Inverse-Volatility Portfolios and Their Performance Questions

Summary

The note introduces the inverse-volatility portfolio (IVP) as a heuristic allocation that assigns more weight to lower-volatility assets. It contrasts this approach with equal weighting and Markowitz mean-variance methods, which require covariance estimation and optimization. It also distinguishes IVP from risk parity, while noting that related literature derives the former from the latter. The low-volatility effect motivates these allocations, but the excerpt does not provide a full construction recipe.

The discussion frames a comparison rather than reporting a completed empirical result. It asks whether IVP beats the global minimum-variance portfolio in and out of sample, and whether it can surpass equal weighting across repeated simulations. It cites prior work describing weak out-of-sample results for optimized portfolios, the resilience of equal weighting, and hierarchical clustering as a potential compromise between risk-minimizing extremes. No simulation design, data, or findings are included here, so the performance questions remain unresolved.

Key ideas

• Inverse-volatility weighting is a heuristic that allocates more to assets with lower volatility.
• The inverse-volatility portfolio and risk parity are related but are not identical concepts.
• Mean-variance optimization relies on covariance estimates and may perform less reliably out of sample than in sample.
• The excerpt raises comparisons against global minimum variance and equal weighting but supplies no empirical answer.
• Hierarchical clustering is presented as a way to form portfolios between contrasting risk-minimization approaches.

Tags

# How good is the inverse-volatility portfolio?


# How good is the inverse-volatility portfolio?












Heuristic portfolio construction techniques include the equally-weighted portfolio (1/N) and the inverse volatility portfolio (IVP), which is based on the low-volatility effect. They can be assembled much more efficiently than Markwoitz' mean-variance model that uses optimization of the asset returns covariance matrix. Markowitz' global minimum-variance portfolio (GMV) is known to promise high in-sample performance, but can disappoint out-of-sample.

Maillard (2010) derives IVP from the risk parity portfolio I think, but to make it clear, the two (IVP and risk parity) are not the same. The paper stops at the derivation, and only evaluates the inverse-beta portfolio instead, which is also based on the low-volatility effect.

de Prado (2018) went on to apply unsupervised learning algorithm called hierarchical clustering to form portfolios that outperform both GMV and the IVP by finding a compromise between the two since they are extreme opposites of one another in terms of the type of risk they minimize.

The (heuristic) 1/N portfolio is a benchmark that is tough to beat out-of-sample (de Miguel 2009), but how about the IVP? does the IVP consistently outperform Markowitz' GMV portfolio in-sample and out-of-sample? and does the IVP outperform the 1/N? Or are these three roughly the same in their chances of coming out on top, based on repeated simulations of new returns/volatility data?

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

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