The model
Given expected returns, standard deviations and correlations, MVO finds the efficient frontier. The investor then picks a point using their risk aversion: maximize U = E(R) − 0.5 λ σ².
Why raw MVO goes wrong
- Input sensitivity. Expected returns are estimated with large errors; the optimizer treats them as exact and overweights assets whose returns happen to be overestimated (“error maximization”).
- Concentration. Unconstrained solutions often load heavily on a few assets.
- Single period, asset-only. It ignores liabilities, taxes, liquidity and the shape of returns (skew, fat tails).
Fixes
| Approach | How it works | What it solves |
|---|---|---|
| Constraints | No shorting, minimum/maximum weights | Prevents extreme weights, but imposes judgment |
| Reverse optimization | Back out the returns that make market-cap weights optimal | Sensible starting returns |
| Black-Litterman | Blend market-implied returns with views, weighted by confidence | Tilts portfolios modestly toward views |
| Resampling | Optimize many times on simulated inputs, average the weights | More diversified, stable weights |
| Monte Carlo simulation | Simulate wealth paths over many periods | Shows path dependence, spending, taxes |
Corner portfolios
With a no-shorting constraint, efficient portfolios between two corner portfolios are linear combinations of them. If a target return of 7% lies between corner portfolio A (8%) and B (6%), the weight on A is (7 − 6)/(8 − 6) = 50%.
Essay practice
Case. An endowment’s consultant shows an MVO result putting 55% in private equity, using the asset class’s 12-year historical return. The investment committee is uneasy.
Task. Give two reasons the result may be unreliable and one way to improve it.
Model answer
- MVO is very sensitive to expected-return inputs, and a 12-year historical average is a noisy estimate, so the optimizer is likely overweighting an overestimated return.
- Private equity’s reported returns are smoothed by appraisal valuations, understating volatility and correlation, which makes it look more attractive than it is.
Improvement: unsmooth the return series and use reverse optimization or Black-Litterman inputs, with a maximum weight constraint for illiquid assets.
Exam focus
Know MVO’s criticisms and the fixes, compute utility and corner-portfolio weights, and explain why private-asset data needs adjusting.