Crisis atlas · back to overview
What the data looks like before a banking crisis
An event study of the run-up to systemic banking crises, in two separate panels. Panel A follows 84 crises in 18 advanced economies with the deep Jordà-Schularick-Taylor macrohistory. Panel B follows 142 crises in 104 mostly emerging and developing economies with the Laeven-Valencia chronology and the Global Macro Database, the breadth Panel A structurally lacks. Each variable plots the median trajectory over the eleven years centred on the onset, with a shaded interquartile band and a flat tranquil-period benchmark. This describes what preceded past crises; it is not an early-warning signal, a forecast, or a probability for any country today.
The two panels are not directly comparable. They use different chronologies (JST vs Laeven-Valencia), different macro panels (JST vs GMD), different country universes (18 advanced economies vs 104 mostly emerging and developing), different eras, and partly different variables (Panel A: credit and house prices; Panel B: inflation and the exchange rate, which GMD supports and the credit story it does not). A level difference between a Panel A line and a Panel B line is as likely to reflect the panel, the chronology, or the era as any real advanced-vs-emerging difference. Read them as complementary windows, not a matched comparison.
Panel A · Advanced economies · 1870–2020 · JST
Median paths around the crisis onset, advanced economies
Each panel is on its own y-scale. The onset year is t0. Onsets are the first year of each systemic banking-crisis episode in the Jordà-Schularick-Taylor chronology, merged and de-overlapped so no recovery is counted twice. Medians (not means) are used because the panel spans wars, hyperinflations, and gold-standard breaks.
Source: Jordà-Schularick-Taylor Macrohistory R6 | BIS credit-to-GDP gap (Data Portal) Methodology
What the advanced-economy panels say
- Credit accelerates, then contracts. The median credit-to-GDP change peaks at +2.13pp at t-2, well above the +0.64pp tranquil pace, before turning negative after the crisis.
- The BIS gap runs hot. The credit-to-GDP gap sits at a +13.51pp median at t-1, above the +10pp Basel upper buffer threshold and far above the -1.02pp tranquil level, on a thinner 24-episode sample.
- House prices peak early. Real house-price growth crests near +5.92% at t-2 (tranquil +1.97%) and turns negative from the onset.
- Growth holds, then breaks. Real GDP growth stays near its tranquil +2.21% until it collapses to +0.33% at t0 and -0.13% at t+1.
Panel B · Emerging & developing economies · 1976–2023 · LV + GMD
Median paths around the crisis onset, emerging and developing economies
The same event-study machinery on a broad country set: onsets are the first year of each systemic banking-crisis episode in the Laeven-Valencia 2026 chronology, restricted to the 104 economies outside Panel A’s 18 (so no crisis is counted twice), then merged and de-overlapped identically. Macro variables come from the Global Macro Database, cut at 2024 to exclude its merged forecasts. GMD carries no credit and thin emerging-market house-price data, so the credit and house-price lines of Panel A are replaced by inflation and two exchange-rate views; the BIS gap remains a thin, small-sample overlay (18 episodes).
Source: Global Macro Database (release 2026_06) | Laeven-Valencia 2026 (IMF WP/26/94) | BIS credit-to-GDP gap (Data Portal) Methodology
What the emerging-market panels say
- A milder, wider growth dip. Real GDP growth holds near its tranquil +3.84% until it slides to +2.17% at t0 and a +1.43% trough at t1, a shallower median dip than Panel A but over a far wider interquartile band.
- Inflation runs hot. CPI inflation sits near a +13.10% median at t0, an order of magnitude above the +3.85% tranquil level and hot throughout the window.
- The current account stays in deficit. It widens to a -4.18% median trough at t0, against a -2.97% tranquil balance.
- Overvaluation, then collapse. The real exchange rate appreciates into the crisis (REER +2.98% median at t-1, tranquil +0.32%), then gives way to nominal devaluation (median depreciation +11.76% at t0 and +14.67% at t+1, tranquil +0.00%): the classic twin-crisis signature.
- The thin credit overlay still spikes. For the 18 larger emerging markets with BIS quarterly credit data, the credit-to-GDP gap still peaks at +11.19pp at t0 from a +0.44pp tranquil level, echoing Panel A’s credit signal where the data exist.
How to read this, and how not to
- Not a prediction. Both panels are unconditional median paths around crises that did happen, not a model. A country can sit inside a shaded band for years without a crisis. Nothing here is an early-warning signal or a probability for any specific country.
- Panel B is not EM-only. It is the complement of the 18 JST countries, tilted to middle- and low-income economies but including some non-JST high-income onsets (Korea, Greece, Iceland, and the transition economies), so it is broad coverage rather than a pure emerging-market panel.
