Correlation from Covariance

Covariance tells you two assets move together, but its size is meaningless on its own — it scales with the volatilities of both assets. Correlation fixes that by dividing out both standard deviations, compressing co-movement onto a universal $[-1, +1]$ scale.

$\rho = +1$ means the assets are the same bet in different sizes; $\rho = -1$ a perfect hedge; $\rho = 0$ no linear relationship. Everything in diversification — why portfolio risk is less than the sum of its parts — flows through this one number.

Standardize each variable by its own dispersion, then measure co-movement of the standardized versions:

The Cauchy–Schwarz inequality guarantees $|Cov_{AB}| \le \sigma_A \sigma_B$, which is why $\rho$ can never escape $[-1, 1]$.

Converting a covariance into a comparable co-movement number, or reasoning about diversification between two assets.

Correlation captures linear co-movement only — assets can be perfectly dependent through their tails and still show low ρ. Worse, correlations are unstable and lurch toward +1 exactly in crises, when diversification is needed most (2008's core lesson). And it says nothing about causation.

Francis Galton discovered correlation in the 1880s studying heredity — why tall parents have tall (but less extreme) children — and Karl Pearson formalized the coefficient in 1896. Markowitz (1952) made it the engine of diversification: portfolio risk depends less on how risky assets are than on how they move together.

Today correlation matrices sit inside every risk model, pairs trade and multi-asset allocation — and its great modern lesson is its instability: in the 2008 crisis, correlations that models assumed were low all lurched toward one at once.