Correlation
Correlation measures how closely two investments move together. It's the decisive quantity for diversification, and unfortunately it isn't stable.
A correlation of plus one means two investments always move together. Minus one means always in opposite directions. Zero means there's no relationship.
For diversification, you want values well below one. Two European stock funds often sit above 0.9 with each other, which is barely any diversification. Stocks and safe government bonds historically sit considerably lower.
The uncomfortable property: correlations change. In calm periods, things drift apart. In panic phases, much of everything falls together, because everyone wants to sell at once.
That's why diversification helps least exactly when you need it most. Knowing that changes how you plan: with a real buffer of safe assets, not with hope for offsetting gains.
The Pearson correlation coefficient measures only linear relationships. Nonlinear dependencies, such as those in options strategies or at the tails of a distribution, aren't captured by it. A correlation near zero doesn't rule out strong dependence in the tails of the distribution.
The rise in pairwise correlations during stress periods is well documented empirically. Causes include shared liquidity needs, leveraged positions with forced selling, and a common risk factor that gets masked by idiosyncratic moves during calm periods. For portfolio construction, that means historically estimated correlations systematically overstate the diversification benefit in a crisis.
A methodologically sound approach is therefore to look at correlations separately: once over the full period, and once exclusively over periods of severe market declines. The second estimate is what matters for whether a holding actually helps when it counts. Assets whose diversification benefit only shows up in the first case are unsuited for genuine protection.
Summary
- Correlation only captures linear relationships.
- Correlations rise in crises, and diversification works worse then.
- Check correlations separately for crisis periods, not just on average.
Did you get it?
What does the correlation coefficient fail to capture?
Nonlinear dependencies, especially in the tails of the distribution.
Why do correlations rise in crises?
Through shared liquidity needs, forced selling of leveraged positions, and a dominant common risk factor.
How do you realistically check diversification benefit?
By additionally estimating correlations only over periods of severe decline.
Related
- DiversificationStage 2
- Reading a chartStage 3
- VolatilityStage 4