When VaR and Expected Shortfall Share the Mean–Variance Frontier
Summary
The document asks whether portfolios minimizing Value at Risk or expected shortfall must lie on the mean–variance efficient frontier. It explains that when asset returns follow an elliptical distribution, including Gaussian and Student t examples, minimizing these risk measures is equivalent to minimizing variance under the stated conditions. In that setting, the risk measures produce the same frontier and minimum-risk portfolio as variance.
Outside the elliptical case, the optimizers and frontiers can differ. The discussion also distinguishes a plot using standard deviation on the risk axis from plots that place VaR or expected shortfall on that axis; those diagrams need not look alike. One answer argues that the displayed graph may be misleading if it presents separate minimum-risk points while using standard deviation as risk, particularly when the source says it does not assume elliptical returns. The conclusion is conditional on distributional assumptions and does not provide empirical portfolio results.
Key ideas
- Under elliptical return distributions, minimizing VaR or expected shortfall can reduce to minimizing variance.
- In that setting, the mean-risk frontiers coincide with the mean–variance frontier.
- For non-elliptical distributions, VaR and expected-shortfall portfolios can differ from mean–variance portfolios.
- A frontier’s appearance depends on which risk measure is plotted on its horizontal axis.
- Claims about shared frontiers require explicit attention to the distributional assumptions.
Tags
Full text
# Do the minimum VaR and minimum ES portfolios lie on the mean-variance efficient frontier? # Do the minimum VaR and minimum ES portfolios lie on the mean-variance efficient frontier? The mean-variance efficient frontier holds the minimum variance portfolio, but in the graph above it shows that the minimum VaR (Value-at-Risk) and minimum ES (CVaR) portfolios (expected shortfall/conditional VaR) lie on and share the same frontier as the minimum variance portfolio. I thought though (and have seen in articles) that there are frontiers unique to the mean-VaR and mean-ES efficient frontiers? Which is right? Source ## Answer by Enrico Schumann (score 4) https://quant.stackexchange.com/a/59028 When returns follow an elliptical distribution (e.g. the Gaussian distribution), then minimising VaR and ES is equivalent to minimising variance. See https://people.math.ethz.ch/~embrecht/ftp/pitfalls.pdf. Then, the frontiers will be the same. ## Answer by Kermittfrog (score 2) https://quant.stackexchange.com/a/59029 This is a result of the two fund separation theorem or mutual fund separation theorem. Any (optimal) portfolio choice will take place on the efficient frontier. In a Markowitzian world, the asset universe is fully characterised by first and second (co-)moments. Hence, for any performance metric, you would always be able to obtain "more return at a given risk" or "less risk at a given return" by simply moving your portfolio towards the efficient frontier. The VaR and ES metrics are (simply) combinations of portfolio mean and risk: Hence, they can be improved by "moving left/up" as well. What may be observed, though, is a different diagram depicting mean-return vs VaR, or mean-return vs ES. They may look different. ## Answer by markowitz (score 0) https://quant.stackexchange.com/a/61284 If we follow the mean-risk ptf optimization strategy and use variance/VaR/ES as risk measure, the results in general are different. In other words the efficient frontier are not shared. However, as already said by Enrico Schumann, if the multivariate returns distribution of the assets is Elliptical (with finite variance) the frontier is shared. Note that the class of Elliptical distributions include the Normal one but is much more general, among others include t-student distribution. The graph above can make sense only in the elliptical case, however It seems me strange yet. In elliptical case entire frontier is shared and the global minimum risk point as well. In the graph are indicated three separated points/portfolios on the frontier; what minimum VaR/ES points mean? These two point not make so sense. These can be labeled as min mean variance too. Indeed in the horizontal line standard deviation is used. At the same points even min VaR and/or ES are achieved but the graph appear missleading to me. Moreover in the source, before the graph, it is said that: > In this chapter we formulate and solve the mean-CVaR portfolio model, where covariance risk is now replaced by the conditional Value at Risk as the risk measure. In contrast to the mean-variance portfolio optimization problem, we no longer assume the restriction consisting in the set of assets to have a multivariate elliptically contoured distribution. In this case is almost impossible that mean-VaR and/or mean-ES solutions are shared with mean-variance one; mean-VaR and/or mean-ES points must stay below the mean-variance frontier. Then I conclude that the graph is wrong.
Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.