Claim check: Correlations go to one in a crisis
“In a crisis all correlations go to 1, so diversification fails when you need it.”
What the data says
Partly Partly: the average pairwise correlation among the ten markets was 0.25 while SPY was more than 10% below its peak, against 0.16 otherwise, a rise of +0.09 (-0.01 in the first half, +0.13 in the second). The link is real but weaker than the saying implies.
It holds in some markets or periods, leans the right way without proof, or helps on risk but not on return.
Who says it
Common refrain after 2008, 2020.
Origins are as commonly cited and were not independently verified.
How we tested it
For every pair of the ten markets we compute the correlation of daily returns on common dates, separately for days when SPY is more than 10% below its previous peak (stress, known at the previous close) and for all other days, then average over pairs. A rise of at least 0.15 in both halves of the history counts as correlations rising meaningfully in stress.
Results
| Period | Stress days | Stress | Calm | Difference |
|---|---|---|---|---|
| Full history | 697 | 0.25 | 0.16 | +0.09 |
| First half | 165 | 0.17 | 0.17 | -0.01 |
| Second half | 532 | 0.27 | 0.14 | +0.13 |
Average pairwise correlation: stress vs calm
Stability across the two halves of history
A saying that only works in one half of the history is not much of a rule. Each row shows the effect in the full history, the first half and the second half; green means the claimed effect was present.
| Market | Full history | First half | Second half |
|---|---|---|---|
| All markets | +0.09 | -0.01 | +0.13 |
Context and cross-checks
Shown for context; not part of the verdict.
The site's own strategies
| Strategy | Market | Sharpe, 5 years | Benchmark, 5 years | Sharpe, full | Benchmark, full | CAGR, full | Benchmark, full | Max DD, full | Benchmark, full | Reading |
|---|---|---|---|---|---|---|---|---|---|---|
| PORT-05 Hierarchical Risk Parity | HRP | 0.95 | 1.11 | 0.14 | 0.47 | +0.5% | +5.3% | -81% | -38% | No effect |
Sharpe ratio minus benchmark, full history · Benchmark = the buy-and-hold shown on the strategy pages.
The honest caveat
Crisis windows overlap, so the number of independent crises is only a few.
Descriptive statistics from historical data, before costs. Not a forecast and not investment advice.
Questions
In a crisis all correlations go to 1, so diversification fails when you need it.
Partly: the average pairwise correlation among the ten markets was 0.25 while SPY was more than 10% below its peak, against 0.16 otherwise, a rise of +0.09 (-0.01 in the first half, +0.13 in the second). The link is real but weaker than the saying implies.
How was this claim tested?
For every pair of the ten markets we compute the correlation of daily returns on common dates, separately for days when SPY is more than 10% below its previous peak (stress, known at the previous close) and for all other days, then average over pairs. A rise of at least 0.15 in both halves of the history counts as correlations rising meaningfully in stress.
Related strategies on Quant Arena
Markets used
S&P 500 · Gold · Silver · Oil · Natural gas · Commodities · Agriculture · EUR/USD · Bitcoin · Ether