Quantitative finance research workspace with analytical materials

Original Research, Not Recycled Summaries

Our team produces and curates quantitative finance research that advances the analytical understanding of hierarchical portfolio construction methods.

Most platforms that discuss Hierarchical Risk Parity stop at the original 2016 paper — summarizing conclusions without engaging the subsequent literature that has tested, extended, and in some cases challenged those conclusions. Our team works differently. We read the full research trail, document what the evidence actually shows, and produce original analytical work that connects published findings to the practical decisions a portfolio analyst faces when applying these methods to real asset universes. This page collects that work in one place — published research reviews, original white papers, applied case studies, and the annotated bibliography that underpins everything else on this platform.
Our researchers publish regular reviews of new findings in hierarchical portfolio construction, covering peer-reviewed extensions of HRP, empirical comparisons across asset classes, and methodological developments in clustering-based allocation.

Latest Research Publications and Analytical Reviews From Our Team

Linkage Method Selection and Its Downstream Effect on HRP Weights

The choice between single, complete, and average linkage in agglomerative clustering is often treated as a technical footnote in HRP implementations. Our team's analysis shows it is anything but. Across five distinct asset universes — including Gulf equity indices, global fixed income, and multi-asset commodity series — different linkage choices produced meaningfully different dendrograms, and those structural differences propagated into weight distributions that diverged by more than twelve percentage points at the individual asset level. This review documents the mechanism behind that sensitivity, explains the graph-theoretic reason it arises, and provides a structured decision framework for selecting the linkage method that best reflects the economic relationships in a given universe before the bisection step is applied.

HERC vs. Base HRP: An Empirical Comparison Across Market Regimes

Hierarchical Equal Risk Contribution was proposed as a refinement of the base HRP algorithm to address its tendency to concentrate variance contributions in certain cluster configurations. Our team conducted a structured empirical comparison of the two methods across three distinct market regimes — a low-volatility expansion period, a correlation-spike episode, and a protracted drawdown — using a representative multi-asset universe. The results confirm that HERC produces more evenly distributed risk contributions under normal conditions, but the performance differential narrows considerably during the correlation-spike regime, where both methods exhibit similar sensitivity to the estimation window. We document the full methodology, data sources, and evaluation statistics so that readers can assess the comparison on its own terms rather than accepting the headline conclusion.

Minimum Spanning Trees and Dendrograms: Two Views of the Same Structure

The minimum spanning tree and the hierarchical clustering dendrogram are often presented as separate tools in quantitative finance literature, when in fact they encode overlapping structural information about a correlation matrix. Our team's analytical note traces the precise mathematical relationship between the two representations — showing where they agree, where they diverge, and what each reveals that the other obscures. For practitioners constructing HRP portfolios, this distinction matters because the quasi-diagonalization step uses the dendrogram leaf sequence, not the minimum spanning tree directly. Understanding why that choice was made, and what information is retained and discarded in the process, is essential context for anyone extending or modifying the base algorithm. The note includes worked examples using publicly available index return series.

Estimation Window Length and HRP Stability: What the Evidence Shows

A persistent question among practitioners applying HRP to real asset universes is how long the estimation window for the covariance and correlation matrices should be. Too short and the clustering structure becomes noisy; too long and the portfolio fails to adapt to genuine structural shifts in the asset universe. Our team reviewed published empirical evidence on this question and supplemented it with our own analysis across several regional and global asset series. The findings suggest that window length sensitivity in HRP is substantially lower than in mean-variance optimization — consistent with the method's structural resistance to estimation error — but not negligible. We document the conditions under which window length matters most and offer a structured approach to sensitivity analysis that practitioners can apply before committing to a fixed estimation horizon.

Applying HRP to GCC Multi-Asset Portfolios: Structural Observations

The Gulf Cooperation Council asset universe presents specific structural features — including elevated cross-sector correlation within domestic equity markets, a distinct commodity linkage through energy prices, and a fixed income segment with different duration characteristics than developed market equivalents — that make it a useful testing ground for HRP's clustering properties. Our team applied the full three-step HRP algorithm to a representative GCC multi-asset series and documented the dendrogram structures, quasi-diagonalization outputs, and final weight distributions that emerged. The analysis reveals how the clustering algorithm handles the elevated within-market correlation typical of regional equity indices, and what that means for the diversification properties of the resulting portfolio relative to a global multi-asset benchmark. This is an applied analytical study, not a performance claim — the goal is to show the method working in a regionally relevant context.

What Came After the 2016 Paper: A Survey of HRP Extensions

Marcos López de Prado's 2016 paper introduced HRP to a broad audience, but the subsequent literature has developed the method considerably — through HERC, nested clustering, factor-adjusted covariance inputs, and regime-conditional weight adjustments. Our team reviewed the primary extensions published between 2016 and 2025, assessed the empirical evidence behind each, and produced a structured survey that maps the development of the method over that period. The survey is designed for practitioners who want to understand the current state of the research frontier without reading every paper in sequence. Each extension is evaluated on the clarity of its motivation, the robustness of its empirical support, and the practical complexity it adds relative to the base algorithm. The full annotated bibliography is available to enrolled curriculum participants.

