Skip to content
FinObservatory

Chinese overseas lending / Two sources

The gap everyone wants to compute does not exist

The obvious question is how much Chinese lending the official statistics miss: take the estimate of what is owed to China, subtract what the borrower reports owing to China, call the difference hidden debt. That subtraction cannot be performed with this data. The World Bank International Debt Statistics table in this estate carries exactly 24 indicator codes, listed in full below, and not one of them is a creditor-country series. “Debt owed to China” is not a quantity the borrowers report here, so no China-versus-reported difference is defined. What remains available is a ratio, and a bound.

Earlier vintage comparison

The two published stock vintages overlap from 2000 to 2017

This is a like-for-like aggregation over borrower-years present in both releases. The solid line is the 2021 HRT replication vintage; the dashed line is the November 2025 refresh, restricted to those same mapped observations. They are separate estimates, not a spliced history, correction, or calculated revision.

2021 replication, 108 observed borrowers, published debt coverage 20002017November 2025 refresh, 126 borrowers, publication coverage 20002021

External debt to Chinese official creditors, aggregate USD bn. Earlier source dollars are nominal USD converted to billions at query time. Each year includes only mapped borrowers with a published earlier observation, from 105 in 2000 to 107 in 2017.

Latest common year: 2017, 107 observed overlapping borrowers

Source vintageExternal debtPublic debt / PPGPublication coverage
2021 HRT replication$391.95bn$362.83bn20002017
November 2025 refresh$387.89bn$294.90bn20002021

The earlier field is external public debt; the refreshed field is public and publicly guaranteed debt. Both are displayed under their source definitions, so the levels should not be read as a borrower-by-borrower revision.

Earlier workbook grid

7,767

rows, 144 labels, 19502017

Published earlier observations

1,927

records, 108 source labels, 20002017

Comparable overlap

1,909

borrower-years, 107 mapped borrowers

Excluded null grid rows

5,840

both earlier China debt fields blank, never zero-filled

Mapping reconciliation: 0 published observations on 0 source labels are unmapped. 18 observations on 1 mapped borrower is outside the refreshed overlap.

Albania (ALB): 18 observations, 20002017, outside current borrower universe.

Sources: Horn, Reinhart and Trebesch (2021) replication archive (hrt_china_debt), and the November 2025 refreshed China debt-stock workbook. Exact source labels are preserved; normalization is explicit, with no fuzzy matching.

Every indicator the reported-debt table carries

The complete list, queried from the table rather than recalled. The table does break debt out by creditor class in places: multilateral debt (DT.DOD.MLAT.ZS), IBRD loans (DT.DOD.MIBR.CD), IDA credits (DT.DOD.MIDA.CD). What it never does is break debt out by creditor country.

  • DT.AMT.DLTF.CD
  • DT.DOD.ALLC.ZS
  • DT.DOD.DECT.CD
  • DT.DOD.DECT.EX.ZS
  • DT.DOD.DECT.GN.ZS
  • DT.DOD.DECT.PC.CD
  • DT.DOD.DLXF.CD
  • DT.DOD.DPNG.CD
  • DT.DOD.DPPG.CD
  • DT.DOD.DSTC.CD
  • DT.DOD.DSTC.ZS
  • DT.DOD.MDRI.CD
  • DT.DOD.MIBR.CD
  • DT.DOD.MIDA.CD
  • DT.DOD.MLAT.ZS
  • DT.DOD.PRVT.CD
  • DT.DOD.PVLX.CD
  • DT.INT.DECT.CD
  • DT.NFL.DECT.CD
  • DT.NTR.DECT.CD
  • DT.TDS.DECT.CD
  • DT.TDS.DECT.EX.ZS
  • DT.TDS.DPPG.CD
  • FI.RES.TOTL.DT.ZS

Source: Horn, Reinhart and Trebesch, China’s Overseas Lending (estimated stocks), and World Bank International Debt Statistics, borrower-reported via the Debtor Reporting System (all-creditor stocks). Selection: SELECT DISTINCT indicator_code from the International Debt Statistics table, all 24 of them. Methodology

What can still be tested: a logical bound

An estimate of what one creditor is owed cannot exceed what the borrower reports owing to all creditors. That is a bound, not a model, and it needs no assumption about what China lent. In 2021 it breaks for 3 of the 102 borrowers that appear in both sources.

