Cross-border bank claims data: who lends across borders, by country, currency and quarter

Cross-border bank claims data records what the banking offices located in one country claim from, and owe to, residents of another. The reference record is BIS Locational Banking Statistics (LBS) & BIS Data Portal: quarterly stocks of outstanding claims and liabilities of internationally active banks in roughly 50 reporting countries against counterparties residing in more than 200 countries, split by counterparty sector, instrument, currency and position type.

Three properties define the series. Positions follow a residence perspective - the location of the banking office - compiled on principles consistent with balance-of-payments statistics under the Committee on the Global Financial System. They are unconsolidated, so intragroup positions between offices of the same banking group stay in. And the BIS estimates the reporting perimeter captures about 95% of all cross-border banking activity - near-census reach no commercial vendor reconstructs from national fragments. Every observation is denominated in millions of US dollars, and the current release carries 609,065 series rows x 225 columns: 24 dimension and metadata columns followed by one column per quarter from 1977-Q4 through 2026-Q1.

Within Datadory's catalog the record scores 10 out of 10 - one of only 145 perfect scores among the 1,744 datasets cataloged, against a catalog-wide average of 7.81. It anchors the eleven-record diversified-banks shelf precisely because nothing else sees the whole world's cross-border bank balance sheet in one schema.

One record answers who lends across borders; four neighbours make it decision-grade. Staged together rather than alone:

• BIS Consolidated Banking Statistics - the nationality-based companion shipped beside the locational panel, netting intrabank flows out to show where ultimate risk comes to rest regardless of which office booked it.
• ECB Statistics - Consolidated Banking Data (CBD2) - quality 9: 66,139 series (40,227 annual, 25,912 quarterly) across 30 reference areas including the UK, with capital, funding and profitability items for EU banking groups from Q4 2007 onward.
• World Bank Global Financial Development Database - quality 9: 108 indicators across 214 economies annually from 1960, holding the system-level stability and depth context around any claims number.
• FDIC BankFind Suite API and SEC EDGAR XBRL Financial Statements API - both quality 10 - reconstruct the American offices behind the US contribution to the global totals at institution and listed-parent level respectively.

The locational panel stops at the border; these start there. Datadory resolves the identifier mappings up front so the stack arrives pre-joined instead of as five exports reconciled by hand.

Eight documented dimensions address a banking stock before any value appears, which is what makes the panel filterable without a join:

• L_MEASURE - S = amounts outstanding / stocks; F = FX- and break-adjusted change computed upstream to approximate underlying flows; G = annual growth; B = break in stocks. Pick the measure before the analysis, not after.
• L_POSITION - side of the balance sheet: total claims (C) or total liabilities (L).
• L_INSTR - instrument cut: all instruments (A), loans and deposits (G), debt securities (D), currency and deposits (C).
• L_DENOM / L_CURR_TYPE - the currency a position is booked in - USD, JPY, TO1 for all currencies - and whether it is domestic or foreign-currency business relative to the reporting country, the two axes behind most global-dollar-credit work.
• L_REP_CTY - whose banks hold the position: 50 reporting-country codes with aggregates beside them, 5A for all reporters and 5C for the euro area.
• L_CP_SECTOR - resident sector of the counterparty: all sectors (A), banks (B), non-banks total (N).
• L_CP_COUNTRY - where the counterparty resides: more than 200 country codes plus 5J for the world.
• L_POS_TYPE - cross-border positions (N) versus local lending by foreign-owned offices to local residents (R), the book most risk registers omit because they count only cross-border claims.

Because the counterparty axis runs to individual countries rather than regions, one filter combination - claims, non-banks, a single country code - isolates every reporter's exposure to that economy for every quarter since 1977. In a Datadory delivery the code-and-label pairs arrive resolved and the wide quarter grid arrives tidy, so consecutive releases diff cleanly instead of by hand.

Two verified series rows from the current release, exactly as they land before normalization:

freq          : Q
measure       : S  (amounts outstanding / stocks)
instrument    : G  (loans and deposits)
currency      : JPY (yen)
reporting cty : 5A (all reporting countries)
cp sector     : N  (non-banks, total)
cp country    : YE (Yemen)
quarters      : 1977-Q4 ... 2026-Q1   # one value column per quarter

freq          : Q
measure       : F  (FX and break adjusted change)
position      : C  (total claims)
reporting cty : BR (Brazil)
curr type     : F  (foreign currency)
pos type      : R  (local positions of foreign-owned offices)
1977-Q4 value : 1643.983

The first row reads plainly once decoded: all reporting countries combined held a stock of yen-denominated loans and deposits against Yemeni non-banks, observed quarter by quarter across five decades. The second shows why the derived measures matter - a foreign-currency local-position series expressed as FX- and break-adjusted change, so valuation swings never impersonate new lending.

