Fourth Annual Risk Conference 2026
What it is about
Dany Cajas, independent consultant and creator of the open-source library Riskfolio-Lib, presented at the Fourth Annual Risk Conference (September 17, 2026) a family of entropic risk measures for managing the extreme risk of a portfolio.
The 34 slides review value at risk (VaR) and conditional value at risk (CVaR), extreme value theory and the dynamic GARCH-EVT model, and then present entropic value at risk (EVaR) and relativistic value at risk (RLVaR), with a comparison of their properties. They apply the measures in loss backtests and in portfolios built with assets of the Dow Jones index, and close with the Entropy Pooling method to incorporate analyst views on the risk measures.
Objectives
- Compare VaR, CVaR, extreme value models, EVaR and RLVaR by their mathematical properties and computational complexity.
- Show how these measures are evaluated in loss backtests and in portfolio optimization.
- Present the Entropy Pooling method to incorporate views on portfolio risk.
Key takeaways
- VaR is not a coherent risk measure because it fails subadditivity and is not convex; CVaR is.
- EVaR (Ahmadi-Javid, 2012) and RLVaR (Cajas, 2023) are coherent and convex measures that diversify risk, are robust and have an optimal portfolio model.
- In loss tests with USD 1 million invested in the assets CMG and TSLA, EVaR and RLVaR showed the lowest failure rates (between 0.7% and 1.2%), all in the green zone of the traffic light.
- In the test of Dow Jones portfolios between April 2015 and July 2026, the maximum return-to-risk portfolio with RLVaR had a total return of 795.21%, against 500.65% with variance and 191.62% for the index.
- With Entropy Pooling (Meucci, 2007) the analyst sets views, for example a VaR of at least 3%, and a posterior distribution is obtained that meets them with the smallest change in information.
Connect with our networks