Cross-Market Volatility Spillovers in Asian Equity Markets: A Transformer-Based Multivariate GARCH Approach

We extend the dynamic-conditional-correlation GARCH framework with a transformer encoder to model time-varying volatility transmission across six Asian equity indices. The hybrid model captures both short-term shocks and longer-range dependencies that conventional MV-GARCH specifications miss. Empirical results across 2010-2023 show that the transformer-augmented model reduces one-day volatility forecast RMSE by 12.8% relative to DCC-GARCH and identifies asymmetric spillover regimes around the 2015 RMB devaluation and the 2020 COVID-19 shock.