Know these first
By the end you will be able to
- 1Measure dependence three ways (Pearson, Spearman, Kendall's τ) and say what each sees and misses.
- 2Compare correlations across calm and stress windows with their uncertainty, and spot when one day drives a Pearson figure.
- 3Measure tail dependence on real data and explain why correlation alone cannot describe joint extremes.
- 4Diagnose a correlation matrix that a stress overlay has broken, and repair it by clipping or with Higham's nearest correlation matrix.
Why it matters
Over twenty years of daily fixings, GBP/USD and EUR/USD have a correlation of 0.645. In September and October 2022, when gilts sold off and the Bank of England stepped in, it was 0.855. In the spring of 2020 it was 0.542. L2.5.03: Correlation and diversification warned that correlations change and can rise in a crisis; L2.6.05: Diversification and its limits that assets fall together in the tails in ways one number misses. This lesson measures both on real data, asks how much of each change is noise, and fixes the matrices that stress testing breaks.