Before 2007, banking crises struck the 18 advanced and the 218 other economies at 3.6% (24/666) and 4.2% (338/7,971) of country-years; across 2007–2017 the advanced share rose to 22.2% (44/198)

What Reinhart and Rogoff claimed

Their paper assembled a long-run banking-crisis chronology for 66 countries and reported that the share of years spent in a banking crisis is close across income groups, so that banking crises are, in their phrase, an equal-opportunity menace, roughly as common in high-income countries as in middle-to-low-income ones. The estate carries a Reinhart-Rogoff chronology among its five sources, but as an annual start-year record covering 70 countries in the data, not the paper’s own income-weighted 66-country vintage, so this note does not reproduce their published shares; it re-runs the same share-of-years comparison on the two sources that survey all countries and reports what they show. Their headline incidence figure also weighted countries by their share of world income; the counting here weights every country-year equally, a different measure, and the two are not interchangeable.

Source: Crisis atlas: banking event-years from crises_events (source ‘laeven_valencia’, the Laeven-Valencia systemic banking crisis database, all IMF members 1970–2017; and source ‘gmd’, the Global Macro Database), the crises_panel country-year frame, and the JST Macrohistory Database for the 18-economy advanced group. The claim under test is Reinhart, C. and K. Rogoff (2013), “Banking Crises: An Equal Opportunity Menace,” Journal of Banking and Finance 37(11), 4557–4573 (first circulated as NBER Working Paper 14587, 2008).

What we run it on

The crisis atlas records banking crises from five chronologies. The comparison here is a share of country-years: banking event-years from one chronology in the numerator, all crises_panel country-years in the window in the denominator, split into the 18 advanced economies and the 218 others, over 1970–2017.

• One chronology at a time. Of the five sources, Laeven-Valencia and the Global Macro Database survey all countries; the Reinhart-Rogoff chronology covers 70 countries in the data, the JST banking flag only the 18 advanced economies, and the ESRB source only European ones, so pooling all five would raise the advanced share by construction, since two of them are advanced-only or Europe-only. The primary test therefore uses Laeven-Valencia alone over 1970–2017, its all-IMF-member window, drawn from crises_events where the source is Laeven-Valencia and the crisis type is banking.
• The advanced group is the 18-economy JST list. The split is JST membership, the same country-level cut the crisis atlas’s patterns page uses for its advanced panel, applied here as a proxy for the paper’s income classification; the 218 others are every remaining economy in the panel window.
• The share weights crisis-years; the onset rate weights episodes. The crisis-year share is the count of panel country-years in which a banking event-year is recorded, over all panel country-years. A long crisis counts every year it runs, so a second table reports the onset rate: a banking event-year is an onset only when no banking event-year falls in the prior 2 calendar years, the atlas’s own episode merge rule, and continuation years are dropped from the at-risk denominator so the rate counts distinct episodes rather than crisis-years.
• Every window is calendar-anchored. The prior-years test is an EXISTS over the calendar range [year − 2, year − 1], not a row offset over the event list, so a gap between two countries’ events cannot be bridged into a spurious continuation. The single pre-declared variant re-runs the same crisis-year comparison on the Global Macro Database source in place of Laeven-Valencia.

Source: Crisis atlas: banking event-years from crises_events (source ‘laeven_valencia’, the Laeven-Valencia systemic banking crisis database, all IMF members 1970–2017; and source ‘gmd’, the Global Macro Database), the crises_panel country-year frame, and the JST Macrohistory Database for the 18-economy advanced group. The claim under test is Reinhart, C. and K. Rogoff (2013), “Banking Crises: An Equal Opportunity Menace,” Journal of Banking and Finance 37(11), 4557–4573 (first circulated as NBER Working Paper 14587, 2008). All numbers on this page are computed at build time from src/lib/equalOpportunityMenace.ts; nothing is typed in.

Layer 1: banking crises on Laeven-Valencia alone

On the full 1970–2017 window the advanced group spends 7.9% (68/864) of country-years in a banking crisis and the rest 4.1% (424/10,322), so on the whole window the advanced share is the higher of the two. But the split by era is the point: before 2007 the two are at parity, with the rest in fact 0.6pp above the advanced group (4.2% (338/7,971) against 3.6% (24/666)), and it is the 2007–2017 decade alone that opens a 18.5pp advanced excess (22.2% (44/198) against 3.7% (86/2,351)).

Source: Crisis atlas: banking event-years from crises_events (source ‘laeven_valencia’, the Laeven-Valencia systemic banking crisis database, all IMF members 1970–2017; and source ‘gmd’, the Global Macro Database), the crises_panel country-year frame, and the JST Macrohistory Database for the 18-economy advanced group. The claim under test is Reinhart, C. and K. Rogoff (2013), “Banking Crises: An Equal Opportunity Menace,” Journal of Banking and Finance 37(11), 4557–4573 (first circulated as NBER Working Paper 14587, 2008). The share is banking crisis country-years over all panel country-years in the era; the two groups are the 18 JST economies and the 218 others.

Layer 1, by episode: the onset rate tells the same story

Counting distinct episodes rather than crisis-years, before 2007 the advanced group starts a banking crisis in 0.9% (6/648) of at-risk country-years against 1.5% (114/7,747) for the rest, so the rest again start crises at least as often. Across 2007–2017 the advanced onset rate is 7.8% (13/167) against 1.1% (25/2,290), about 7 times the rest’s, so the inversion is not an artifact of a few long crises being counted many times: it holds when each episode is counted once.

