Methodology

The source

The External Wealth of Nations database (Lane and Milesi-Ferretti): 212 economies, 1970 to 2024, holding each country's foreign assets and liabilities broken down by instrument. 195 of those economies have at least 15 years with a net position and a complete instrument split, and those are the ones with a page.

Equity-like and debt-like

External liabilities are split into two economically different kinds. Equity-like liabilities are inward direct investment and portfolio equity: the foreign investor owns a claim whose value falls when the country does badly, so losses are shared. Debt-like liabilities are portfolio debt and other investment (mostly cross-border bank lending): fixed claims that do not fall when the country does badly, and that get larger in local-currency terms when the currency drops, which is exactly when they are hardest to service.

equity share = (FDI + portfolio equity) / (FDI + portfolio equity + portfolio debt + other investment)

A country is only given a composition share in a year where all four instruments are reported. A share computed with a missing leg is not a share. The denominator is the sum of those four legs, not EWN's total-liabilities aggregate: that aggregate carries a fifth leg, financial derivatives, which the instrument split on these pages leaves out on both sides of the ratio. The four shares therefore sum to 100 percent of the four legs, and for a country with a large derivative book they describe a smaller stock than its reported total liabilities.

Crisis episodes are deduplicated, and that matters

The FinObservatory crisis atlas holds one row per source, across five chronologies. A raw row count therefore inflates a country's crisis history by up to five times: 6,518 rows in the atlas are 4,798 actual episodes. Every crisis figure on these pages counts DISTINCT (country, year, type). Getting this wrong would not have changed the shape of the composition table, but it would have made every rate printed in it a fiction.

The result we did not want

The argument that motivates this module predicts that debt-financed countries are more crisis-prone than equity-financed ones. Sorting economies into terciles by their average equity share and counting distinct episodes from 1970 onward against the country-years that carry a complete instrument split, that ordering does not appear:

Two things about that table. The terciles cover 211 economies, more than the 195 with a page: a tercile needs only 15 years of a complete instrument split, while a page also needs a net position and GDP in those years, and a set of small jurisdictions reports the split without the position. And the two sides of the rate cover different years. The numerator is every distinct episode from 1970 onward; the denominator counts only the years with a complete split. The comparison across the three groups is the point of the table, and the printed level is not a hazard rate.

The middle group has the most crises. The page reports that rather than quoting the two end groups, which do line up with the story, and calling it a finding. The index page asserts this non-monotonicity in code: if a future data vintage ever makes the relationship monotonic, the build fails rather than shipping a page whose prose contradicts its own table.

What is not here

• The currency of the liabilities. Whether a country owes the world in a currency it can print is arguably the single most important fact about its external position. All 25 EWN variables were checked: there is no currency breakdown. The original-sin story cannot be told from this file, so it is not told here.
• A reported-versus-estimated flag. EWN is a reconstruction. Where a country never reported a position, the authors estimate it, and the released file carries no flag distinguishing the two. These pages therefore cannot mark estimated cells, and do not pretend to.
• Causation. Nothing assigns an economy to a tercile at random. The three groups differ in far more than the composition of their liabilities, and any of those differences could produce the gap between the outer groups with external composition doing no work at all.

The other net-position series on this site

The country pages plot a net international investment position taken from the IMF balance-of-payments data. It is a different vintage of the same concept, from a different compiler, and the two will not agree to the decimal. EWN is canonical here because it is the only one of the two with the instrument split these pages are built on.