Podcast: Alexandre Antonov turns down the noise in Markowitz

Adia quant explains how to apply hierarchical risk parity to a minimum-variance portfolio

Portfolio construction is the art (and science) of allocating weights to a collection of assets to achieve a given objective – typically, a target volatility or risk-adjusted return.

The Markowitz Mean-Variance Optimization Model, introduced in 1952, remains the benchmark for doing this, despite its well-known drawbacks. The process involves calculating a correlation matrix for a set of assets and then inverting it. However, running the calculation for a large number of assets can be overwhelming and the estimates of the weights tend to be unstable over time.

Quantcast – a Risk.net Cutting Edge podcast · Alexandre Antonov 04/02/2025

 

Alexandre Antonov, quant researcher and development lead at the Abu Dhabi Investment Authority (Adia), sums up the problem. “There is a certain noise in [the correlation] matrix. When we take its inverse, this noise can be amplified,” he says. This in turn can result in optimal portfolios varying significantly from day to day, potentially increasing transaction costs.

In this episode of Quantcast, Antonov explains how an approach called Hierarchical Risk Parity (HRP) can be used to overcome these issues and produce more robust and stable estimates for asset weights.

Related

Overcoming Markowitz’s instability with hierarchical risk parity 

Portfolio optimisation via HRP provides stable and robust weight estimates

HRP was developed by Marcos Lopez de Prado, global co-head of quantitative research and development at Adia, in 2016. The approach consists of two steps. First, clusters of assets are optimised to build a number of optimised sub-portfolios. Then, the sub-portfolios are combined and a second optimisation is performed. Essentially, the Markowitz optimisation is done twice, each time on a portfolio that has fewer components than would otherwise be the case.

The approach sits somewhere between Markowitz’s minimum-variance portfolio, which assumes perfect knowledge of the covariance matrix, and risk parity, where no knowledge of correlations is assumed.

The main application of HRP is in estimating the confidence levels of optimisation weights, as a direct derivation from the noise estimates.

As a natural consequence of this, clusters can be selected based on their noise level to build portfolios that are more stable over time.

Antonov’s latest paper, co-authored with Lopez de Prado and Adia’s co-head of quant R&D Alexander Lipton, compares HRP with the original Markowitz method and finds that it produces more robust risk weights. Anotonov’s next project is to test HRP against more modern and sophisticated portfolio construction techniques.

Index

00:00 Introduction

01:45 Sell-side vs buy-side research

03:47 Markowitz and noisy data

09:07 Hierarchical risk parity

15:27 Calculation of asset weights

17:00 Why is Markovitz still relevant?

21:21 Applications of HRP

27:20 Future research projects

To hear the full interview, listen in the player above, or download. Future podcasts in our Quantcast series will be uploaded to Risk.net. You can also visit the main page here to access all tracks, or go to Spotify, Amazon Music or the iTunes store to listen and subscribe.