E-Backtesting Expected Shortfall: What Defines a “Good” Forecasting Method for Chinese Regulators?

Following the implementation of Basel IV, China’s financial regulators have replaced Value-at-Risk (VaR) with Expected Shortfall (ES) as the standard market risk measure, necessitating regulatory-oriented evaluation of ES forecasts. This study examines what constitutes a prudent ES forecasting method using e-backtesting, a sequential and model-free evaluation framework designed for regulatory monitoring. We evaluate 11 forecasting methods, including parametric, semiparametric, empirical, and deep-learning models, across four asset classes and four portfolio strategies in the Chinese market under Basel IV-consistent settings. Results show that parametric and semiparametric candidates exhibit clustered backtesting detections and increased computational burden around major market regime shifts, whereas the deep-learning model demonstrates greater resilience and produces more conservative ES forecasts during turbulent periods. These findings suggest that robustness to regime shifts should be considered a key criterion in the regulatory evaluation of ES forecasting models in the Chinese market.