Risk models occupy a curious place in modern finance. They are treated as essential infrastructure, being embedded in portfolio construction, risk reporting, regulatory processes. Yet, they are rarely questioned at a conceptual level. The implicit assumption is that risk modelling is largely a solved problem: choose a well-known framework, calibrate it carefully, and the rest is execution.

A closer look suggests otherwise.

In a recent paper published in the Journal of Risk (Risk.net) by the authors from Scientific Portfolio, “When betas meet the cross-section: a hybrid risk model for equity portfolios”, we revisit this assumption and argue that many institutional investors, and notably those who manage their portfolios via funds and mandates, generally do not have the data required to use risk models. Specifically, they do not have detailed nor up-to-date stock-level data of their holdings, as a result of their dependency on third-parties to manage their portfolios. Because these investors control about 80% of equity market capitalisation, this is a real issue: many of them do not have an aggregate view of their risks.

The broader message of the paper is not that existing risk models are wrong—in fact commercial models are excellent for those who have the data—but that they are not adequate for many investors. Risk modelling is not merely about explaining past returns; it is about producing stable, interpretable, and actionable inputs for decision-making under uncertainty with available data.

By rethinking the role of betas and explicitly embracing dimension reduction as a design principle, it is possible to move beyond the long-standing trade-offs that define much of today’s practice. The result is a model that is simpler to implement, more robust in use, and better aligned with the realities of portfolio management.

“This new model will bring relief to many institutional investors who are currently struggling to obtain a full view of their risks.”
— Benoit Vaucher, Director of Research, Scientific Climate Indices

“Using machine learning to compress risk information from factor betas, we are able to build a reliable risk model from just price data, which produces effective covariance matrices for portfolio optimization.”
— Matteo Bagnara, PhD, Senior Quant Researcher, Scientific Climate Indices

Read the full version published in the Journal of Risk
When betas meet the cross section: a hybrid risk model for equity portfolios