A0220
Title: A varying-coefficient model of expected shortfall and its application to mixed-frequency data
Authors: Jiangtao Wang - Huazhong Normal University (China) [presenting]
Abstract: The focus is to develop a nonparametric varying-coefficient approach for modelling the value-at-risk (VaR) and expected shortfall (ES) simultaneously since the ES is not elicitable but can be elicitable combined with VaR. Previous studies on conditional ES estimated only considered parametric model see-ups, which account for the stochastic dynamic of asset returns but ignore other exogenous economic variables and the investment situation. The approach overcomes this drawback and allows VaR and ES to be modelled directly in a flexible way using covariates that may be exogenous, especially sampled at different frequencies compared with the return series. A three-step procedure is developed based on the local linear smoothing technique for estimating the coefficient functions and establishing the consistency and asymptotic normality of the resultant estimator. To overcome the challenge associated with calculating the asymptotical variance, a random weight resampling approach is designed by perturbing the loss function directly. Simulation studies are presented to demonstrate the finite-sample performance of the proposed estimator. The favorable performance of the proposed method is further illustrated via an application for forecasting ES with mixed-frequency data.