Values ship unrounded exactly as published, genuinely blank where a series has no observation rather than zero-filled, and the collection indicator travels with each row (end-of-period versus summed-through-period) so timing assumptions stay explicit instead of inherited.

Temporal: quarterly observations from 1977-Q4 through 2026-Q1 - 195 quarters per series. Long enough to carry funding-stress and carry signals through several full credit cycles, which is why backtests keep returning to this panel rather than to shorter commercial alternatives. The break-in-stocks flag (measure B) marks structural discontinuities, so point-in-time research can exclude revised history instead of discovering it mid-model.

Geography: roughly 50 reporting countries across Europe, the Americas, Asia-Pacific and the offshore centres, against counterparties in more than 200 countries, with world and euro-area aggregates sitting beside the individual codes.

Granularity: reporting country x counterparty country x counterparty sector x instrument x currency x position type x measure, per quarter, in millions of US dollars or growth ratios where the measure calls for them.

Two seams deserve planning rather than discovery. Russian public-authority reporting stopped after February 28, 2022, so those series end there abruptly - a collection fact, not a gradual fade. And officially compiled figures get revised between releases, which means provenance belongs beside every value rather than in a folder name. Datadory ships the vintage with each observation and flags the breaks upstream of your warehouse, before they contaminate anything downstream.

Both describe international banking; they disagree on purpose. The locational view follows the residence of the office, so a Japanese bank's New York branch lending to France books as US claims. The consolidated view follows the nationality of the group and strips intragroup positions, so the same exposure nets back to Japan's ultimate risk.

The practical consequence: funding-chain and liquidity questions want locational, because money moves through offices; country-risk and contagion questions want consolidated, because loss absorption happens at the group. Analysts who pick one and call it the other produce the classic mismatch - a "Japanese exposure" number that is really an American office's balance sheet. Delivered together under one schema, the disagreement itself becomes informative: where residence-based and nationality-based totals diverge widest, intragroup plumbing is doing the heavy lifting.

Global-liquidity monitoring. Dollar- and yen-denominated cross-border credit to non-bank borrowers is the canonical gauge of the global financial cycle. Measure F strips the valuation noise, and switching the counterparty sector to N isolates non-bank borrowing alone - the raw material for funding-stress dashboards.

Country-risk screening. Counterparty-level claims by sector let credit committees size what a given sovereign's banking system actually owes abroad before the headline gets politicized - China as a counterparty code is one filter, not a modeling project.

Backtesting across regimes. Five decades of quarterly claims and liabilities give strategies multiple cycles to prove themselves against, with the break flag keeping revised history out of point-in-time tests.

Offshore-exposure mapping. Position type R isolates local lending by foreign-owned offices - the exposure most registers miss because they count only cross-border claims.

Reconciling the global book against US microdata. The FDIC BankFind Suite API tracks roughly 4,700 active insured institutions with quarterly financials reaching toward the 1930s and more than 4,100 recorded failures, grounding America's contribution to the BIS totals charter by charter; the FDIC Quarterly Banking Profile adds the industry aggregate across 160+ quarters archived since Q1 1986. SEC XBRL companyfacts contributes per-concept GAAP fundamentals for the listed parents - a single large-bank file runs about 8 MB across 900+ concepts, with JPMorgan facts verified back to 2008. The Wikipedia List of Largest Banks - 100 banks across 24 countries and territories, ranked on S&P Global Market Intelligence figures from April 2026 - supplies a quick entity spine for resolving reporters to groups.

Every one of these starts the same way: tell Datadory the counterparties, currencies and measures, and get a sample cut from the same normalized feed production would use.

The table lines up the stack as of August 2026. They differ by grain, history and the job each settles - not by which one is 'best':

Files, feeds, or your warehouse. Daily, weekly, or hourly - your call.

For the wider pool, see how the slice ranks in the best diversified-banks datasets list, browse the whole shelf on the diversified-banks data hub, or read the head-to-head argument in BIS LBS vs FDIC Quarterly Banking Profile. When you want rows instead of reading, request a sample on any dataset page - the field dictionary travels with it.