Disclaimer: This post summarizes and comments on a publicly available research paper. All interpretations are solely my own.

Brent Neiman gave a talk on his paper with Gregor Matvos, Financial Sanctions and the Global Payments Network. The data come from LexisNexis Risk Solutions: standard settlement instructions (SSIs), which record, for each currency a bank can receive, the correspondent(s) it receives through. Stacked together, they give a map of the correspondent network by currency: 17,403 EMDE financial institutions, six snapshots from September 2021 to August 2026, merged with an OFAC/EU/HMT sanctions indicator.

The plumbing. Say Bank A wants to pay Bank B, abroad, in dollars. Each keeps a dollar account at a large clearing bank, its correspondent: A at Citi, B at JPMorgan. Citi debits A, JPMorgan credits B, and the two settle with each other at the central bank. That’s three hops: A to Citi, Citi to JPMorgan, JPMorgan to B. Most dollar chains run through a handful of such banks: the top 10 hold 77% of the dollar network’s PageRank, a measure of how central a node is.

What sanctions do. They remove edges at those hubs. Targeted banks rarely drop off the network; they reroute. Among sanctioned banks that keep dollar access, the PageRank of their dollar correspondents falls from about 15% to 1.8%, and hops to their top-5 trading partners rise from 2.07 to 2.81, about 36%. Each extra hop means another compliance screen, more pre-funded nostro balances, counterparty credit risk and no settlement finality. The paper’s reading: sanctions make payments costlier, slower and riskier rather than impossible.

Spillover. Untargeted banks in heavily targeted jurisdictions also lose their major-currency correspondents, likely through de-risking. A correspondent prices a respondent partly by its neighbors, and once enough of a banking system is designated, the whole system gets repriced. These banks shift toward alternative currency networks. Elsewhere, little has moved: the dollar or euro still offers the shortest path for about three-quarters of EMDE–foreign bank pairs, as in 2021.

The point I liked. The concentration that makes the dollar network efficient is the same concentration that makes it sensitive to the removal of a few nodes. Few hops require few hubs. A hub is a shortcut and a chokepoint at once; which one you see depends on which end of the edge you sit.

Sources