Methodology and sources

This site shows India's economy for one financial year, 2023-24, as a single table of payments called a social accounting matrix (SAM). Every account in the economy appears twice: once as a row, where it receives money, and once as a column, where it pays money. So every cell answers one question: how much did the column pay the row? For every account, what comes in equals what goes out. That is what "balanced" means.

The SAM has 128 accounts: industries and the products they sell, wages and profits, households in ten groups, companies, banks and other financial firms, the Union government (the Centre), the States, each kind of tax, the rest of the world, saving and investment, and financial instruments. It is built on India's new national-accounts series, whose base year is 2022-23, and uses the latest published figures for 2023-24.

India's GDP in 2023-24 was ₹290.73 lakh crore Official. Every "share of GDP" on this site divides by that figure.

Why the flows add up to far more than GDP

GDP counts each final good or service once. The map shows gross flows: it also shows the inputs firms buy from each other, the money that passes through financial markets, and transfers such as taxes, pensions and remittances. A single rupee can appear in several flows as it moves: a household pays a shop, the shop pays its workers, the workers pay tax. Add up every cell and you get ₹2,532.21 lakh crore, several times GDP. Nothing is double counted; the money is simply followed at every step.

What this is not

This is not a forecast, and it is not a model of behaviour. It records what was paid in a year that has ended. It does not predict what will happen next, and it does not say what would happen if a policy changed. Where a story says "of every ₹100", it means the average share in 2023-24, not what an extra rupee would do. Working out ripple effects needs a model with its own assumptions, which comes in a later part of this project.

Step one: write down what the sources say. Nine building blocks (production, factor incomes, households, taxes, government, financial firms, the rest of the world, capital and finance, and the closing accounts) wrote 3,893 entries into 2,169 cells, using 84 kinds of flow. Each entry carries a source record and one of three statuses: official (a published figure used as it is), derived (arithmetic on published figures) or estimated (a key, a structure or a model applied to a published total).

Step two: write down what must hold. There are 429 constraints. Most are published totals that the map must reproduce, each with a tolerance: GDP, consumption, investment, exports and imports to within a crore or two; some taxes within a few per cent of cash collections, because the national accounts record taxes when they are due, not when the cash arrives. A few are floors, caps or judgement bands: ranges we chose where no source pins a figure.

Step three: balance. The first assembly did not balance. Some accounts received more than they paid out, mostly because the new national-accounts series does not yet publish who pays interest, dividends and transfers to whom, so older structures and survey keys had to fill those flows. Balancing finds the smallest set of proportional changes that makes every account balance while keeping every constraint (a cross-entropy method). Each entry has a role:

• fixed entries never move;
• anchored entries are official or derived figures that may move only inside a small box around the published number;
• free entries are estimates; the less confident the estimate, the more freely it may move.

The solver finished in 83 iterations, and its answer comes with a mathematical certificate that, by that measure, no smaller set of changes keeps every constraint. Every account's receipts now equal its payments.

How much did balancing move? In all, ₹28.38 lakh crore (the sum of every change, ignoring sign). That is 9.8% of GDP, and 1.1% of the absolute value of the whole SAM. Almost all of it fell on estimated, freely adjustable entries; no fixed entry moved.

Large moves are findings, not noise. Every move of at least ₹5,000 crore, or of at least ₹500 crore and 10.0% of the entry, needs a written explanation with a verdict. 223 entries and 103 cells needed one. For entries, 151 are explained, 64 are explained in direction but not in level, and 8 are unexplained: balancing moved them and no source says why. For cells the figures are 73, 26 and 4. The unexplained ones all involve public sector companies: the interest and dividends they pay to banks and NBFCs, and the interest and grants they receive. The map flags every one of them, and the flow panel shows each entry's value before and after balancing.

Not every limit was moved by data. 174 constraints end exactly on the edge of their tolerance. Where that tolerance was a published figure's rounding or a cash-versus-accrual allowance, the edge is evidence. Where it was a judgement band, the edge is our choice: see the limits.

Every flow on the map and every figure in the stories carries a badge that answers two separate questions, because a total can be firmly published even when its parts are estimates. On this page, the economic values (GDP, household factor income, the discrepancy items and the sums of flagged entries) carry the same badge; counts, rule thresholds and totals over the whole SAM carry none: they describe how the map was built, not a flow in the economy. A zero amount has no badge.