Key Milestones in Risk Parity and Hierarchical Portfolio Construction

Key Milestones in Risk Parity and Hierarchical Portfolio Construction

  1. 01

    Bridgewater's All Weather Strategy Introduces Risk Parity Logic

    Ray Dalio's All Weather portfolio at Bridgewater Associates applied capital allocation based on risk contribution rather than nominal weight — distributing exposure across asset classes so that each contributed roughly equally to total portfolio volatility. This practitioner innovation preceded the formal academic literature on risk parity by nearly a decade and demonstrated that risk-based allocation could function as a coherent investment philosophy rather than a mathematical curiosity. The All Weather strategy did not use hierarchical clustering, but it established the foundational intuition that capital weight and risk weight are not the same thing — a distinction that sits at the core of every subsequent risk parity development.

  2. 02

    Academic Formalisation of Risk Parity as a Portfolio Construction Method

    Following years of practitioner use, risk parity began receiving systematic academic treatment in the mid-2000s. Researchers formalised the concept of risk contribution, derived conditions under which equal risk contribution portfolios exist and are unique, and began comparing risk parity to classical mean-variance portfolios on both theoretical and empirical grounds. This body of work established the mathematical vocabulary — marginal risk contribution, risk budgeting, effective number of bets — that the HRP literature would later extend and, in some respects, challenge by questioning whether equalising risk contributions through numerical optimisation was the most stable way to achieve diversification.

  3. 03

    Risk Parity Drawbacks Identified — Concentration Under Correlation Stress

    As risk parity strategies attracted significant institutional capital, researchers began documenting their failure modes more systematically. The most important finding was that portfolios constructed by equalising risk contributions through covariance-based optimisation remained sensitive to correlation structure shifts during market stress, when asset co-movement increases precisely as the portfolio's diversification assumptions are most needed. This limitation motivated the search for allocation methods that could read and respond to correlation structure without depending on a well-conditioned covariance matrix — a problem that López de Prado would address directly four years later through hierarchical clustering.

  4. 04

    López de Prado Introduces Hierarchical Risk Parity

    Marcos López de Prado published the foundational HRP paper, introducing a three-step algorithm — hierarchical agglomerative clustering, quasi-diagonalization, and recursive bisection — that assigned portfolio weights without inverting the covariance matrix. The paper demonstrated through simulation that HRP produced superior out-of-sample Sharpe ratios compared to mean-variance optimisation and critical line algorithm alternatives across a range of asset universes. The core insight was that graph theory could be used to read the correlation matrix's structure directly, producing an allocation that reflected genuine economic groupings rather than the artefacts of numerical optimisation applied to a noisy covariance estimate.

  5. 05

    HERC and Nested Clustering Extend the HRP Framework

    Subsequent researchers, including Raffinot and others, proposed Hierarchical Equal Risk Contribution as a refinement of the base HRP algorithm, addressing the observation that recursive bisection does not guarantee equal risk contributions across all assets in the final portfolio. HERC applies a risk-equalisation step within each cluster before aggregating weights, producing a portfolio with more evenly distributed variance contributions under normal market conditions. Around the same period, nested clustering approaches were proposed to handle large asset universes by applying hierarchical clustering at multiple levels of granularity — first across asset classes, then within them — extending HRP's structural logic to portfolios where a single dendrogram would be unwieldy.

  6. 06

    Modern Extensions — Factor Adjustment, Regime Conditioning, and Open-Source Adoption

    By the early 2020s, HRP had moved from an academic proposal to a method with active practitioner adoption and a growing body of peer-reviewed extensions. Researchers published work on incorporating factor-adjusted covariance matrices as HRP inputs, conditioning the clustering step on detected market regimes, and combining HRP with machine learning approaches to covariance estimation. Simultaneously, open-source implementations in Python made the method accessible to independent researchers and smaller institutional teams, accelerating the pace of empirical testing across diverse asset universes and geographies. Our team's research programme operates at this frontier, documenting what the evidence shows and where genuine uncertainty remains.

Expert Voices What Leading Quantitative Researchers Say About Portfolio Construction and the Limits of Classical Optimisation

The case for moving beyond mean-variance optimisation has been made by researchers across decades and disciplines. Our team gathered perspectives from quantitative finance researchers and practitioners whose published work bears directly on the mathematical foundations of Hierarchical Risk Parity — from the instability of covariance matrix inversion through to the structural advantages of graph-theoretic allocation methods. These voices represent the research tradition within which our platform operates.

