a valid historical annual comparison. The chosen monthly statement within each lender universe is kept intact. All-SCB Section-42 endpoints and selected-bank SIBC endpoints can differ within a month; one cannot displace the other simply because its release was newer. Chart JSON includes comparison dates, source checksum, history segment and the comparison basis separately from numeric plot columns. Null plot markers have no source-observation provenance.

Reporting changes that affect interpretation

• August 2018 consumer-durables credit includes a ₹17,300 crore correction of housing loans that a bank had misclassified. It is not solely a borrowing decline.
• From May 2024, a bank reclassified a category of agricultural loans into retail gold loans. Published gold-loan growth includes that transfer.
• July 2023 onward includes the bank/non-bank merger in reported stocks. These calculations do not manufacture merger-adjusted histories.
• Priority-sector definitions and bank-sample coverage changed. Old-format priority data are explicitly based on the older definition; matched labels do not imply a constant regulatory definition or sample.

These changes are disclosed rather than repaired through interpolation or guessed adjustments. The first two are also shown beside the relevant household charts.

The two files do not supply deposit stocks, fortnightly matched credit/deposit comparisons, or NBFCs’ own lending histories. Those need separate exports.

September 2025 omission

A monthly selection bug let the all-SCB annual endpoint dated 5 September 2025, printed in a newer release, displace the detailed selected-bank rows dated 19 September 2025. These are distinct reporting universes. Vintage and month selection now partition by lender universe as well as statement; each retains its own actual date. A regression test covers this exact failure class.

Source-native 2019 comparison recovery

The initial portal cutoff discarded valid overlapping old-format 2019 sector comparisons. That cutoff has been removed. Old-format current/prior pairs and native release pairs restore the household and services histories without any cross-format division. For industry, eight accessible 2019 releases plus four later comparison releases (October 2020–January 2021) provide printed 2019 annual pairs. Four original late-2019 workbooks returned HTML error payloads and were quarantined, including retry attempts; they are not used as Excel data.

The legacy parser supports separate category-code and label columns, source amounts in INR billion (multiply by 100 to obtain crore), month-first dates, date footnote markers, and plain-date spacing variants. Printed annual rates must reconcile with the paired amounts. Each target statement must contain actual parsed facts; merely finding its title is insufficient. Legacy sampled headline rows explicitly labelled Non-food Credit (1 to 4) are not relabelled as all-SCB totals. Their four-sector sum is checked with source rounding tolerance, while the detailed selected-bank categories are retained. Source codes remain available in the chart JSON even when display codes have been harmonised.

Any category without a valid same-source annual pair stays blank. Chart JSON lists unavailable 2019 growth months by category; the explanatory note is no longer a blanket statement that every 2019 rate must be missing.