First: is the total held to a published figure? If it is, the badge says Official total (or Derived total). This happens when the figure's parts are exactly covered by published totals that balancing had to respect within a tight tolerance: at most ₹1.0 crore or 0.1% of the figure, whichever is larger. A matching annual series whose check against the map passes counts too. Select the badge to see which published figure holds it. Household consumption, household saving and remittances received are examples. If no published total holds a figure, the badge shows only how its parts were made.

Second: how were the parts made? Each entry gets one of four labels:

• Official: a published cell or total, used as it is. 1,127 entries, worth ₹1,261.66 lakh crore (summed ignoring sign).
• Derived: arithmetic on published figures, such as a difference or a sum of two tables. 114 entries, worth ₹223.94 lakh crore.
• Estimated: a published total split with a key, a structure or a model, for example consumption split across household groups with survey shares. 2,180 entries, worth ₹802.71 lakh crore.
• Set by balancing: balancing moved a free estimate enough to need a written explanation, or the entry has an explanation verdict, or it sits in a judgement band whose edge it reached. Its level comes from the balancing, not from a source. 472 entries, worth ₹389.81 lakh crore.

Balancing nudged many other entries by small amounts. Those keep their original label, and the flow panel says "adjusted by" the percentage, so a move of a fraction of a per cent does not hide an official figure behind a "balancing" label.

A flow on the map usually contains many entries. Its badge shows the least certain label that carries at least 10.0% of its value, and a small bar shows the full mix of the four labels. Across the whole SAM the mix is: official 47.1%, derived 8.4%, estimated 30.0% and set by balancing 14.6%. On the map, estimated ribbons are hatched and balancing-set ribbons cross-hatched, so colour is never the only signal.

Flags add warnings in plain words. On the map they read:

• "Level set by a judgement band, not by data." The entry belongs to one of the bands described under limits: the balanced value sits on the band's edge, so our choice of width sets it.
• "Balancing moved this and the reason is not yet documented." Its explanation verdict is "unexplained".
• "Direction explained, level not." A source supports the direction of the move but not its size.
• "Estimated with the old-base structure." The split uses the pattern of the previous national-accounts series, because the new series does not publish it yet.
• "Includes a discrepancy or reconciliation item." The flow panel says which published residual it is.
• "Includes negative amounts." Part of the flow is negative, such as a subsidy (a negative tax) or a sector that sold more of an asset than it bought, so it is drawn in the opposite direction.
• "Uses a stated assumption." Somewhere in its sources there is an assumption rather than a published figure; the method is one click away.

The words "official", "derived", "estimated" and "official total" appear on this site only where these rules put them.

India publishes reliable totals: GDP and its three approaches, value added, consumption, government consumption, investment, exports and imports, the production structure, tax devolution, each sector's saving, the balance of payments, and household financial saving. Inside those totals, many splits are not published. The map fills them, and labels them estimated. These are the main ones; each link opens an example on the map.

• Every split by household group: income, consumption, taxes, transfers and saving across the ten rural and urban groups. The totals are national; the shares come from household surveys and other studies. Examples: income earned by households, household consumption, household saving.
• Interest, dividends and other property income between institutions: who pays whom. Where the split follows the previous national-accounts series it is flagged "estimated with the old-base structure". Examples: dividends and distributed income from companies, interest and insurance income from financial firms, companies' payments to financial firms.
• The split of financial firms' saving and income among the RBI, banks, NBFCs, and insurers and pension funds. Examples: financial firms' saving, their operating surplus.
• The split between the Centre and the States of government consumption, saving and shared taxes. Examples: the Centre's spending on public services, the States' spending, the Centre's negative saving.
• Who holds which financial asset from whom, taken from RBI's flow of funds, and the residuals that reconcile it with the national accounts. Examples: financial assets acquired, new liabilities.

Some figures are set by balancing rather than estimated from a source, and the map labels them so: estimates that balancing moved by enough to need a written explanation, including those whose move is unexplained or explained only in direction, and flows that sit on the edge of a judgement band. For these the flow panel shows the value before balancing, the value after, and the change.

These are the main ways the map could mislead you if you did not know them.

