Correlation - The Foundation of Diversification
By EC Assets Research Team, Quantitative Foundations · Published · Updated
Correlation: Correlation measures the degree to which two return series move together. It is the foundational input to portfolio construction, ranging from -1 (perfectly opposite) through 0 (independent) to +1 (identical movement).
What Correlation Measures
Correlation measures how closely two return series move together. It is the covariance of the two series divided by the product of their standard deviations, which gives a number between -1 and +1:
$$\rho_{XY} = \frac{\text{Cov}(X, Y)}{\sigma_X,\sigma_Y}$$
At +1 the two always move in the same direction in proportion, at 0 there is no linear relationship, and at -1 they always move in opposite directions in proportion. Because it is scaled by the standard deviations, correlation can be compared across pairs of assets with very different volatilities.
How It Works
Markowitz (1952) put correlation at the centre of portfolio theory. For two assets with weights w and volatilities σ,
$$\sigma_p^2 = w_1^2\sigma_1^2 + w_2^2\sigma_2^2 + 2,w_1 w_2,\sigma_1 \sigma_2,\rho_{12}$$
The correlation term decides how much the two risks reinforce or offset each other. For two assets with 20 percent volatility each, combined 50/50:
| Correlation | Portfolio volatility | Reduction against either asset |
|---|---|---|
| +1.0 | 20.00% | 0.0% |
| +0.5 | 17.32% | 13.4% |
| 0.0 | 14.14% | 29.3% |
| -0.5 | 10.00% | 50.0% |
| -1.0 | 0.00% | 100.0% |
Portfolio volatility equals 20% × √((1 + ρ) / 2).
The benefit is not linear. Moving from +1 to +0.5 cuts volatility by 2.68 points, moving from +0.5 to 0 by 3.18, and every further step down is worth more, which is why genuinely low or negatively correlated assets are so valuable, and so rare. The correlation bench shows the effect for other mixes.
For a portfolio of many assets, the average correlation sets a floor: with equal weights and equal volatilities, portfolio variance approaches the average covariance as the number of assets grows, so diversification can remove only the uncorrelated part of the risk.
Pearson correlation is not the only measure. Rank correlation, such as Spearman's, is less sensitive to outliers, and tail-dependence measures look only at joint extreme moves, which matter most for drawdowns. The estimation window matters too: a one-year window tracks the current regime but is noisy, while a ten-year window is stable but can blend opposite regimes into a misleading average.
Worked Example
A 60/40 portfolio holds equities with 16 percent volatility and 10-year government bonds with 7 percent. Its variance is 0.6² × 16² + 0.4² × 7² + 2 × 0.6 × 0.4 × 16 × 7 × ρ = 100 + 53.76ρ, so the stock-bond correlation alone moves its volatility:
| Stock-bond correlation | Portfolio volatility |
|---|---|
| -0.45 (2010s) | 8.71% |
| 0.00 | 10.00% |
| +0.28 (1990s) | 10.73% |
| +0.60 (stress assumption) | 11.50% |
Correlations labelled by decade are the daily S&P 500 and 10-year Treasury correlations computed for this entry.
Those decade figures come from daily S&P 500 price returns and approximate 10-year Treasury returns built from US Treasury par yields. The correlation was +0.28 in the 1990s, -0.34 in the 2000s, -0.45 in the 2010s and -0.07 from 2020 to October 2026. In 2022, when inflation and rate rises hit both assets, it was +0.17, after -0.48 in 2020. The same allocation carried about 23 percent more risk at the 1990s correlation than at the 2010s one.
When It Applies (and Limitations)
It changes with the regime. The stock-bond correlation has changed sign across decades. Portfolios that assume the recent average will hold through a change in inflation or policy can carry far more risk than planned.
Stress raises measured correlations, partly mechanically. Longin and Solnik (2001) found that international equity correlations rise in bear markets. Forbes and Rigobon (2002) showed that higher volatility alone inflates measured correlation, so part of the apparent rise in crises is statistical rather than a change in underlying linkages. For risk management the distinction matters less: losses arrive together either way.