Source: Crisis atlas: banking event-years from crises_events (source ‘laeven_valencia’, the Laeven-Valencia systemic banking crisis database, all IMF members 1970–2017; and source ‘gmd’, the Global Macro Database), the crises_panel country-year frame, and the JST Macrohistory Database for the 18-economy advanced group. The claim under test is Reinhart, C. and K. Rogoff (2013), “Banking Crises: An Equal Opportunity Menace,” Journal of Banking and Finance 37(11), 4557–4573 (first circulated as NBER Working Paper 14587, 2008). An onset is a Laeven-Valencia banking event-year with no such event-year in the prior 2 calendar years; continuation years are dropped from the at-risk denominator.

The variant: the Global Macro Database source

The single pre-declared variant swaps the chronology for the Global Macro Database and repeats the crisis-year share. It agrees on both halves: before 2007 the two groups are again at parity, this time with the advanced group 0.7pp above the rest (2.4% (16/666) against 1.7% (137/7,971)), the opposite tilt to Laeven-Valencia but the same order, and across 2007–2017 the advanced share again pulls clear, 8.1% (16/198) against 0.8% (18/2,351), a 7.3pp gap on the second all-country chronology.

Source: Crisis atlas: banking event-years from crises_events (source ‘laeven_valencia’, the Laeven-Valencia systemic banking crisis database, all IMF members 1970–2017; and source ‘gmd’, the Global Macro Database), the crises_panel country-year frame, and the JST Macrohistory Database for the 18-economy advanced group. The claim under test is Reinhart, C. and K. Rogoff (2013), “Banking Crises: An Equal Opportunity Menace,” Journal of Banking and Finance 37(11), 4557–4573 (first circulated as NBER Working Paper 14587, 2008). Same design and eras as the Laeven-Valencia table, with GMD banking event-years as the numerator.

Layer 2: what not to do, and why the atlas scores agreement

Running the same comparison on the pooled all-source flag, which sets banking_crisis to one whenever any of the five chronologies records a crisis, the full-window advanced share climbs to 22.1% (191/864) and the rest to 7.0% (724/10,322). That is a coverage artifact, not a finding: the single-source Laeven-Valencia numbers over the same window are 7.9% (68/864) and 4.1% (424/10,322), so pooling adds 14.2pp to the advanced share and 2.9pp to the rest, and it adds more to the advanced group because two of the five pooled chronologies, JST and the ESRB, carry only advanced and European countries and so contribute advanced crisis-years the other economies have no counterpart source to match. Pooling therefore manufactures an advanced-economy excess even over the full window, where the era-by-era single-source test shows parity before 2007; this is why the atlas scores agreement across chronologies rather than taking their union.

Source: Crisis atlas: banking event-years from crises_events (source ‘laeven_valencia’, the Laeven-Valencia systemic banking crisis database, all IMF members 1970–2017; and source ‘gmd’, the Global Macro Database), the crises_panel country-year frame, and the JST Macrohistory Database for the 18-economy advanced group. The claim under test is Reinhart, C. and K. Rogoff (2013), “Banking Crises: An Equal Opportunity Menace,” Journal of Banking and Finance 37(11), 4557–4573 (first circulated as NBER Working Paper 14587, 2008). This row uses crises_panel.banking_crisis, the union of all five chronologies; it is shown as the artifact to avoid, not as the test.

Reading the result

• Before 2007, the groups are at parity, on both all-country chronologies. Laeven-Valencia puts the rest 0.6pp above the advanced group (4.2% (338/7,971) against 3.6% (24/666)) and the Global Macro Database puts the advanced group 0.7pp above the rest (2.4% (16/666) against 1.7% (137/7,971)); the two sources tilt opposite ways by less than a percentage point, which is exactly the near-equality the paper described.
• The crisis decade inverts it. Across 2007–2017 the advanced group spends 22.2% (44/198) of country-years in a banking crisis against 3.7% (86/2,351) for the rest, a 18.5pp gap, and starts episodes about 7 times as often (7.8% (13/167) against 1.1% (25/2,290)); on this atlas the equal-opportunity menace became an advanced-economy menace in that decade, and the variant source agrees.
• The pooled number is the trap. The all-source flag reads 22.1% (191/864) against 7.0% (724/10,322), a gap manufactured by pooling chronologies that cover only rich countries, and it appears even over the full window where the single-source test shows parity before 2007, so it should be read as a warning about coverage, not as evidence on the paper’s claim.

What this cannot tell you

• The advanced split is a proxy. The 18 advanced economies here are JST membership, not the paper’s own high-income versus middle-to-low-income classification, so the two groups are close to but not the same as the paper’s.
• Laeven-Valencia records systemic crises only. The numerator counts systemic banking crises; borderline and non-systemic events the database sets aside are not in the 492 crisis country-years counted over 1970–2017, so the shares are incidence of systemic crises, not of all banking distress.
• The denominator counts every panel country-year. All 10,322 other-economy country-years in the window sit in the denominator, including small and micro states Laeven-Valencia may never have flagged, which lowers the rest’s share relative to a denominator restricted to countries the database actively tracked.
• Shares weight long crises; onset rates do not. The crisis-year share counts every year a crisis runs, so a country with one long crisis outweighs one with several short ones; the onset table is shown beside it for exactly this reason, and the two agree on the 2007–2017 inversion.
• The era boundary is chosen, not pre-registered. The 2006/2007 split is drawn at the global financial crisis, not by the paper, so the two eras are a lens on the same 1970–2017 window and not an out-of-sample test.
• Counting is not their weighted incidence. The paper’s headline measure weighted countries by their share of world income; every share here weights country-years equally, so agreement on the direction of the paper’s claim does not reproduce its magnitude.