The household survey sees only part of consumption. The household consumption survey (HCES) for 2023-24 adds up to 44.3% of the consumption in the national accounts. We spread each product's national consumption across household groups with the survey's shares, which assumes the missing part is spread like the part the survey sees. Studies suggest it is concentrated among richer households, so the richest groups' consumption is probably understated and their saving overstated.

Income by household group comes from an older study's pattern. Factor income of households, ₹192.46 lakh crore Derived total · split set by balancing, is held to totals derived from the published national accounts, but it is split across the ten groups with income gradients from an IFPRI social accounting matrix built on the old national-accounts base (D2-19). We did not use survey microdata (D1-2), so the pattern within rural and urban areas is a modelling choice. Income tax by household group has no open source at all and is estimated.

Non-profit institutions serving households (charities, religious bodies, many private schools) are inside the household account, as in India's national accounts.

Old-base structures fill gaps in the new series. The new national-accounts series does not yet publish who pays interest, dividends, transfers and some taxes to whom. Where nothing newer exists, the map uses the pattern of the previous series as a starting point and holds it to the new totals (D2-21). 36 entries, worth ₹19.15 lakh crore Set by balancing, are flagged this way. The old and new series do not line up evenly, so this is the weakest part of the map.

Some levels are set by tolerance bands we chose. Where no source pins a flow, we let balancing move it only within a band, for example within a twentieth or a tenth of a related published figure. In 11 such bands the balanced value ended exactly on the band's edge, so the width of the band, a judgement, sets the level. They cover 97 entries worth ₹30.12 lakh crore Set by balancing:

• general government's current transfers and its capital transfers, around the published spending by function;
• railway and postal pensions;
• money for central schemes released outside State treasuries;
• revenue grants to public sector companies;
• Union CGST, the GST compensation cess and Union excise, which sit on the ceiling of their allowance above cash collections;
• the Centre's share of government depreciation;
• public sector companies' corporation tax.

Every entry in these bands is flagged, and every one that balancing was free to move is labelled set by balancing. The owner accepted these widths provisionally at the second project checkpoint (Checkpoint B); a test of how the map changes if they are halved or doubled has not been run.

Discrepancies are shown, not hidden. Official statistics for 2023-24 leave some gaps open. The map carries them in a separate account instead of balancing them away, and balancing never moved them:

• Household residual, ₹5.71 lakh crore Derived total · split estimated: household disposable income less consumption less saving, each as published. The total is derived from those published figures; its split across household groups is estimated. It is never counted as consumption or saving.
• Services import reconciliation, ₹2.12 lakh crore Official total: the difference between imports of services in the national accounts and in the balance of payments (a CIF adjustment). It is an accounting item, not money paid abroad.
• Saving-investment errors and omissions, ₹3.59 lakh crore Estimated: investment less saving less net capital inflow. It is one residual, derived from the published national accounts; the map's total equals that derived figure (the export checks it), but no balancing constraint holds it, so its badge shows only how its two parts were made. Of it, ₹2.04 lakh crore Derived is attributed to valuables, such as gold and silver, that the national accounts assign to no institutional sector; it is derived from published figures and carried into the households' capital account. The remaining ₹1.55 lakh crore Estimated is spread over the capital accounts of companies and financial firms in proportion to their investment, so that split is estimated.
• GDP expenditure discrepancy, ₹0.0 crore in 2023-24.

One year in full, other years in part. The balanced map exists only for 2023-24; see the time series.

The stories use average shares, not multipliers. "Of every ₹100" means the average in 2023-24. It does not say what an extra rupee of spending, tax or remittance would do, and it follows only the first round of payments.

Some findings are still open. The States' subsidies come out well above what State budgets report; public sector companies keep more than the national accounts say they save; two official sources disagree on how saving is split between banks and NBFCs; and the rural richest fifth shows a saving rate below the rural poorest fifth, which looks implausible. These are listed under open issues.

The balanced SAM exists for one year only, 2023-24. No balanced map exists for any other year, and this site does not invent one: it does not interpolate between years or project the 2023-24 structure backwards.

For other years the time slider shows a lighter view: the top flows for which annual series exist. There are 69 such series (54 core flows and 15 memo items such as GDP), running from 1950-51 to 2025-26, though few series cover the early decades. This view is not complete and not balanced: it shows only the flows we have series for, a banner says so, and a flow with no figure for a year is listed as missing, not drawn.

