Covariance
Covariance measures how two random variables move together: Cov(X,Y) = E[(X−μ_x)(Y−μ_y)]. It is the off-diagonal that makes a book more than a list of variances.
Definition
The covariance matrix Σ is positive semidefinite if it is a real second-moment matrix. Correlation is covariance scaled by the two standard deviations. Diversification is a statement about off-diagonals, not about owning many tickers.
Why it matters
Risk parity, PCA, and beta are covariance objects. When Σ breaks (crisis correlations → 1), the optimization was fiction.
Case
A 60/40 book’s variance is 0.6²σ_e² + 0.4²σ_b² + 2·0.6·0.4·Cov. In 2022 Cov flipped sign versus the 2010s — the formula did not change, the number did.
How to read it
Estimate horizon and whether you shrink. A sample Cov from 60 days on 400 names is garbage without structure.
Key papers
Landmark references with real publisher or arXiv links — not token-overlap noise.
- Honey, I Shrunk the Sample Covariance MatrixLedoit, Wolf · 2004 · Journal of Portfolio ManagementShrinkage for usable Σ.