Hierarchical Risk Parity¶
Allocates asset weights by clustering return behavior, ordering assets by the tree, then recursively splitting weight between branches by their estimated variances.
Core Idea¶
Hierarchical Risk Parity (HRP) assigns portfolio weights from asset-return data. It builds a similarity tree, puts related assets together in a covariance layout, and repeatedly splits weight between branches.[^ref-17c62578cd37] The lower-variance branch gets more weight; within-branch inverse-variance weights estimate each branch's variance.[^ref-17c62578cd37] The original method avoids inverting the full covariance matrix.[^ref-17c62578cd37]
The original paper illustrates a correlation-derived clustering distance but permits other metrics.[^ref-17c62578cd37] A later Indian-stock application uses Ward linkage on transformed returns.[^ref-55c56b9dc187] Both retain the tree, ordering, and recursive allocation stages.
Scope of Application¶
HRP applies when asset returns support a hierarchy, covariance estimates, and a portfolio-weight decision:
- Simulated investments: The original ten-investment numerical example yields a tree, ordered correlation display, and actual HRP weights.[^ref-17c62578cd37]
- Historical stocks: An Indian auto-sector portfolio uses 2016–2020 returns to form a dendrogram and calculate stock weights.[^ref-55c56b9dc187]
- Changed asset universe: Applying the same stages to another financial asset set with adequate return data would conditionally yield a new allocation.[^ref-17c62578cd37]
- Changed return window: Re-estimating similarity and branch variances for another period conditionally reruns HRP with potentially changed weights.[^ref-17c62578cd37]
Clarity¶
At each split, HRP estimates branch variances from each branch's covariance and internal inverse-variance weights, then gives the lower-variance branch more parent weight.[^ref-17c62578cd37] The original baseline yields nonnegative weights totaling one, while the author permits altered constraints through modified split equations.[^ref-17c62578cd37] Clustering distance and branch variance serve different stages; Sen's Ward tree does not establish use of the original correlation-distance formula.[^ref-55c56b9dc187] A dendrogram alone cannot show equal asset risk contribution or future variance minimization.
Manages Complexity¶
The hierarchy turns pairwise return relationships into an ordered route for local weight decisions. It reveals which branch receives weight at each step and avoids full covariance-matrix inversion.[^ref-17c62578cd37] A mistaken tree can misdirect later splits; the method name cannot guarantee stable weights or better returns. Comparing allocations under alternate plausible trees tests that sensitivity without changing the defining three-stage procedure. Such a comparison can expose whether one branch dominates the allocation.
Abstract Reasoning¶
If one branch's estimated variance rises relative to its sibling's, the split rule lowers its share of their parent weight.[^ref-17c62578cd37] Its assets inherit that reduction. If the tree changes, assets can meet different siblings and receive different weights under the same rule. Hierarchy, ordering, and allocation must therefore be examined together.
Knowledge Transfer¶
The procedure can be rerun on different asset universes and data windows, as the simulated and Indian-stock cases show.[ref-17c62578cd37][ref-55c56b9dc187] Its recursive division contains an Allocation operation: a finite portfolio budget is assigned among asset uses under a stated rule. Other allocation methods need not use HRP. Clustering, Diversification, and Portfolio Optimization are related concepts, not synonyms for this computation. Outside finance, a tree-shaped budget split is only an analogy unless a comparable branch-risk calculation is specified and tested.
Relationships to Other Abstractions¶
Current abstraction Hierarchical Risk Parity Domain-specific
Parents (1) — more general patterns this builds on
-
Hierarchical Risk Parity is part of Allocation Prime
The recursive branch split assigns a finite portfolio budget among competing assets; this Allocation operation is an identity-bearing internal stage of every HRP run.
Hierarchy path (1) — routes to 1 parentless root
- Hierarchical Risk Parity → Allocation → Scarcity → Constraint
Neighborhood in Abstraction Space¶
Hierarchical Risk Parity sits in a sparse region of the domain-specific corpus (79th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.
Family — Scaling Laws & Growth Patterns (13 abstractions)
Nearest neighbors
- Portfolio Optimization — 0.84
- Merton's portfolio problem — 0.82
- Incremental Capital–Output Ratio — 0.82
- Ulcer Index — 0.82
- Chow–Liu Tree — 0.82
Computed from structural-signature embeddings · 2026-10-17
References¶
[^ref-17c62578cd37]: Marcos López de Prado. (2016). Building Diversified Portfolios that Outperform Out-of-Sample, SSRN 2708678. Full author paper: Stages 1–3, PDF pp. 4–8; A Numerical Example, PDF p. 8; Exhibits 4–7, PDF pp. 21–24. The title reports the paper's research aim and is not a universal performance claim in this entry. [^ref-55c56b9dc187]: Jaydip Sen, Sidra Mehtab, Abhishek Dutta, and Saikat Mondal. (2022). Hierarchical Risk Parity and Minimum Variance Portfolio Design on NIFTY 50 Stocks, arXiv:2202.02728, preprint of the 2021 IEEE DASA paper. Methodology step 6, PDF pp. 1–2; Auto sector and Figures 1–2, PDF pp. 2–3.