Both views size each flow as a share of that year's GDP, so widths can be compared across the switch.

Breaks in the series are marked, never smoothed.

• The base year changed. From 2022-23 the national accounts use the new base; earlier years are on the 2011-12 base. Levels on the two bases are not directly comparable; shares of GDP, each on its own base, are closer. Some household financial saving series also change base earlier, at 2011-12 (7 such changes); those are marked too.
• Sources and concepts change. Long series are joined from several sources (34 joins in all), and a concept can change, as State VAT did when GST replaced most of it in 2017-18 (1 such break in all). Every join, base change and concept break appears as a marker on the slider and in the trend charts.
• Linked series are derived. For 10 flows, such as consumption, investment and GDP, old-base values are also shown ratio-linked to the new base at 2022-23. That linked series is derived, labelled so, and switched off by default. All other series show the break as it is.
• The latest year is provisional. The most recent figures are provisional or revised estimates and will change; the stage of each estimate is marked.

Each series is checked against the map. For 2023-24 each series is compared with the SAM cells it measures: 57 pass, 11 differ for a documented reason, and 1 has no check. "Explained" means the difference is understood (for example cash versus accrual, or a balancing move), not that it is confirmed. Every trend chart shows the SAM value for 2023-24 as a separate point, so any gap stays visible.

Building the map needed some choices that no published source settles. The project owner approved the balanced SAM as the base for this site at its second checkpoint (Checkpoint B), and accepted the choices below provisionally, on the option we suggested; where that option is not yet in the data, the item says so. Each stays open: questions to the National Accounts Division of MoSPI are pending, and any of them can be revisited when it answers. The flows they touch are labelled estimated on the map (or set by balancing, where balancing set their level), with the method one click away.

• Imputed pension contributions of government employees are not written (D2-20). The evidence favours leaving them out; whether the national accounts include them is a question for MoSPI.
• Factor income by household group uses the same fifth in an older national study (D2-19): the rural and urban groups take the income gradient of the matching national group.
• PM-KISAN is a cash transfer to farm families, not a production subsidy (D2-24), because it pays a flat amount per family.
• Part of personal income tax is paid by unincorporated firms treated as companies (D2-11), following the household share in the previous national-accounts series.
• The GST compensation cess used to repay the back-to-back loans is a current transfer to the States (D2-13), not a capital transfer.
• The previous series' structures fill flows the new series does not yet publish (D2-21), until MoSPI publishes new sector accounts.
• The States' motor-vehicle tax is a tax on products (D2-4), not a direct tax on households.
• The Centre and the States are separated as the national accounts separate them (D2-8): money for central schemes released outside State treasuries counts as the States'.
• The Union's income from the IMF and its SDR charges are recorded on the RBI (D2-7). We suggested moving them to the Centre, and that suggestion was accepted, but the data have not been rebuilt since: on this site both still sit with the RBI. They are small next to GDP.
• The tolerance bands that set levels (see limits) keep their current widths.
• Open findings and smaller choices are accepted as they stand in the data:
  • the surplus that public sector companies keep, with no evidenced outlet (the moves it causes are labelled unexplained);
  • the split of saving between banks and NBFCs, which follows the flow of funds;
  • the household saving profile across the ten groups;
  • the consumption the survey does not capture, spread like the part it does (no top-heavy variant);
  • the banks' service margin (FISIM) on household loans, kept on the current method;
  • the key for who receives private companies' dividends and other distributed income, kept on the flow-of-funds holder shares;
  • the unexplained remainder in the bridge from government budgets to general-government saving, kept as a remainder;
  • a possible overlap between revenue grants to public sector companies and the "business" subsidies in the Finance Commission's data, left as it is;
  • insurance premiums and claims by payer, which the balancing cannot yet require to match;
  • the uneven band on households' fund-held interest from public sector company bonds;
  • a few further time series that could be ratio-linked to the new base, kept unlinked.
• Two proposed adjustments are not yet in the data: adding urban local bodies' capital spending to the key that splits government investment between the Centre and the States, and bounding the share of foreign direct investment income paid abroad by financial firms to the range in the RBI's census of foreign liabilities and assets. Both wait for the next rebuild; until then the flows they touch are as balanced.

