Hvass Laboratories
Blog on Investment Finance

All this was made by a single person without any grant-money.
If you find it useful then please donate securely using PayPal.
Even a few dollars are appreciated. Thanks!


Investing
Books
Services
Blog
S&P 500

Video Talks
Investing
Deep Learning
Optimization

Source Code
GitHub
SwarmOps
RandomOps
ArrayOps
NeuralOps

Author
Publications
Thesis
Citations
Schoolwork
Contact

Hvass Labs

Portfolio Variance is Not Risk

by Magnus Erik Hvass Pedersen, May 17, 2014

Mean-variance portfolios are commonly believed to minimize risk for a given level of expected return, because variance is believed to measure risk. But this is incorrect as proven with a short example.


Negative Returns

Let asset A be a stochastic variable with negative returns (4%), (5%) or (6%) and let asset B be a stochastic variable with positive returns 5%, 10% or 15%. The returns have equal probability of occurring and are dependent in the order they are listed so that if asset A has return (4%) then asset B has return 5%, etc. The asset returns are perfectly anti-correlated with coefficient -1.

The long-only, minimum-variance portfolio lies on the efficient frontier and is when the weight for asset A is 5/6 and the weight for asset B is 1/6. This gives a portfolio with mean (2.5%) and zero variance, that is, all possible returns of the portfolio are losses of exactly (2.5%). But an investor could instead have chosen a portfolio consisting entirely of asset B which would always give a positive return of either 5%, 10% or 15%. An investment entirely in asset B is clearly superior to an investment in the minimum-variance portfolio even though asset B has higher variance.

Positive Returns

The above example had an asset with negative returns which could be avoided by adding the constraint that returns must be positive; but the problem also exists for assets with partly negative returns or all positive returns. To see this, change asset A's possible returns to 3%, 2% or 1%. Then the minimum variance portfolio still has asset A weight 5/6 and asset B weight 1/6 which gives a portfolio return of about 3.3% with zero variance. But asset B alone would give a higher return of either 5%, 10% or 15%. So although the minimum-variance portfolio has no return spread, it has a lower return with certainty.

Overlapping Return Distributions

The problem also exists for assets that have overlapping return distributions. Let asset A's possible returns be 6%, 5% or 4%. Then the minimum variance portfolio still has asset A weight 5/6 and asset B weight 1/6 which always gives a portfolio return of about 5.8% with zero variance. But asset B alone would give a higher return of either 10% or 15% with probability 2/3 (or about 67%) and a slightly lower return of 5% with probability 1/3 (or about 33%).

Variance is Not Risk

The reason asset A is included in the efficient frontier and minimum-variance portfolio is that its return has a low (sample) standard deviation of 1% while asset B has a higher standard deviation of 5%. The two assets have negative correlation so combining them in a portfolio lowers the combined standard deviation. The mean-variance efficient frontier is optimized for low variance (and standard deviation) which gives a low spread of the possible returns from the portfolio. But the spread of possible returns is not a useful measure of risk because it does not consider the probability of loss and the probability of other assets having a higher return. So the mean-variance portfolio is not optimized for risk in the traditional sense of the word which is defined in a dictionary as "the chance of injury or loss".

Further Reading

This example is taken from the paper Portfolio Optimization and Monte Carlo Simulation.
It is also explained in this video.


Hvass Laboratories