Trade and Cross-Border Bank Claims

Abstract

Cross-border bank lending is often assumed to follow trade, because banks finance and insure international transactions. This paper tests that idea by matching merchandise imports from UN Comtrade with cross-border bank claims from the BIS Locational Banking Statistics: 493 observations on 62 country pairs over 2015–2023. Across pairs, a 1% larger import flow is associated with 0.9–1.2% larger bank claims; within pairs over time the relationship is much weaker and mostly insignificant.

Research focus

Whether bilateral trade is associated with cross-border bank lending, using a country-pair panel for 2015–2023.

Paper title

Is Bilateral Trade Associated with Cross-Border Bank Claims? Evidence from a Country-Pair Panel, 2015–2023

Keywords

Bilateral trade; cross-border banking; BIS locational banking statistics; gravity model; PPML; fixed effects

Research question

Do country pairs with larger bilateral imports hold larger cross-border bank claims — and within a pair, do claims rise when trade rises?

Motivation

Banks support trade through trade credit, guarantees and payment services, so financial linkages should be larger where trade is larger. The original aim was a two-step test — whether tariffs reduce imports, and whether trade is associated with bank claims — but tariff data (World Bank WITS) were unavailable, so the paper concentrates on the second step.

Hypothesis H1 (cross-section)

Country pairs with larger bilateral imports hold larger cross-border bank claims (positive association).

Hypothesis H2 (within pair)

Within a given pair, years with higher imports are associated with higher bank claims — a stricter test, because pair-level differences are removed.

Hypothesis H3 (heterogeneity)

The trade–banking association differs by claimant type (banking centres, emerging markets, other advanced economies) and between the pre-2020 and post-2020 periods.

Theoretical background

Gravity model of trade (Tinbergen 1962; Anderson & van Wincoop 2003; Head & Mayer 2014); PPML estimation (Santos Silva & Tenreyro 2006); gravity in cross-border equity flows (Portes & Rey 2005); joint trade–asset holdings with two-way causality (Aviat & Coeurdacier 2007). Results are therefore interpreted as associations, not causal effects.

Data — imports

UN Comtrade: annual total merchandise imports reported by the importing country, current US dollars, 14 reporters and 10 partners over 2014–2023 (1,400 observations, no missing cells).

Data — bank claims

BIS Locational Banking Statistics: total cross-border claims of banks in the reporting country by counterparty residence, all instruments, currencies and sectors; year-end (Q4) stocks in USD millions; seven reporting countries (BR, CA, CL, MX, CH, UK, US).

Matching and sample

A pair-year is kept only when both series exist. Seven of the 14 Comtrade importers are BIS reporters; the BIS file omits 2014 and some series are suppressed for confidentiality. Final sample: 493 observations on 62 country pairs (of 70 possible), 2015–2023, with 53–57 pairs per year. No values imputed.

Summary statistics

Bank claims: mean USD 7,736 m, median 1,258 m, max 103,771 m. Imports: mean USD 20,729 m, median 3,704 m, max 575,688 m.

Empirical strategy

Six models: pooled OLS through to pair, claimant-year and counterparty-year fixed effects (the most demanding specification); a one-year trade lag to reduce simultaneity; and PPML with claims in levels, pair and year effects, plus a 300-draw pairs bootstrap. Standard errors clustered by country pair (62 clusters).

Results — cross-section (H1)

Across pairs, a 1% larger import flow is associated with 0.9–1.2% larger bank claims. The association holds before and after 2020 and across claimant types. Supported.

Results — within pair (H2)

Much weaker: log-linear elasticity 0.20 and statistically insignificant; PPML gives 0.36 (p = 0.03); the long-difference regression gives 0.45 (p = 0.27). Not supported, except in the PPML specification that emphasises large pairs.

Results — heterogeneity (H3)

No statistically significant differences between banking centres (UK, CH), emerging-market claimants (BR, CL, MX) and other advanced economies (CA, US), nor between 2015–2019 and 2020–2023. Not supported.

Interpretation

A strong structural link between trade and banking exists across country pairs, but there is little evidence that year-to-year changes in a pair's trade are matched by changes in its bank claims. The cross-sectional association likely reflects common structural factors — size, proximity, history, financial development — rather than a short-run link between the two flows.

Limitations

Small sample (seven claimants, ten counterparties, including two major banking centres); imports only, not total bilateral trade; associations, not causal effects; banking data omit 2014; dollar claim stocks are affected by exchange rates and valuation changes; 62 clusters limit the power of within-pair estimates.

Future work

Add bilateral tariff data (WITS) to test whether tariffs reduce imports and, through that channel, claims; extend the banking panel to 2014 and more claimants; use total bilateral trade. A gravity specification with tariffs, importer-year and exporter-year effects and PPML is already prepared.

Key references

Amiti, Redding & Weinstein (2019); Fajgelbaum et al. (2020); Anderson & van Wincoop (2003); Portes & Rey (2005); Aviat & Coeurdacier (2007); Santos Silva & Tenreyro (2006); Tinbergen (1962); Yotov et al. (2016).

Publication status

Working paper, October 2026.