Elicitability and its Application in Risk Management

TL;DR

This paper reviews the concept of elicitability for statistical functionals, explores conditions for scoring functions, and highlights its applications in risk management, notably showing that Value at Risk and Expected Shortfall are jointly elicitable.

Contribution

It provides a comprehensive overview of elicitability, including new examples and conditions, with a focus on its application to risk measures like VaR and ES.

Findings

Value at Risk and Expected Shortfall are jointly elicitable

Conditions for strictly consistent scoring functions are characterized

The paper includes new examples and applications in risk management

Abstract

Elicitability is a property of $\mathbb{R}^k$-valued functionals defined on a set of distribution functions. These functionals represent statistical properties of a distribution, for instance its mean, variance, or median. They are called elicitable if there exists a scoring function such that the expected score under a distribution takes its unique minimum at the functional value of this distribution. If such a scoring function exists, it is called strictly consistent for the functional. Motivated by the recent findings of Fissler and Ziegel concerning higher order elicitability, this thesis reviews the most important results, examples, and applications which are found in the relevant literature. Moreover, we also contribute our own examples and findings in order to give the reader a well-founded overview of the topic as well as of the most used tools and techniques. We include necessary and sufficient conditions for strictly consistent scoring functions, several elicitable as well as non-elicitable functionals and the use of elicitability in forecast comparison, regression, and estimation. Special emphasis is placed on quantitative risk management and the result that Value at Risk and Expected Shortfall are jointly elicitable.