Every number on this site traces back to a public source: a publisher, a table or series, the date we retrieved it, its unit and whether it is in current prices. Most come from MoSPI's national accounts and Supply-Use Tables, the household surveys, the Union and State budgets and finance accounts, the RBI's statistics on the balance of payments, flow of funds and State finances, and the regulators of banks, insurance and pensions. The flow panel on the map lists the sources behind each flow, closest inputs first.

Links to RBI pages are withheld. The RBI's website terms do not allow deep links to its pages or files, or caching them, without permission, and that permission has not yet been obtained. Until the RBI grants it, RBI sources are cited by name, publication, table and retrieval date, with no link. The links of 514 source records on RBI sites are withheld this way.

Licences. No source we use publishes its data under India's open-data licence (GODL) on its own site, so reuse depends on each publisher's terms. MoSPI allows reproduction with prominent acknowledgement. The RBI needs to give permission for two things, not one: to link to and cache its pages and files, and to republish figures derived from its statistics. Its Bulletin allows reproduction with acknowledgement (the older issues we use are assumed to say the same), but its Handbook, balance-of-payments, State finances and other releases state no reuse terms and reserve all rights, and the project downloaded every RBI file for internal processing only (D1-1) until the RBI agrees. This site republishes figures derived from RBI statistics, among them the balance-of-payments flows, the flow of funds, household financial saving and the remittance survey shares used in the Dubai story, so that permission must be granted before any public launch. The Ministry of Finance, SEBI, GSTN and DGCI&S also require permission. None of these permissions has been obtained yet; that is why this is a private preview, not a public launch. A few publishers (the Controller General of Accounts, the Sixteenth Finance Commission and PFRDA) state no licence at all; we cite them as they ask, and the first two are on the list of reuse terms to confirm before any public launch. Figures adapted from IFPRI's social accounting matrix are used under its open licence; IFPRI has not reviewed or endorsed this adaptation. Each source's licence and attribution are in the table below, as recorded in the project's source registry.

32 sources are cited, of which 12 are cited without links.

Everything on this site is built from files in the project's repository, by code anyone can run. The raw source files are pinned by a fingerprint (a cryptographic hash), the pipeline rebuilds the same output byte for byte, and the website's data are exported by one command. The data version below names the exact export this page was built from; it changes whenever any input or the export code changes. The table lists each input file with its fingerprint.

The project keeps a public log of everything that does not close (the ISSUES file in the repository). These open items affect what you see here:

• Who receives property income is the weakest link. Without new-base sector accounts, households' interest and dividend income is filled with old-base structures and then moved by balancing to reach the national total of household saving. The total is evidenced; the split is not.
• The Centre and the States. Balancing moved saving, government consumption and some tax receipts between the Centre and the States. The direction has support in the flow of funds; the levels are set by tolerance bands.
• Public sector companies keep more than the national accounts say they save, and the balancing had nowhere evidenced to put it. The resulting moves in the interest, dividends and grants they pay and receive are labelled unexplained.
• The States' subsidies come out well above what State budgets report, because of how the national-accounts subsidy total is defined. This is a question for MoSPI.
• Financial firms' saving. Two official sources disagree on how it is split between banks and NBFCs; the map follows the flow of funds.
• The household saving profile. Saving rates by group rest on survey keys and on an older study's income pattern, and the rural richest fifth looks implausibly low.
• National accounts and the flow of funds disagree on how much each sector lends or borrows. The national accounts are the control, and the differences sit in named residuals.
• Old and new series do not line up evenly. MoSPI expects to publish the new-base back series later; until then the time series shows the break.

Pending permissions before any public launch. The RBI (deep links, caching, and republishing figures derived from its statistics: the balance of payments, the flow of funds, household financial saving, and the remittance survey used in the Dubai story), the Ministry of Finance, SEBI, GSTN, DGCI&S and CAG. Until they are granted, the site is a private preview (ISSUES P1-9).

Questions for MoSPI's National Accounts Division. New-base sector accounts; how departmental enterprises' surplus and pensions are recorded; whether government income includes imputed pension contributions; the subsidy concept; how PM-KISAN, motor-vehicle tax, securities transaction tax and the cess-funded loan repayment are classified; depreciation of dwellings by sector; and the source of imputed rent. One answer could replace several judgement calls on this page.