A minimum variance portfolio is the combination of a given set of risky assets that has the lowest possible variance of returns, and therefore the lowest standard deviation or volatility. It is found using only the assets’ volatilities and the covariances between them. Unlike most optimised portfolios, it does not require any forecast of expected returns.
The concept comes from the mean-variance framework set out by Harry Markowitz in his 1952 paper “Portfolio Selection”, the foundation of modern portfolio theory. When every possible portfolio of the assets is plotted by risk and expected return, the minimum variance portfolio is the left-most point of the curve, at the start of the efficient frontier. It is sometimes called the global minimum variance portfolio to distinguish it from the lowest-risk portfolio for a particular target return.
Why diversification lowers variance
The variance of a portfolio depends not only on the variance of each asset but also on how the assets move together. When two assets are less than perfectly correlated, losses on one are partly offset by gains on the other, so a mix can be less volatile than either asset alone. The minimum variance portfolio pushes this effect of diversification as far as the data allow.
Formula and example
For two assets, the weight of asset 1 in the minimum variance portfolio is:
- and are the volatilities of assets 1 and 2.
- is their covariance, equal to , where is the correlation.
- The weight of asset 2 is .
The portfolio variance is then:
Suppose asset 1 has a volatility of 20 per cent, asset 2 a volatility of 10 per cent, and the correlation is 0.2. The covariance is . The weight of asset 1 is , so the portfolio holds about 14.3 per cent in asset 1 and 85.7 per cent in asset 2. The variance is . The square root gives a volatility of about 9.6 per cent, lower than the 10 per cent of the safer asset on its own.
With many assets, the weights are found from the covariance matrix as , where is a vector of ones. In practice most investors add constraints, such as no short positions and maximum weights per asset, and solve the problem numerically through portfolio optimisation software.
Minimum volatility investing
From the 2000s, index providers and asset managers built equity products based on minimum variance optimisation, usually called minimum volatility or low volatility strategies. Index providers publish minimum volatility versions of broad indices such as the MSCI World Index, and many exchange-traded funds track them. These indices apply constraints on sector, country and single-stock weights so that the portfolio does not become too concentrated.
Interest in the approach was supported by research on the low-volatility anomaly, the observation that lower-risk stocks have historically delivered returns similar to or higher than higher-risk stocks, contrary to the simple prediction of the capital asset pricing model. A related strategy is betting against beta. Minimum volatility portfolios have tended to fall less than the market in sell-offs but to lag in strong rallies, particularly those led by fast-growing, high-volatility sectors such as technology. Investors therefore often use them to reduce the risk of an equity allocation rather than to replace it, and judge them over a full market cycle rather than over a single year.
Limitations
The minimum variance portfolio depends entirely on the estimated covariance matrix. Estimates based on historical data contain errors, and the optimiser tends to place large weights on assets whose risk happens to have been underestimated. Analysts reduce the problem with techniques such as shrinkage of the covariance matrix, factor models and weight limits.
The result is often concentrated in defensive sectors, such as utilities and consumer staples, and in bond-like stocks that are sensitive to interest rates. That concentration can cause losses when rates rise. Because the portfolio ignores expected returns, it is not necessarily the best risk-adjusted choice: the portfolio with the highest Sharpe ratio usually sits further up the frontier. Correlations can also change sharply in a crisis, weakening the diversification on which the portfolio relies. Alternatives that use risk estimates without full optimisation include risk parity and the maximum diversification portfolio.