Structuring Portfolio Optimization Across Instruments and Forecast Rules

Summary

The post considers how to organize portfolio fitting across a grid of instruments and trading forecasts, such as momentum and carry rules. It compares fitting all rule and instrument combinations together, clustering correlated combinations, fitting first across instruments within each rule, and fitting first across rules within each instrument. The author describes an initial clustering exercise using weekly returns over roughly ten years, with thousands of combinations, and reports that the resulting groups did not offer a clear rationale for changing the existing approach.

A simpler comparison finds that forecasts for the same instrument have a higher average correlation than forecasts sharing the same rule across instruments. The author takes this as support for fitting within instruments first, then combining instruments. The evidence is exploratory: the post does not provide details of the cluster plots or a performance comparison of the alternative fitting methods. It also begins a smaller-universe follow-up but ends before giving its results, so the recommendation is suggestive rather than conclusive.

Key ideas

• Portfolio fitting can be organized jointly, by correlation clusters, or in a staged order across instruments and forecast rules.
• The author compares fitting forecasts within instruments first with fitting instruments within each rule first.
• In the reported data, average forecast correlation within an instrument exceeds average correlation within a rule.
• The clustering exercise did not reveal an interpretable advantage over the author's existing fitting approach.
• The post's smaller-universe follow-up is incomplete, limiting the strength of its conclusion.

Tags

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