Abstract
Expected Shortfall (ES) has been widely accepted as an intuitive coherent alternative to Value-at-Risk (VaR). However, recently ES has been found not to be elicitable. This means that backtesting ES is less straightforward than backtesting elicitable risk measures like VaR or expectiles. We revisit the properties of ES and expectiles in order to assess which is the more suitable risk measure for practical purposes. We find that expectiles lack an intuitive interpretation in terms of solvency standards and are not comonotonically additive. They may hence see diversification where there is none. We observe that ES can be backtested despite not being elicitable and, therefore, conclude that ES still should be the first choice as a risk measure.
[PDF] Slides of the presentation.