Project Details
Projekt
Expected Shortfall Modeling: Advances for Cross-Sectional and Time Series Data
Subject Area
Statistics and Econometrics
Term
since 2025
Project identifier
Deutsche Forschungsgemeinschaft (DFG) - Project number 568876076
As a conditional tail mean, the Expected Shortfall (ES) has recently risen to prominence in time series analysis, largely due to its appearance in the Basel framework. Spawned partly by this, the ES also gained popularity in the microeconometric literature as a quantile-truncated expectation. The overarching methodological goal of this proposal is to refine the (time series and cross-sectional) modeling of the ES along several distinct, but related, dimensions. Specifically, the first aim is to devise new ES elicitation methods that yield more powerful forecast comparisons and better interpretability. Second, we will develop ES regressions at extreme risk levels, which are of particular interest in financial and macroeconomic risk management. Our third goal is to construct ES regressions that impede crossings with the auxiliary quantile regression. Such crossings regularly occur in empirical work, yet are nonsensical, so that avoiding them improves model interpretability and (quite likely) forecast performance. The fourth work package explores the use of adaptive learning rates in dynamic (score-driven) time series models for the ES. The aim is to achieve a better model fit, particularly in turbulent financial and macroeconomic times, where accurate risk forecasts are needed most. In light of these advances, the overall empirical objectives of the project are to improve ES forecasts together with their evaluation methodology as well as to enhance the estimation and interpretation of (dynamic) ES regressions.
DFG Programme
Research Grants
International Connection
Netherlands, Switzerland
Cooperation Partners
Dr. Tobias Fissler; Professor Rutger-Jan Lange, Ph.D.; Ramon de Punder, Ph.D.
