External debt methodology

Every figure in the /external-debt pages comes from a build-time query against the World Bank International Debt Statistics extract in this repository (120 economies, 24 indicators, 1970 onward), except in the clearly-marked sections that carry the BIS debt-service-ratio series. This page records what each series is, what was derived rather than read, and where the data cannot support a claim.

Source and universe

IDS is a borrower-sidecollection. The World Bank’s Debtor Reporting System requires countries that borrow from the Bank to report their external obligations loan by loan. Two consequences run through everything on these pages.

• The reporting population is a queried set, not a description. The World Bank classifies the 120 reporters as follows: 49 upper middle income, 46 lower middle income, 23 low income, 1 with no income group assigned, 1 high income. G7 members in the file: 0 of 7, and the build throws if that ever stops being true, because the pages say so in prose. The 120-country total is not world external debt and must never be read as such.
• It is reported, not observed. These are the numbers borrowing governments gave the World Bank, not positions measured in markets or at creditor banks. Unreported or late-reported debt is not here, and IDS gives no way to size what is missing.

Source: World Bank International Debt Statistics. Tables: ids_debt (long format: iso3, country_name, indicator_code, year, value), ids_indicators (the 24-row dictionary), ids_countries (the 120-row country list, with income group, region and lending category). The iso3 sets of ids_debt and ids_countries are identical, so every reporting economy has a page and no page is built for an economy with no data.

Units

Indicator codes ending .CD are raw current US dollars, not thousands and not millions: China’s 2024 total external debt is stored as 2,419,840,000,000. Codes ending .ZS are percentages. Dollar series are current-price throughout; IDS carries no constant-price variant, so no chart on these pages is deflated and none should be read as a real-terms path.

Actual years versus the contracted schedule

This is the largest trap in the file. ids_debt runs to 2032, but only 2 of its 24 indicators carry values past 2024, and both are debt service:

• Debt service on external debt, total (TDS, current US$)
• Debt service on external debt, public and publicly guaranteed (PPG) (TDS, current US$)

Those out-years are the amortisation profile of debt already contracted: principal and interest that existing loans and bonds fall due to pay. They are not a forecast of what a country will owe or pay, because borrowing not yet contracted cannot appear in them. The aggregate therefore declines mechanically toward 2032, and a reader who takes that decline for a projection has been misled. On these pages the schedule is drawn dashed, on a shaded panel, labelled SCHEDULED inside the plot, and never blended into a stock series or a ratio.

The boundary is derived, not typed: the last actual year is the maximum year of the anchor stock series DT.DOD.DECT.CD, currently 2024. Every stock and ratio query is cut there. When IDS publishes a vintage whose stocks reach further, the cut moves on its own and no page needs editing. The build throws if that anchor series ever returns no years.

The creditor decomposition, and what is derived in it

IDS does not hand over a clean creditor split. It gives the three legs asymmetrically, and one of them not at all.

• Private creditors (DT.DOD.PRVT.CD) is a reported dollar level of PPG debt. IDS defines it as bonds, commercial bank loans, and other private credits, together. There is no bond-versus-bank split in this file, so no page claims one.
• Multilateral (DT.DOD.MLAT.ZS) is reported only as a percent of total external debt. It is re-levelled onto dollars (share times DT.DOD.DECT.CD) before being compared with PPG.
• Bilateral and other official is not a series. It is derived as the residual: PPG minus private minus multilateral. It therefore contains bilateral official lending, any other official creditor, and any private PPG debt a country did not report. Chart legends and table rows label it “bilateral and other official (residual)”; running prose shortens that to “bilateral and other official”, and it is a residual wherever it appears.

Two things must hold for that derivation to mean anything, and both are asserted at build time. First, multilateral lending must sit inside PPG, or dividing it by PPG is nonsense. IDS does not state this in the indicator definition, so it is treated as an assumption. The evidence for it is that the residual comes out non-negative in all 5,705 country-years where the three inputs exist, which is consistent with containment but does not prove it. Second, the residual must be non-negative; the build throws if it ever is not, and the composition chart does not ship.

The residual also absorbs a measured amount of unreported private debt. In 2024, 20 of the 120 reporters carry no private-creditor value at all. A missing value is treated as zero, so whatever private PPG debt those countries hold lands in the residual. They account for 1.32% of total PPG, which bounds the error from this source.

