Input-Output Tables — Algeria (2000-2021)

What every branch buys from all the others in order to produce. 17,251 observations across twenty-two years, 2000 to 2021: nineteen branches down the rows, nineteen branches and ten final uses across the columns. The full matrix of exchanges in the Algerian economy.

The Input-Output Table reads two ways. The row says which product is consumed — agriculture, hydrocarbons, metals and machinery, construction, textiles, transport, household services. The column says by whom: another branch that transforms it, or a final use — household consumption, government, gross fixed capital formation, change in inventories, exports. The cell says how much.

To which is added the lower block, worth the price on its own: value added, wages and salaries, production taxes, gross and net operating surplus, consumption of fixed capital, gross output, imports, VAT, customs duties and trade margins — each broken down across the nineteen branches. Who creates value, who pays wages, who clears a surplus.

Three accounting identities, verified row by row. Total intermediate consumption equals the sum of the nineteen branches; total uses equal intermediate consumption plus final uses; final uses equal the sum of their seven components. Sixty-three tables out of sixty-three pass all three, and 675 rows out of 675 still satisfy the first in the delivered file — after arbitration between editions, which was not a given.

A cross-check nothing obliged to succeed. Total gross output in this table equals that of our "Integrated Economic Accounts" across the sixteen overlapping years, to the dinar. Two different documents, two unrelated layouts, two extraction pipelines written separately. Agreement does not prove both are right, but it would make it very unlikely that both are wrong in the same way.

What the layout demanded, and what ignoring it cost. One year spans two pages, and the right-hand page carries no labels: its rows are identified only by rank. Assuming rank zero answers rank zero is a bet — the two pages share neither row heights, since a wrapping label deepens its row on the left and not on the right, nor necessarily row counts. The offset is therefore measured, not assumed: we keep the one that most often satisfies the identity crossing both pages. Across sixty-three tables, exactly one was not at offset zero. Without that measurement it would ship, wrong, with nothing to flag it.

Three further traps are worth naming, because they say what the file avoided. Columns are centred: reading their left edge spreads one column over eighteen points where its centre holds it within seven, and values fell outside tolerance. A branch's code, label and values occupy up to four lines whose order settles nothing — one fragment follows the values of its own row exactly as another follows those of the preceding row, to which it does not belong; only position separates them. Finally, two neighbouring wide totals end up glued into one fragment: "1726162 2515442", which stripping spaces without looking turns into a fourteen-digit number of entirely normal appearance. That single fix took the extraction from sixteen tables to thirty-nine.

Unit, scope and reservations. Millions of current dinars, as published, unprocessed. The eighteen branches the source publishes are present every year; the sixteenth, Trade, is absent from every edition — a property of the document, not of the reading. 10,320 observations are confirmed by two distinct publications. 362 rows were revised between editions: the most recent is kept, and a column flags it, so the buyer knows where the source corrected itself.

This product reads alongside our "Integrated Economic Accounts", which take the economy by institutional sector where this one takes it by branch.