A1952
Title: Joint tail risk forecasting for high-volatility assets based on copula semiparametric framework
Authors: Aaron E-P Lin - Feng Chia University (Taiwan) [presenting]
Cathy W-S Chen - Feng Chia University (Taiwan)
Takaaki Koike - Hitotsubashi University (Japan)
Abstract: A copula-based bivariate ES-CAViaR framework is proposed to jointly forecast value-at-risk (VaR) and expected shortfall (ES), while investigating downside dependence between financial assets. Four ES-CAViaR-type marginal specifications are considered, including the baseline ES-CAViaR model, ES-CAViaR-st with a scaled exceedance ratio, ES-CAViaR-v incorporating Rogers-Satchell volatility, and ES-CAViaR-R incorporating Garman-Klass volatility. To model cross-asset dependence, the Clayton and BB1 copulas are employed based on pseudo-observations. The Clayton copula is designed to capture lower-tail dependence and joint downside risk, whereas the BB1 copula accommodates both lower- and upper-tail dependence, allowing for greater flexibility in modeling asymmetric tail behavior. Model parameters are estimated using a Bayesian adaptive Markov chain Monte Carlo (MCMC) framework that combines random-walk Metropolis and independent-kernel Metropolis-Hastings sampling. An empirical application to Bitcoin and Ethereum is conducted using rolling-window one-step-ahead forecasts at the 1\% risk level over a five-year evaluation period. Forecast performance is evaluated using violation-based backtests and scoring criteria. The empirical results indicate that the scaled exceedance ratio specification provides more robust joint VaR and ES forecasts, especially during periods of elevated downside risk.
