Stablecoin risk – a hybrid Copula-GARCH–QT framework for early warning, tail quantiles, and co-depeg dynamics

This study proposes a peg-centric risk framework for USD-denominated stablecoins, combining a Q-Transformer (QT) for next-day tail quantiles with a Copula–GARCH layer for cross-asset tail dependence. All inputs and outputs are expressed in basis points around a one-dollar anchor for like-for-like comparisons across DAI, TUSD, USDC, USDP, and USDT since 2020. In addition, this study designs a family of seven peg-sensitive indicators tailored to stablecoin market mechanics – spot peg deviation, 20-day band width, intraday/high–low volatility, volume–band resonance, peg mean-reversion half-life, deviation-based momentum, and asymmetric downside depth–so that signals align naturally with VaR scale. Empirically, QT produces VaR paths that remain compact in calm periods and rise smoothly and decisively as stress builds, improving timing and readability relative to ARIMA, GBR, GRU, LSTM, and QREG. Event-aligned views show stable thresholds around exceedances and warnings emerging about three trading days in advance. Grouped SHAP reveals a consistent mechanism shaped by the above indicators: skewness and deviation persistence raise next-day tail risk; faster mean reversion suppresses it; volatility–volume resonance amplifies reach; spot deviation and band width strengthen when co-moving with the dominant drivers. Regime-split results confirm that these roles intensify under stress, supporting calibrated, interpretable VaR for both single-coin monitoring and basket-level assessment.