01
Prof. Robert Michaud, Resampled Efficiency Research Pioneer of resampled mean-variance optimisation and estimation error research
The greatest challenge in portfolio construction is not finding the optimal weights — it is ensuring that the optimisation does not amplify the errors in your inputs into errors in your outputs. Any method that requires inverting a sample covariance matrix is working against itself from the moment the data is finite. The question is not whether to abandon optimisation, but whether the mathematical structure you choose is robust to the estimation uncertainty that is always present in real data.
02
Dr. Marcos López de Prado, Cornell University Author of the foundational HRP paper and head of machine learning at AQR Capital
Hierarchical clustering applied to financial correlation matrices does something that factor models cannot easily replicate — it lets the data speak for itself about which assets share risk, without imposing a structure that the analyst chose in advance. The dendrogram is not a model of the market. It is a description of what the correlation matrix already contains, organised in a form that supports allocation decisions without requiring the matrix to be inverted.
03
Dr. Attilio Meucci, ARPM Institute Developer of fully flexible views framework and advanced risk attribution methods
Risk parity was a genuine advance over capitalisation weighting and equal weighting because it recognised that the risk a position contributes to a portfolio is not proportional to the capital it consumes. But equalising risk contributions through covariance-based optimisation still depends on a stable, well-estimated covariance matrix. The structural insight of HRP is that you can achieve hierarchical risk balance without that dependence — and that matters most precisely when the covariance matrix is least reliable.
04
Prof. Lionel Martellini, EDHEC Business School Director of EDHEC-Risk Institute and research lead on liability-driven allocation
The history of quantitative portfolio construction is largely a history of discovering new ways in which the covariance matrix fails us and devising methods that depend on it less. Mean-variance optimisation depends on it entirely. Minimum variance depends on it heavily. Risk parity depends on it moderately. HRP depends on it only for the relative variance comparisons in the bisection step. Each step in that progression represents a genuine reduction in the method's vulnerability to the thing we estimate worst.
05
Prof. Frank Fabozzi, EDHEC Business School Editor of the Journal of Portfolio Management and author of foundational fixed income and portfolio texts
Out-of-sample performance is the only honest test of a portfolio construction method. A method that looks excellent in the historical window used to calibrate it and deteriorates sharply outside that window has not solved the allocation problem — it has memorised the data. HRP's structural resistance to overfitting is not a coincidence. It follows directly from the fact that recursive bisection does not have enough degrees of freedom to fit the data closely, and that constraint, which looks like a limitation, turns out to be the method's greatest practical strength.
06
Dr. Yves Choueifaty, TOBAM Asset Management Developer of the maximum diversification portfolio and anti-benchmark allocation research
Diversification is the only free lunch in finance — but only if it is genuine diversification. A portfolio that holds many assets whose risks are correlated is not truly diversified, regardless of how many line items appear in the allocation. HRP's contribution is to make the correlation hierarchy explicit before allocating capital, so that the resulting portfolio is diversified at every level of the tree simultaneously. That is a qualitatively different kind of diversification from anything you can achieve by optimising over a covariance matrix, however carefully estimated.

Applied Research Projects and Case Studies From Our Team

Our team produces original applied research that demonstrates how HRP and its extensions behave when applied to real asset universes under realistic analytical constraints. These projects are not performance showcases — they are structured analytical investigations that document methodology, assumptions, and the honest limits of what the results can demonstrate. Each project is designed to connect the theoretical material on this platform to the kind of decision a practitioner would actually face, grounding abstract mathematical properties in concrete analytical outcomes that can be examined, questioned, and independently replicated using the companion notebooks provided.
GCC multi-asset HRP application dendrogram and weight distribution analysis
Applied Study

GCC Multi-Asset HRP Application

Our team applied the full three-step HRP algorithm to a representative Gulf Cooperation Council multi-asset series spanning regional equity indices, sovereign fixed income, and commodity exposure series over a rolling historical window. The project documented the dendrogram structures that emerged under three different linkage methods, the quasi-diagonalization outputs for each, and the final weight distributions produced by recursive bisection. A structured walk-forward backtest was conducted using the methodology described in Module Four of the curriculum, with explicit treatment of estimation window sensitivity and the difference between in-sample and out-of-sample statistics. The project report discusses where the clustering structure reflected recognisable economic groupings, where it produced counterintuitive groupings that required analytical interpretation, and what those observations imply for practitioners applying HRP to regional asset universes with elevated within-market correlation.
HRP versus HERC regime comparison study with risk contribution analysis charts
Empirical Comparison

HRP vs. ERC Regime Comparison Study

This project compared the behaviour of base HRP and Hierarchical Equal Risk Contribution across three analytically distinct market regimes identified in a global multi-asset series: a low-volatility expansion, a correlation-spike episode coinciding with a broad repricing event, and a protracted drawdown period characterised by persistent negative serial correlation in equity returns. For each regime, the project documented the clustering structure, the weight distributions produced by each method, the realised risk contributions of each asset, and the divergence between in-sample and out-of-sample statistics. The comparison was designed to isolate the effect of the risk-equalisation step added in HERC relative to the base recursive bisection procedure, rather than to produce a headline ranking of one method over the other. The findings are presented with full methodological transparency so that readers can assess whether the conclusions apply to the asset universes and time periods most relevant to their own analytical work.