Hierarchical Risk Parity Allocation

Article Hudson & Thames

Summary

The document introduces Hierarchical Risk Parity (HRP) as a portfolio allocation method intended to reduce sensitivity to noisy return estimates and covariance-matrix inversion in traditional mean-variance optimization. It explains HRP in three stages: hierarchical clustering groups similar assets, quasi-diagonalization reorders the covariance matrix to reflect those clusters, and recursive bisection assigns weights from the top of the hierarchy downward. The tutorial also compares common linkage choices and describes the intuition behind allocating within clusters.

A PortfolioLab example applies HRP to a diverse historical asset dataset and shows how users can supply prices or custom inputs, inspect clusters, and specify long and short directions. The tutorial is instructional rather than a controlled performance comparison: it offers no evidence that the example portfolio outperforms a benchmark. It also notes that hierarchical methods work best when data contain clear clusters, limiting how readily the approach may transfer to other universes. The long-short option is an implementation extension beyond the original algorithm’s default long-only setup.

Key ideas

HRP uses clustering and recursive allocation rather than directly inverting a covariance matrix for mean-variance weights.
Hierarchical clustering groups assets by similarity, with results affected by the chosen linkage method.
Quasi-diagonalization orders similar assets together before recursive bisection assigns portfolio weights.
The tutorial demonstrates price and custom-input workflows and an implementation option for long-short portfolios.
The example is instructional and does not establish superior performance; clear asset clusters are favorable to the method.

Hierarchical Risk Parity for Portfolio Allocation

Summary

Key ideas

Tags