Robeco logo

Disclaimer Robeco Switzerland Ltd.

The information contained on these pages is solely for marketing purposes.

Access to the funds is restricted to (i) Qualified Investors within the meaning of art. 10 para. 3 et sequ. of the Swiss Federal Act on Collective Investment Schemes (“CISA”), (ii) Institutional Investors within the meaning of art. 4 para. 3 and 4 of the Financial Services Act (“FinSA”) domiciled Switzerland and (iii) Professional Clients in accordance with Annex II of the Markets in Financial Instruments Directive II (“MiFID II”) domiciled in the European Union und European Economic Area with a license to distribute / promote financial instruments in such capacity or herewith requesting respective information on products and services in their capacity as Professional Clients.

The Funds are domiciled in Luxembourg and The Netherlands. ACOLIN Fund Services AG, postal address: Leutschenbachstrasse 50, CH-8050 Zürich, acts as the Swiss representative of the Fund(s). UBS Switzerland AG, Bahnhofstrasse 45, 8001 Zurich, postal address: Europastrasse 2, P.O. Box, CH-8152 Opfikon, acts as the Swiss paying agent.

The prospectus, the Key Investor Information Documents (KIIDs), the articles of association, the annual and semi-annual reports of the Fund(s) may be obtained, on simple request and free of charge, at the office of the Swiss representative ACOLIN Fund Services AG. The prospectuses are also available via the website https://www.robeco.com/ch.

Some funds about which information is shown on these pages may fall outside the scope of CISA and therefore do not (need to) have a license from or registration with the Swiss Financial Market Supervisory Authority (FINMA).

Some funds about which information is shown on this website may not be available in your domicile country. Please check the registration status in your respective domicile country. To view the Robeco Switzerland Ltd. products that are registered/available in your country, please go to the respective Fund Selector, which can be found on this website and select your country of domicile.

Neither information nor any opinion expressed on this website constitutes a solicitation, an offer or a recommendation to buy, sell or dispose of any investment, to engage in any other transaction or to provide any investment advice or service. An investment in a Robeco Switzerland Ltd. product should only be made after reading the related legal documents such as prospectuses, annual and semi-annual reports.

By clicking “I agree” you confirm that you/the company you represent falls under one of the above-mentioned categories of addressees and that you have read, understood and accept the terms of use for this website.

I Disagree

19-03-2024 · Research

Covariance rhapsody: A reality check for evaluating risk models

Portfolio management is all about trading off expected return and risk. The key ingredient to measuring and managing portfolio risk is the variance-covariance (VCV1) matrix which needs to be estimated for the given investment universe.

    Authors

  • Maarten Jansen - Researcher

    Maarten Jansen

    Researcher

  • Harald Lohre - Head of Quant Equity Research

    Harald Lohre

    Head of Quant Equity Research

At its core, the VCV informs about assets’ riskiness and their inter-dependencies, as measured by their variance and covariance, respectively. The natural candidate to use is the sample covariance matrix; however, this estimator is prone to error and not suitable when the number of assets under consideration is large, as is often the case when optimizing equity portfolios. To this end the academic literature proposes myriad alternative VCV estimators to address these limitations. But how can we best evaluate the practical relevance of a given VCV estimator?

In a recent research paper, quant researchers Clint Howard, Maarten Jansen, Harald Lohre, and M. Sipke Dom set out to answer this question, putting a wide range of alternative VCV estimators to the practical test. Importantly, they challenge the common academic practice of evaluating the relevance of novel VCV estimators using the unconstrained global minimum variance (GMV) portfolio. Indeed, when validating VCV estimators based on the ex-post volatility of this portfolio, the researchers confirm the academic backing for considering shrinkage and covariance dynamics in modelling the VCV for equity portfolio construction.

This is evident in the leftmost bar of the below figure that highlights a wide range of volatility outcomes for unconstrained GMV portfolios that differ only in the choice of underlying VCV estimator. Yet, these portfolios are often impractical due to their high leverage, concentration, turnover, and transaction costs. The researchers therefore investigated how the opportunity set for volatility improvement changes when making the GMV test portfolios more investable. Although long-only constrained GMV portfolios still allow for meaningful volatility improvements, their overly concentrated stock allocation calls for further constraining portfolio weights.

Active Quant: finding alpha with confidence

Blending data-driven insights, risk control and quant expertise to pursue reliable returns.

Find out more

Figure 1: Volatility outcomes for test GMV portfolios

Figure 1: Volatility outcomes for test GMV portfolios

Source: Robeco 2024

Resorting to truly investible GMV portfolios, the researchers reveal a considerably reduced opportunity set for alternative VCV estimators. Similar findings hold for alternative risk-based portfolio construction approaches, such as risk parity portfolios that aim to maximize portfolio diversification.

These findings highlight the discrepancies between the optimal VCV matrix estimator across different portfolios, suggesting that what works best for an unconstrained GMV portfolio may not hold under more realistic conditions with significant investment constraints. Such realistic test portfolios suggest that the overall room for improvement from a given VCV estimator is limited. Nevertheless, statistically significant improvements can still be made under the right circumstances.

Read the full paper on SSRN.

Footnote

1From Investopedia: ‘Variance refers to the spread of a data set around its mean value, while a covariance refers to the measure of the directional relationship between two random variables… Variance is used by financial experts to measure an asset's volatility, while covariance describes two different investments' returns over a period of time when compared to different variables.’

Get the latest insights

Subscribe to our newsletter for investment updates and expert analysis.

Don’t miss out