Linear only. Correlation misses non-linear relationships. Option positions, structured products and strategies with tail exposure can have low measured correlation and still lose together in a crash.
Estimation noise. A matrix for 50 assets has 1,225 distinct correlations, usually estimated from a few years of data. Ledoit and Wolf (2004) show that shrinking the sample matrix toward a structured target gives more reliable portfolios.
Why It Matters for Institutional Investors
Portfolio construction. Correlation assumptions drive optimiser output more than almost any other input, apart from expected returns. Small changes can swing weights substantially, so robust estimates and constraints matter.
Stress testing. Running portfolios under higher, stress-period correlations shows how much diversification is likely to remain in a selloff, a better guide to drawdown risk than long-run averages.
Hedging. The quality of a hedge depends on the correlation between the hedge and the exposure, and a hedge that works in calm markets can fail when correlations shift.
Smoothed returns hide correlation. Private equity, real estate and some hedge fund returns are based on appraisals or stale prices, which smooth them and lower their measured correlation with public markets. Getmansky, Lo and Makarov (2004) show how this serial correlation understates risk; unsmoothing the returns usually reveals much higher correlation with equities.
Correlation as an asset. Implied correlation can be traded directly through dispersion trading, and its level is a gauge of how much systemic risk the market is pricing.
References
- Markowitz, H. (1952). Portfolio Selection. Journal of Finance, 7(1), 77-91.
- Longin, F., & Solnik, B. (2001). Extreme Correlation of International Equity Markets. Journal of Finance, 56(2), 649-676.
- Forbes, K. J., & Rigobon, R. (2002). No Contagion, Only Interdependence: Measuring Stock Market Comovements. Journal of Finance, 57(5), 2223-2261.
- Getmansky, M., Lo, A. W., & Makarov, I. (2004). An Econometric Model of Serial Correlation and Illiquidity in Hedge Fund Returns. Journal of Financial Economics, 74(3), 529-609.
- Ledoit, O., & Wolf, M. (2004). Honey, I Shrunk the Sample Covariance Matrix. Journal of Portfolio Management, 30(4), 110-119.
- Ang, A. (2014). Asset Management: A Systematic Approach to Factor Investing. Oxford University Press.
Frequently asked questions
Why does correlation matter more than individual volatility in a portfolio?
Because it decides how much the assets' risks offset each other. Two assets with 20 percent volatility each, mixed 50/50, give a portfolio volatility of 20 percent at correlation +1, 14.1 percent at zero and nothing at -1. The individual volatilities are the same in each case; only the correlation changes the result.
What does it mean that correlations rise during stress?
In severe selloffs many risk assets fall together, because of forced deleveraging, withdrawal of liquidity and common macro shocks. Longin and Solnik (2001) found that equity market correlations rise in bear markets but not in bull markets. Diversification is therefore weakest exactly when it is needed most.
How is correlation different from covariance?
Covariance measures co-movement in the units of the two series, so its size depends on their volatilities. Correlation divides covariance by the product of the standard deviations, giving a unit-free number between -1 and +1 that can be compared across pairs of assets.
Can correlation be misleading?
Yes. It measures only linear relationships, so assets with option-like payoffs can be strongly related with low correlation. It is also unstable: an estimate from a calm period can understate co-movement in a crisis, and two assets can look uncorrelated while sharing a driver that only matters occasionally.
What is the long-run correlation between stocks and bonds?
It depends on the period. Using daily S&P 500 returns and approximate 10-year Treasury returns, it was +0.28 in the 1990s, -0.34 in the 2000s and -0.45 in the 2010s, and close to zero from 2020 to 2026, with 2022 positive. Inflation and the direction of policy rates appear to drive the sign.
Stay informed
Market commentary, firm news and research from EC Assets - direct to your inbox.