A1248
Title: Quantile-based modeling of scale dynamics in financial returns for value-at-risk and expected shortfall forecasting
Authors: Xiaochun Liu - University of Alabama (United States)
Richard Luger - Laval University (Canada) [presenting]
Abstract: The purpose is to introduce a new approach for forecasting value-at-risk (VaR) and expected shortfall (ES) by modeling the dynamics of the conditional scale of financial returns. Focusing on downside market risks, the conditional scale is defined as the difference between key quantiles of the return distribution, which are modeled using conditional autoregressive VaR (CAViaR) specifications. VaR is obtained by estimating the left-tail quantiles of the rescaled returns, while ES is approximated by averaging these quantiles over a range of levels below the VaR threshold, providing robust and distribution-free estimates of potential extreme losses. Simulation experiments show that this approach markedly improves the accuracy of VaR and ES forecasts, particularly in scenarios involving skewness, heavy tails, and leverage effects. The method consistently outperforms several established models, including GARCH and joint VaR and ES conditional quantile models. An empirical application using daily returns of major international stock market indices further demonstrates the model's effectiveness in accurately forecasting risk over the period studied, which includes the recent COVID-19 pandemic.
