Correlation

Correlation measures how closely two assets move together over time. It is the foundation of diversification: combining assets that are not perfectly correlated produces a portfolio with less risk than the average of its components.

How to read it

Correlation is expressed as a number between -1 and +1.

• +1.0 — perfectly correlated. The two assets move identically. Adding the second one provides no diversification benefit.
• +0.7 to +0.9 — strongly positively correlated. Common between stocks in the same sector. Provides limited diversification.
• +0.3 to +0.6 — moderately correlated. Typical between stocks in different sectors. Meaningful diversification.
• 0 — uncorrelated. The two assets' movements are independent. Strong diversification benefit.
• -1.0 — perfectly negatively correlated. When one rises, the other falls. Rare in practice.

Why it matters for portfolios

Two stocks each with 25% volatility and correlation of +1.0 produce a portfolio with 25% volatility — no benefit. The same two stocks with correlation of +0.3 produce a portfolio with roughly 19% volatility. The lower the correlation, the more risk diversification removes.

This is the math behind why holding twenty stocks in twenty different sectors reduces risk far more than holding twenty stocks in the same sector. The ticker count matters less than the correlation structure underneath.

Correlation in stress

Correlations are not constant. They rise sharply during market crashes — when investors sell indiscriminately, almost everything falls together. A diversified portfolio with average pairwise correlations of 0.3 in calm markets can see those correlations jump to 0.7 or higher during a crisis. The diversification you thought you had partially evaporates exactly when you need it.

Limitations

• Backward-looking. Historical correlations do not always predict future correlations.
• Linear measure. Correlation captures linear relationships only. Two assets can be uncorrelated and still related in non-linear ways (e.g., both crash together only during tail events).
• Window-dependent. 1-year, 5-year, and 20-year correlations can differ substantially. Choosing a window is a modeling decision, not a neutral fact.