CountryEst. owed China, externalReported to all creditorsRatio, %Est. owed China, PPGReported PPG, all creditorsRatio, %
Eritrea$2.14bn$772m277.5%$488m$691m70.6%
Equatorial Guinea$1.63bn$1.50bn109.2%$1.63bn$1.13bn144.0%
Iran$10.25bn$10.35bn99.1%$9.88bn$332m2972.2%

Source: Horn, Reinhart and Trebesch, China’s Overseas Lending (estimated stocks), and World Bank International Debt Statistics, borrower-reported via the Debtor Reporting System (all-creditor stocks). Selection: estimated external (PPG + private non-guaranteed) and PPG stock owed to China in 2021, against borrower-reported DT.DOD.DECT.CD and DT.DOD.DPPG.CD for the same year; every borrower where either estimate exceeds the reported all-creditor stock. Drawn swap balances are excluded from both numerators: they are a central-bank liability the Debtor Reporting System's long-term-debt concepts do not carry. Methodology

A violated bound is a disagreement, not a proven concealment

Two readings survive the data, and this page asserts neither.

The reported record is incomplete

The Debtor Reporting System is a return filed by the borrowing government about its own obligations. How thin those returns can be is measurable: of the 102 borrowers with any row in 2021, 24 file nothing at all for private non-guaranteed debt, and just 1 reports a genuine zero. A category left blank by 24 of 102 filers is not a category a bound can lean on. If a borrower under-reports, the ratio above is measuring that.

The estimate is wrong

The China figure is not observed either. It is reconstructed from loan-level commitment records, and the source imputes the missing maturities and grace periods by creditor entity and instrument type. A commitment that was announced but never fully drawn, or repaid faster than the imputed schedule assumes, leaves the estimate too high. Nothing in either dataset says which of these two boxes is the right one, so a violated bound cannot arbitrate between them, and this page does not try.

The borrowers with no row in the reported-debt table

24 of the 126 borrowers in the China panel have no row in the reported-debt table at all. Between them these 24 hold $95.29bn of the 2021 stock, 20.18% of the total. That headline share is mostly one country: Russia alone accounts for $67.50bn of it, or 70.8%. Reporting the 20.2% without naming Russia would be a true number carrying a false implication. 4 of the 24 carry a reported zero stock in 2021.

CountryEst. stock owed to China, 2021Share of the $95.29bn
Russia$67.50bn70.8%
Malaysia$7.15bn7.5%
Venezuela$6.59bn6.9%
Oman$3.71bn3.9%
Brunei Darussalam$1.86bn1.9%
South Sudan$1.58bn1.7%
Bahamas$1.29bn1.4%
Chile$1.18bn1.2%
Cuba$1.01bn1.1%
Israel$814m0.9%
Bulgaria$654m0.7%
Namibia$489m0.5%
Costa Rica$353m0.4%
Trinidad and Tobago$295m0.3%
Romania$251m0.3%
Panama$176m0.2%
Antigua and Barbuda$164m0.2%
Barbados$159m0.2%
Kiribati$54m0.1%
Cook Islands$19m0.0%
Micronesia$00.0%
Seychelles$00.0%
Uruguay$00.0%
Libya$00.0%

Source: Horn, Reinhart and Trebesch, China’s Overseas Lending (estimated stocks), and World Bank International Debt Statistics, borrower-reported via the Debtor Reporting System (all-creditor stocks). Selection: every borrower in the China panel with no row in the International Debt Statistics table, and its estimated stock in 2021. A reported zero stock stays 0; it means no known loans outstanding, not a missing figure. Methodology

Why these two numbers must never be subtracted

  • One is estimated, one is reported. The China figures are reconstructed from loan-level commitments with imputed schedules. The all-creditor figures are what borrowers themselves filed. A difference between them mixes a measurement error with a reporting gap and cannot separate the two.
  • They cover different instruments. Per its own documentation, the China estimate excludes short-term trade debt and portfolio debt, and does not account for missed principal payments. The reported total external debt stock (DT.DOD.DECT.CD) includes short-term debt and IMF credit. Neither is a subset of the other.
  • Swap drawings belong to neither comparison. A drawn central-bank swap is not a category the Debtor Reporting System’s long-term debt concepts carry. Every ratio on this page therefore uses the estimated external stock (public and publicly guaranteed plus private non-guaranteed), never the total including swaps. Including them would manufacture bound violations out of a definitional mismatch. The swap numbers get their own page: /china-lending/swaps.
  • Arrears are not in the stock. Debt in default, including debt in default to China, lives on /sovereign. The estimated stocks here do not account for missed principal payments, so the two can neither be summed nor used to validate each other.