Quant Analytics: from scratch to pro · Portfolio, sizing and risk metrics · Advanced
Portfolio maths: variance, correlation and the limits of diversification
Portfolio variance depends on every pair of assets, not just on each one alone. The formula shows how much diversification can remove, why the benefit flattens as assets are added, and why it shrinks when correlations rise in a sell-off.
In this lesson
- Compute the volatility of a two-asset portfolio from weights, volatilities and correlation
- Compute the volatility of an equal-weight portfolio of n assets with average correlation
- State the floor that average correlation puts under diversification
- Explain why correlation estimates from calm periods overstate protection in a crisis
Read first: Correlation and its limits, Diversification: not keeping everything in one place, Volatility explained
Diversification is the one free improvement in investing: combine assets that do not move together and risk falls faster than return does. The portfolio maths shows how much, and where the free lunch runs out.
Two assets
For weights w1 and w2, volatilities s1 and s2 and correlation rho:
variance_p = w1^2 s1^2 + w2^2 s2^2 + 2 w1 w2 rho s1 s2
The last term holds the covariance (rho s1 s2). It is the only place where the relationship between the assets enters. At rho = 1 the formula collapses to sigma_p = w1 s1 + w2 s2, the weighted average of the volatilities. Anything lower gives a portfolio less volatile than that average. See covariance and correlation.
Worked example
Asset 1 has volatility 20% and weight 60%. Asset 2 has volatility 12% and weight 40%.
- At correlation 1: volatility = 0.6 x 20% + 0.4 x 12% = 16.8%.
- At correlation 0.3: variance = 0.0144 + 0.002304 + 0.003456 = 0.02016, so volatility is 14.2%.
- At correlation 0: volatility is 12.9%.
- At correlation -0.5: volatility is 10.5%.
Going from perfect correlation to 0.3 removes 2.6 percentage points of volatility with no change in the holdings. This is the effect described in why diversification works and first formalised by Markowitz in 1952 (history).
Many assets
Hold n assets in equal weights, each with volatility s, and let rho be the average pairwise correlation. Then:
sigma_p = s x sqrt(1/n + (1 - 1/n) x rho)
Two terms drive it. The first, 1/n, is each asset's own variance, which shrinks as n grows. The second is the shared part, rho, which does not.
With s = 20% and rho = 0.3: one asset gives 20%, ten assets give 12.2%, fifty give 11.2%. As n grows without limit, volatility approaches s x sqrt(rho), which is 10.95%. Adding assets beyond a couple of dozen removes little: the chart flattens because the floor is set by correlation, not by count.
Correlations in a crisis
The floor assumes rho stays put. It does not. Longin and Solnik (2001) studied monthly returns of the five largest stock markets from 1959 to 1996 and found correlation rising in bear markets, not in bull markets. If the average correlation moves from 0.3 to 0.8, the ten-asset portfolio has volatility of 18.1% instead of 12.2%, almost the volatility of a single asset. Protection is weakest at the moment it is needed. Estimates from a calm window are optimistic, and a stress test with higher correlations belongs next to the base case. See the 2008 crisis and tail risk.
Common mistakes
- Counting assets instead of independent sources of risk. Fifty stocks in one sector are closer to a few.
- Using one correlation for all conditions. Estimate it in stress windows too.
- Taking a correlation computed from prices instead of returns (it is inflated by shared trends).
- Forgetting that correlation measures linear co-movement and says little about joint extreme losses.
- Reading a precise correlation from a short sample. With 60 monthly returns the standard error of a correlation near zero is about 0.13.
In practice
Check diversification by looking at the average correlation and at the portfolio's volatility against the weighted average of its parts. Then repeat with correlations raised towards 0.8 to see the loss of protection. Analytics on FinUties applies measures like these to the markets it covers, so they are at hand when you read a chart.
Portfolio volatility against the number of assets
- correlation 0
- correlation 0.3
- correlation 0.6
Worked numbers
Each result is recomputed from the formula shown.
| What | Formula | Result |
|---|---|---|
| 60/40 volatility at correlation 1 | 0.6 * 0.2 + 0.4 * 0.12 | 16.8% |
| 60/40 variance at correlation 0.3 | 0.6**2 * 0.2**2 + 0.4**2 * 0.12**2 + 2 * 0.6 * 0.4 * 0.3 * 0.2 * 0.12 | 0.02016 |
| 60/40 volatility at correlation 0.3 | sqrt(var_rho3) | 14.2% |
| 60/40 volatility at correlation 0 | sqrt(0.6**2 * 0.2**2 + 0.4**2 * 0.12**2) | 12.9% |
| 60/40 volatility at correlation -0.5 | sqrt(0.6**2 * 0.2**2 + 0.4**2 * 0.12**2 + 2 * 0.6 * 0.4 * (-0.5) * 0.2 * 0.12) | 10.5% |
| 10 assets, volatility 20%, correlation 0.3 | 0.2 * sqrt(1/10 + (1 - 1/10) * 0.3) | 12.2% |
| 50 assets, same inputs | 0.2 * sqrt(1/50 + (1 - 1/50) * 0.3) | 11.2% |
| limit as n grows | 0.2 * sqrt(0.3) | 10.95% |
| 10 assets at correlation 0.8 | 0.2 * sqrt(1/10 + (1 - 1/10) * 0.8) | 18.1% |
Check yourself
Two assets each have 20% volatility and a correlation of 1. What is the volatility of a 50/50 mix?
20%. With perfect correlation there is no diversification: sigma_p = 0.5 x 20% + 0.5 x 20%.
Two uncorrelated assets, each with 20% volatility, are held 50/50. What is the portfolio volatility?
sqrt(0.25 x 0.04 + 0.25 x 0.04) = sqrt(0.02) = 14.1%.
With 10 equal-weight assets of 20% volatility and average correlation of 0.5, what is the volatility, and what is the limit as assets are added?
20% x sqrt(0.1 + 0.9 x 0.5) = 20% x sqrt(0.55) = 14.8%. The limit is 20% x sqrt(0.5) = 14.1%.
Why does adding the 51st asset help much less than adding the 2nd?
The term 1/n shrinks quickly, leaving the average-covariance term, which does not fall with n. The remaining risk is the part the assets share.
A portfolio was built using a correlation of 0.3 estimated in calm years. What goes wrong in a sell-off?
Correlations between equity markets tend to be higher in falling markets, so realised volatility is higher than the formula with 0.3 gave.
Key terms
Sources
- Portfolio Selection. Journal of Finance, vol. 7, no. 1, pp. 77-91 (Harry Markowitz), 1952
- Extreme Correlation of International Equity Markets. The Journal of Finance 56(2), 649-676 (Francois Longin, Bruno Solnik), 2001. Accessed 9 Nov 2026