Risk measures with the CxLS property (Q287670)

The authors characterise risk measures that have convex level sets at the level of distributions (CxLS) which means that \[ \rho(F)=\rho(G)=\gamma\quad \Rightarrow\quad \rho(\lambda F+(1-\lambda)G) = \gamma, \qquad \gamma\in(0,1), \] where \(F\) and \(G\) are probability distributions and \(\rho\) is a risk measure. The financial interpretation of this property is that any mixture of two equally risky positions remains with the same risk. The authors show that this risk measures can be identified with a class of generalized shortfall risk measures and based on this result confirm that expectiles are the only elicitable coherent risk measures. They also provide characterization of robustness for convex risk measures in terms of a week notion of mixture continuity.