Robust Portfolio Construction: Overview

Mean-Variance Optimisation delivers the best portfolio for a given expected-return vector $\mu$ and covariance matrix $\Sigma$. In practice both are estimated, and the optimiser treats the estimates as truth. It overweights assets whose expected returns are overestimated and pairs of assets whose correlations are mismeasured, so the portfolio it calls optimal is largely a bet on estimation error. Realised risk comes out above the promised risk, weights swing from one rebalance to the next, and the manager's forecasting skill is diluted. This topic covers the two standard repairs, one for each input. Both work the same way: they pull a noisy estimate towards a structured prior, accepting a little bias for a large reduction in variance. The equal-risk-contribution portfolio, which avoids expected returns altogether, is the third response and sits in the risk-budgeting topic because it is built on risk contributions.