The balanced panel

Countries enter IDS over time. Aggregating a composition across a changing membership turns an entry effect into an apparent shift in creditor structure. Every creditor-composition series on these pages therefore runs on a balanced panel: the 112 countries that report PPG stocks, private-creditor debt and the multilateral share in every year from 2000 through 2024. That covers the stacked-area chart on the index and the country-by-country direction table beneath it. The membership test is rebuilt from the data at query time, against the derived last actual year, so a refreshed vintage automatically tightens it rather than silently keeping a stale panel. 2000 is an analytical window choice, not a data literal: stocks begin in 1970, so the window can never select zero rows.

The debt-service aggregate is the exception, and it is not balanced. It is a level in dollars, not a share, so an entry effect cannot masquerade as a composition shift in it. It sums whichever economies carry DT.TDS.DECT.CD in each year: 77 in 1970, 120 in 2024. Its early years are therefore a sum over fewer countries than its recent ones, and the counts are printed under the chart.

The country-page composition charts are notbalanced-panel: each shows that country’s own reported years, which is why a country page and the index can start in different years.

The stock decomposition

IDS defines total external debt (DT.DOD.DECT.CD) as the sum of public and publicly guaranteed long-term debt, private nonguaranteed long-term debt, short-term debt, and use of IMF credit. The index page reports that split with PPG (DT.DOD.DPPG.CD), private nonguaranteed (DT.DOD.DPNG.CD) and short-term (DT.DOD.DSTC.CD) read directly, and use of IMF credit taken as the residual: total minus long-term (DT.DOD.DLXF.CD) minus short-term. That residual is IMF credit by the definition above, not an assumption. It is non-negative for every reporter in 2024, and the build throws if it is not.

The BIS debt-service ratio, and why it is kept apart

The BIS debt-service-ratio series covers 32 economies, of which 8 also report to IDS. It measures the debt service of the private non-financial sector on allof its debt, domestic and external, as a percent of income. IDS debt service measures a country’s external obligations, public and private, against exports. Different borrower, different debt, different denominator, and a universe that overlaps by 8 of 120. The two are shown side by side once, on the index, precisely so a reader can see that they do not correspond. They are never summed, ranked against each other, or plotted on shared axes.

Coverage, and what a blank means

Coverage varies by indicator, and a missing value has three possible causes that IDS does not distinguish: the country did not report the series; the series does not apply to it (a country with no IBRD loan has no IBRD stock, which is why DT.DOD.MIBR.CD looks sparse); or the series has ended for everyone. One indicator has ended: Debt forgiveness grants (current US$), last value in 2023. Two blanks are summed as zero, and both are stated where they occur: PPG owed to private creditors, which pushes any unreported private debt into the bilateral residual and is bounded above at 1.32% of PPG; and private nonguaranteed debt in the stock split, which 93 of 120 reporters carry.

What this data cannot support

• No creditor country. The extract has no creditor-country dimension. It cannot say who lent, only what class of lender. In particular there is no China series, so the shift from Paris Club bilateral lending toward Chinese bilateral lending cannot be shown or tested from these 24 indicators. What is visible is that the bilateral and other official leg shrank; who left it, and who remains inside it, is not in this file.
• No bond-versus-bank split. DT.DOD.PRVT.CD bundles bondholders with commercial banks and suppliers.
• No currency, no maturity beyond one year. The only maturity cut in the file is short-term (one year or less) against long-term. There is no currency composition, so nothing here speaks to original sin or to FX mismatch.
• No interest rate, no spread.There is a present-value series (DT.DOD.PVLX.CD) and a concessional share (DT.DOD.ALLC.ZS), but no contract rate, so “on what terms” is answerable only through concessionality and maturity, not price.
• Not world external debt. Repeated here because it governs every aggregate above: the universe is the 120 DRS reporters, 0 of them G7 members, so the totals are the external debt of those reporters and of nobody else.

Indicator dictionary

All 24 indicators in ids_indicators, with the observed span and country count of each series in ids_debt. Definitions are the World Bank’s own.

Source for every definition above: World Bank, International Debt Statistics. The Years column is the span of non-null values actually present in this extract, which is why the two debt-service indicators reach past 2024: those years are the contracted schedule, not observations.

Back to external debt