Portfolio Optimisation Under Estimation Error and Changing Data Windows

Summary

This brief document frames a historical portfolio optimisation problem around choices that affect both the estimates and the resulting weights. It asks how a backtest should handle information that would not yet have been available at each point in time, and whether the fitting period should use a rolling window or an expanding history. It also raises the need to account for substantial differences in asset volatility when expressing portfolio weights.

The central concern is that conventional optimisation can produce unstable, extreme allocations, especially when small changes in estimated mean returns drive large weight changes. The document poses bootstrapping and the assumption of equal expected returns as possible alternatives or safeguards. It signals a comparison of in-sample, rolling-window, and expanding-window approaches, but the supplied text contains only section headings and questions, not methods, results, or a conclusion. It therefore offers a useful checklist of modelling issues rather than evidence favoring a particular optimisation procedure.

Key ideas

• Historical portfolio tests must account for which data would have been available at each date.
• Rolling and expanding estimation windows represent different ways to choose the fitting history.
• Portfolio weights need to reflect differences in asset volatility.
• Small differences in estimated mean returns can produce unstable and extreme optimised weights.
• Bootstrapping and equal-mean assumptions are raised as possible ways to address estimation noise, without reported results.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.