QuarcHome
Model catalogueBack to the model index

Expected Shortfall (ES) and CVaR optimisation

Measures mean loss in the worst tail, with the appropriate quantile convention, and supports tail-risk optimisation.

How it works

Expected Shortfall is the mean loss across the worst specified tail probability, including the appropriate portion of any probability mass at the quantile. In the Rockafellar–Uryasev representation, an optimisation over a threshold adds the average excess above that threshold divided by tail probability to the threshold itself. This supports portfolio optimisation over simulated or historical loss scenarios.

Worked example

For 100 equally weighted scenarios with the worst five losses of 4, 5, 7, 8 and 16, the 95% Expected Shortfall is (4 + 5 + 7 + 8 + 16) ÷ 5 = 8. Different conventions for the reported empirical VaR near a tie need not change the definition of the worst five-percent mean.

Validation and limitations

Report the quantile convention, scenario weights and sampling uncertainty. Expected Shortfall responds to the severity of the worst observations, so small tail samples can be unstable. Test portfolio results under alternative scenarios and compare with Expected Shortfall backtesting.

Research sources