Estimation Error in Portfolio Optimisation

The Limits of Optimisation: Why Simpler Portfolios Often Win

Mean-variance optimisation is the theoretical cornerstone of modern portfolio construction. In practice, however, estimated parameters are noisy, and that noise can render sophisticated optimisation worse than doing nothing at all. This paper uses a controlled simulation framework to quantify the performance cost of estimation error across five portfolio strategies: the unconstrained plug-in tangency portfolio, a long-only constrained tangency portfolio, the global minimum-variance portfolio, a risk parity allocation, and the naïve equal-weight portfolio. We find that under realistic estimation conditions, the equal-weight and risk parity strategies deliver near-optimal Sharpe ratios, while the unconstrained plug-in portfolio achieves only a fraction of the theoretical optimum. These results carry direct implications for systematic portfolio construction: in regimes of high parameter uncertainty, analytical simplicity is not a concession. It is a competitive advantage.