Differential Evolution for Robust Portfolio Construction

Summary

This report discusses using optimization methods for asset allocation and portfolio construction, especially when objectives are complex, high dimensional, discontinuous, or heavily constrained. It argues that the in-sample optimum may not remain optimal in future periods, so a solution with a lower in-sample score can perform better out of sample. This is presented as a practical modeling principle rather than a formal financial theorem.

The report introduces differential evolution, a population-based search method that uses real-valued candidates, difference-based mutation, and competitive selection. It describes the method as suitable for difficult optimization problems that conventional mathematical programming may not handle well. Two backtests are summarized: a quarterly rebalanced index portfolio and an annual currency-fund enhancement portfolio using money-market, short-duration, and bond funds. The reported results are historical and the document provides no detailed test design, transaction-cost assumptions, or evidence that performance persists. It also suggests applications such as fund-of-funds allocation and robo-advisory portfolios.

Key ideas

• The in-sample best portfolio may not deliver the best out-of-sample results.
• Differential evolution searches across populations of real-valued candidate solutions.
• Difference-based mutation and competitive selection help the algorithm explore complex objectives.
• The report applies the method to a quarterly index portfolio and an annual cash-fund enhancement portfolio.
• The reported backtests lack detailed assumptions needed to judge robustness or future performance.

Tags

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