Abstract:
In the recent years, the notion of elicitability has gained considerable attention, especially in the discussion about the choice of a good risk measure and the prominent debate between Value at Risk (elicitable) and Expected Shortfall (not elicitable). In a nutshell, elicitability is concerned with the existence and optimal choice of a `good’ loss function. This talk gives a concise introduction to the concept of elicitability, presents the most relevant research questions related to elicitability, and shows its relevance in the fields of M-estimation, regression, and especially forecast comparison. In the latter framework, elicitability of a risk measure leads the way to comparative backtests. Theoretical considerations are illustrated with a special emphasis on financial applications focusing on the pair (Value at Risk, Expected Shortfall). The talk closes with some outlooks on possible extensions, including set-valued measures of systemic risk.