Hierarchical Clustering of Trading Rule Returns for Portfolio Weights

Summary

The author checks whether a heuristic hierarchy for allocating forecast weights across trading rules is supported by correlations in rule returns. To build the correlation matrix, each rule is treated as a portfolio across the instruments actually weighted in the system; returns are pre-cost and are combined using instrument weights. This approach aims to avoid letting asset classes with many instruments or short histories distort the comparison, while retaining broad information even where a rule is not traded in every market.

Hierarchical clustering at different cluster counts broadly separates convergent rules, such as carry and mean reversion, from divergent rules such as momentum and breakout. As the number of clusters grows, the divergent group further divides by trading speed, while carry, skew, and relative momentum sometimes form their own groups. The author uses these patterns to propose a more detailed manual weighting hierarchy. The evidence is a sense check on one system’s rules and chosen return construction, not proof that the clusters are stable across samples or improve realized performance; cluster assignments also depend on the selected number of groups.

Key ideas

• Correlations of rule returns can be used to assess a heuristic hierarchy for forecast weights.
• The author builds rule portfolios across weighted instruments to reduce distortions from unequal market representation.
• Clustering separates convergent rules from divergent trend and breakout rules.
• Additional clusters tend to distinguish divergent rules by speed and separate some carry, skew, and relative momentum rules.
• The proposed hierarchy is a system-specific diagnostic and does not establish improved portfolio performance.

Tags

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