What exactly does the Pearson correlation coefficient measure and when does it fail?

Question

AcadiFi · Accepted Answer

The Pearson correlation coefficient ρ is defined as Cov(X,Y) / (σₓσᵧ). It measures the strength and direction of the linear relationship between two continuous variables. It ranges from -1 (perfect negative linear) to +1 (perfect positive linear), with 0 meaning no linear relationship.

Pearson works well when: (1) both variables are roughly continuous, (2) the relationship is approximately linear, (3) outliers are limited, and (4) both variables are normally distributed.

It fails when:

1. Relationships are nonlinear. Two variables related by Y = X² might show near-zero Pearson correlation even though they are perfectly dependent.
2. Outliers dominate. A single extreme observation can push correlation from 0.1 to 0.7 or back.
3. Data are ordinal or ranked. Pearson assumes interval-scale data.
4. Distributions are heavy-tailed. Financial returns are notoriously leptokurtic, which inflates estimation error.

Example: Returns of Pembroke Biotech during a 2-year stable period show ρ = 0.15 with the S&P 500. After a single crisis month where biotech fell 30% alongside a 15% market drop, recomputed ρ jumps to 0.55. One observation changed the interpretation dramatically.

For CFA candidates, alternatives include Spearman rank correlation (non-parametric, robust to outliers), Kendall's tau, and distance correlation for nonlinear dependence. Always plot scatterplots before trusting a correlation number — the Anscombe quartet famously shows four datasets with identical ρ but completely different structures.

What exactly does the Pearson correlation coefficient measure and when does it fail?

Master Level I with our CFA Course

Related Questions

Practice Questions