← Daybreak
150 YEARS OF CRISES, HONESTLY · A DATA-DRIVEN OVERVIEW

What comes before a banking crisis?

0 countries·0 years·0 crises

→ One warning sign shows up again and again: a credit boom.

Across 150 years and 18 advanced economies, the single most reliable thing that comes before a banking crisis is a rapid run-up in private credit. After the hottest credit booms, a crisis follows within two years about one time in six — against a one-in-sixteen base rate. It shifts the odds; it is not an alarm clock.

Built from the Jordà-Schularick-Taylor Macrohistory database (18 advanced economies, annual, 1870–2020) and a model trained on 17 countries and tested on the country it never saw. We sell no predictions. The job is to separate what 150 years of data actually support from the headlines. Observation, not advice.

THE WHOLE PICTURE

A century and a half of banking crises.

Since 1870 the 18 advanced economies in this study have lived through 88 systemic banking crises — the kind where banks fail in numbers, lending seizes up, and governments step in. That is the full set we will try to explain.

They do not arrive evenly. They come in waves: the 1870s, the run into the 1907 panic, the 1929–31 cluster, and 2008 each light up; the 1950s and 1960s are almost empty. A crisis is a recurring, clustered feature of advanced economies — not a freak one-off — which is exactly why a century of them can teach us what tends to come first.

overview

loading overview…
Each bar counts the banking crises that began in that decade across all 18 countries. The clusters — 1870s, 1907, 1929–31, 2008 — are the waves; the long mid-century calm is the gap between them. The rest of the report asks what those waves had in common.
PART 1 — THE ONE WARNING SIGN

The clearest warning is a credit boom.

Of everything you can measure about an economy — growth, inflation, house prices, the stock market, government debt — one thing stands out before banking crises again and again: how fast private credit (what households and companies owe to banks) has been growing relative to the size of the economy.

Sort every country-year in 150 years of history into ten equal groups by how hot that credit growth ran, and read how often a crisis followed within two years. The result is not a gentle slope — it is flat, and then it jumps.

The credit-boom signal, in three numbers
0.0%Chance of a crisis within 2 years after the hottest credit booms (the top tenth)
0.0%Chance in a typical year — the base rate to beat
0.0×More likely after the hottest booms than after the calmest credit (16.6% vs 2.6%)
0%Of all crises were preceded by a credit boom

credit_boom_decile

loading credit_boom_decile…
The bottom eight groups all hug the 6.4% line of a typical year. Then the ninth jumps to 11.8% and the hottest tenth to 16.6% — about one crisis in six. The danger is concentrated in genuine booms, not spread smoothly across "above-average" credit.

precrisis_credit

loading precrisis_credit…
Looked at the other way round: in the year before an actual crisis, credit had typically run up by about +6.6 points of GDP over three years — roughly three times the +2.1-point pace of an ordinary stretch. The boom is the common thread through the crises, not a coincidence around them.
PART 2 — HOUSING ONLY MATTERS WHEN LEVERAGED

A housing boom is only dangerous when credit fuels it.

"Housing bubble" is the headline that usually gets the blame for a crash. But house prices and credit do not always rise together — and the data is blunt about which one matters. Split every country-year by whether credit was booming and whether house prices were booming, and read the crisis rate in each of the four corners.

Only one corner is dangerous: when credit and housing boom together. A housing boom on its own is actually below average risk.

twin_boom

loading twin_boom…
Credit booming and housing booming together: a 16.4% crisis rate within two years — about two and a half times a normal year. Housing booming alone: just 3.8%, below the 6.4% base rate. A housing mania becomes a banking crisis only when it is financed with a credit boom. Credit is the necessary ingredient.
PART 3 — THE WARNING TRAVELS

The same warning works in countries it never trained on.

A pattern that only fits the history it was built on is worthless. The real test is whether the credit-boom warning generalises: train the model on 17 countries, then ask it to rank the crisis years of the one country it has never seen. Do that for each country in turn.

It travels. 17 of the 18 countries come out better than chance, with a typical accuracy around 0.75 (where 0.50 is a coin flip and 1.00 is perfect). The warning is reading something real about how credit booms turn into crises — not memorising one nation’s past.

country_accuracy

loading country_accuracy…
Each bar is one country, scored only on data the model never trained on. 17 of 18 land above the chance line. The lone failure, Belgium, has its crises bunched in the distorted world-war years — the exception that names the rule.

country_trajectories

loading country_trajectories…
Each panel is one major economy’s private credit relative to its economy, 1870–2019, with amber dots on the years a banking crisis began. The level differs by country and era, but the run-up-then-break rhythm is shared — which is exactly why a warning learned on one set of countries works on the rest.
SYNTHESIS — WHERE THE RISK SITS NOW

What the signal says at the end of the data.

So where do the major economies stand? The honest answer reads the changein credit, not its level — a high but stable credit stock is far less dangerous than a fast-rising one. We anchor on 2019, the last pre-pandemic year, because 2020’s reading is distorted by the COVID GDP collapse rather than any real lending boom.

current_risk

loading current_risk…
At the data’s pre-COVID end, France stood out — a credit run-up that put it in an elevated bucket (about 11.8% two-year crisis odds, roughly twice a normal year). The United States was benign, having deleveraged for a decade after 2008; Italy, Spain and the Netherlands were actively shrinking credit.
See all 7 charts in the appendix →