Portfolio optimization under estimation error and out-of-sample evaluation

Research significance

Allocators, risk committees, and model reviewers

Portfolio managers, asset allocators, investment committees, and model validators use optimizers to convert noisy expected returns and covariances into precise weights. The evidence distinguishes numerical sophistication from decision reliability.

Decision context

Authority assigned to an optimized allocation

Do not interpret an in-sample efficient frontier or Sharpe ranking as a forecast. Decide first whether the mandate is return maximization, concentration control, or a stable risk budget, then evaluate the construction method against that stated objective.

Four construction methods sat the same test. The experiment drew 2,891 daily observations for 15 liquid ETFs from a frozen adjusted-close snapshot. Maximum-Sharpe MVO, box-uncertainty robust MVO, equal-risk-contribution risk parity, and hierarchical risk parity were each fit on 2015–2021 data. Their weights were then frozen and applied to one untouched 2022–2026 window.

The robust portfolio posted the best out-of-sample Sharpe, about 0.6. It also stayed concentrated in QQQ and drew down more than 30% in 2022, then rode the 2023–2025 mega-cap rally. The box uncertainty set lowered estimated return levels without changing their ranking, so concentration survived. One 4.5-year window cannot separate a robust method from a fortunate regime.

Differences of ±0.2 in Sharpe sit well inside an estimated sampling error near 0.5 for this evaluation window. So the defensible conclusion describes realized risk shape. It does not name the method that will earn the highest future return.

• Separate the return objective from the risk-budget objective before comparing methods.
• Report concentration, risk contribution, drawdown, and out-of-sample behavior beside the in-sample optimum.
• State the uncertainty treatment plainly: shrinking return levels does not change an uncertain ranking.
• Rank methods only after multiple rolling-origin evaluations. This study ran one window.
• State each method's regime dependency. Risk-based allocations can embed a bond and stock–bond-correlation bet.

Analyze allocation sensitivity → Inspect the research record → Reproduce from the executed notebook ↗

Lab measurement Maximum-Sharpe optimization scored about 1.12 in the 2015–2021 fit period. It scored 0.44 in the sealed 2022–2026 evaluation. Risk-based methods held a steadier risk shape without establishing a return advantage.

Institutional record The two windows straddle a documented change in stock–bond correlation and monetary conditions. That change brought inflation pressure, rapid tightening, and weakness in equities and nominal bonds at once.

Interpretive synthesis An optimized portfolio assumes its fitted return and covariance relationships still bear on the decision. Precise weights carry relationships measured in one observed environment into a different one.

Causal boundary: the experiment does not identify regime change as the cause of the performance ranking. Correlation, estimation error, concentration, asset selection, and the realized equity path can each move the result.

Examine the allocation regime →

Regime riskStock–bond diversification across changing correlation environmentsThe stock–bond correlation assumption that 2022 called into question. Model calibrationHistorical VaR under coverage and independence testsThe calibration tests a portfolio risk forecast has to pass.