Data to 20 November 2026
About

Methods

Sector and state directories

One page per sector and per state, bringing together what Tipsheet already publishes about it: a sector’s stock returns beside the bank credit it gets, the foreign money in it, what mutual funds hold of it and how its companies are doing; a state’s investors beside its fund assets, prices, output and borrowing.

Code: pipeline/tipsheet/compute/directories.py, publish/directories.py, step directories (steps/site_directories.py). Crosswalks: pipeline/tipsheet/reference/crosswalk/. Tests: pipeline/tests/test_directories.py.

What a directory page shows, and what it doesn’t

Every number on a directory page is read from a bundle the site already publishes, exactly as that bundle has it, with its own date. The page links each number to the topic it comes from. Nothing is computed from the join: no correlations across sources, no “leads”, no rankings of sectors or states that would imply a cause. A new claim of that kind needs pre-registration in docs/research/site_calculations_spec.md first.

Two things are added on top of the source numbers, and both are labelled:

The canonical sectors

Twenty-one sectors (sectors.csv). The grain is NSE’s 22-sector level of its four-level industry classification (12 macro-economic sectors, 22 sectors, 59 industries, 197 basic industries; NSE Indices, Industry Classification Structure, March 2022). It’s the level where the market sources meet: NSDL’s FPI tables have used it since 30 April 2022, and the foreign-ownership estimate uses it. Four changes:

The states

All 28 states and 8 union territories (states.csv), with ISO 3166-2:IN codes as revised on 23 November 2023 (Chhattisgarh IN-CG, Odisha IN-OD, Telangana IN-TS, Uttarakhand IN-UK; the old codes are kept in iso_previous). History is explicit:

state_map.csv holds every label a source prints that isn’t a state’s own name: old spellings (Orissa, Pondicherry, Uttaranchal), NSE’s “Keralam”, ”&” forms, a source’s misspelling (“Lakhswadeep”), the combined regions sources use to keep long series comparable, and the labels that are not places (Others, Army personnel, a generic “Union Territory”, SEBI’s regions, the national total).

How matching works

sector_map.csv lists every category of every source that classifies by sector, keyed by the classification scheme it uses (nse_index, nse_sector, nse_macro, fpi_group, amfi_industry, rbi_bank_credit, rbi_nbfc_credit, rbi_nbfc_sectors, rbi_companies, iip_industry, wpi_group, isp_subsector, gva_sector, pulse_indicator, capital_ownership). Each row gives the canonical sector, a match quality and a note on what differs. A category may map to two sectors (Financial Services to Banks and to Non-bank financials), each partial. Labels are compared ignoring case, ”&” and punctuation.

What the match qualities mean to a reader:

Household loans for a sector’s products (home loans, vehicle loans, consumer-durable loans) are mapped partial, with a note that they measure demand for the sector rather than lending to it. Priority-sector categories map to none: they are a regulatory slice of loans already counted in the main categories.

Known mismatches

Checks

The step, and tests/test_directories.py against published bundles where they exist, fail when:

On failure nothing is published and the last published directories stand. The fix is a row in reference/crosswalk/. The CSVs are edited by hand and reviewed like code.

Bundles

Slugs: lowercase ASCII with hyphens; states by full name (tamil-nadu), agreed with the taxonomy (ia).

This note is the repository file docs/methods/directories.md, rendered as-is.