- No credit line in Panel B. GMD carries neither credit nor broad emerging-market house-price data, so Panel B cannot show the credit boom or the property cycle for most of its countries; the credit dimension survives only as the thin BIS overlay, and real house-price growth is dropped entirely.
- Annual granularity. Onsets are dated to the year and variables are annual in both panels, so sub-year lead-lag structure is invisible.
- Thin cells shown. Per-panel episode counts are printed on every chart, and the BIS-gap overlays in both panels rest on small, recent-crisis-heavy samples that should not be over-read.
Full construction, source chronologies, exclusion and merge rules, and the recompute targets for both panels are in the crisis methodology.
Panel A: Jordà, Schularick and Taylor (2017), Macrohistory Database Release 6 (macrohistory.net, consulted 2026-07-09), free for non-commercial use with citation and consultation date. Panel B: Global Macro Database (Müller, Xu, Lehbib and Chen 2025, NBER WP 33714, release 2026_06, globalmacrodata.com, consulted 2026-07-09), free for academic and non-profit research only, citation required; onsets from Laeven and Valencia (2026), “Systemic Banking Crises Database: 1970-2025”, IMF Working Paper WP/26/94. Both panels use the Bank for International Settlements credit-to-GDP gap (BIS Data Portal). Shown for non-commercial research use only.
Where gaps stand today · 2025-Q4 · BIS
Where gaps stand today
The event study above described the run-up to crises that happened. This closing section places the current BIS credit-to-GDP gap for every reporting country on the same axis as that benchmark, and reports how often an elevated gap was actually followed by a banking crisis. It is not a forecast and names no country as at risk; the readings are facts and the historical base rate is shown beside them.
Latest credit-to-GDP gap, every reporting country
Each bar is a country’s most recent quarterly reading (all 43 currently at 2025-Q4), sorted from highest to lowest. The dashed lines mark the Basel activation (+2pp) and elevated (+10pp) thresholds and the +13.51pp median the gap reached at t−1 before past advanced-economy crises. Not one country reaches the elevated line today.
Source: BIS credit-to-GDP gap (Data Portal) | Jordà-Schularick-Taylor Macrohistory R6 | Laeven-Valencia 2026 (IMF WP/26/94) Methodology
These are current values of a backward-looking indicator, not a forecast.
The bars show where the credit-to-GDP gap sits now, drawn against how past crises looked. An elevated gap was followed by a banking crisis in about one in five past episodes (21.8%); it is not an early-warning signal, a ranking of danger, or a probability for any country here. No country on this page is named as at risk or vulnerable. Today, 0 of 43 reporting countries sit at or above the +10pp elevated line; the 4 in the buffer-activation band are Saudi Arabia (+6.99), Japan (+6.78), Argentina (+4.96) and Israel (+3.60).
How to read the current values
- The latest gap is the least reliable point. The credit-to-GDP gap is a one-sided HP filter of the data to date, so the trend is weakest at the sample end and the most recent gap (the one shown) is the most revision-prone. BIS revises the series as new quarters arrive.
- The filter itself is contested. Hamilton (2018) shows the HP filter induces spurious dynamics and behaves badly at endpoints; the standard credit-gap smoothing (λ = 400,000) is a convention, not an estimate. The gap is reported because it is the Basel buffer guide, not because the filter is above criticism.
- Extreme negatives are GDP artifacts. The gap is credit over GDP, so a distorted GDP distorts the gap. Ireland’s -82.09pp reading is the clearest case: Irish GDP is inflated by multinational balance-sheet relocation (“leprechaun economics”), which mechanically deflates credit/GDP. The deep negatives are national-accounts artifacts, not deleveraging.
- Chronology endpoints are conservative. The JST chronology stops flagging advanced-economy onsets after 2008 (coded to 2020) and Laeven-Valencia runs through 2023, so the most recent readings are right-censored out of the base rate rather than counted as “not followed”. If anything this understates the followed rate for post-2008 Europe.
- The base rate is unconditional. The 21.8% pools advanced and emerging economies across the whole sample and conditions on nothing else, not the level above +10pp, not the era. It is a historical frequency, not a model output and not a country-specific probability.
BIS credit-to-GDP gap: Bank for International Settlements, Data Portal (credit-to-GDP gaps, private non-financial sector). Banking-crisis onsets used for the base rate: JST Macrohistory Database Release 6 (Jordà, Schularick and Taylor 2017, macrohistory.net, consulted 2026-07-09) for the 18 advanced economies; Laeven and Valencia (2026), “Systemic Banking Crises Database: 1970-2025”, IMF Working Paper WP/26/94, for the rest. Non-commercial research use only, consistent with the rest of the crisis layer. Full construction and recompute targets are in the crisis methodology.