Multi-Scale Markov-Switching GARCH: Volatility Regime Detection in EUR/USD

Battle Count: 71

Highly relevant to quantitative trading in FX markets. The framework provides statistically validated volatility regime detection (DM test p=1.28e-6, KS test p=1.35e-153) with a multi-scale architecture that captures macro, meso, and micro dynamics simultaneously. The 27-expert MoE routing and Shannon entropy filter are directly applicable to regime-aware trading systems. However, the paper explicitly positions itself as a risk management tool rather than an alpha engine (IC=+0.0252), and the walk-forward validation across 16 diverse quarters (2021-2024) demonstrates robustness. The cross-scale probability tensor and regime disagreement metrics offer novel signal construction for systematic FX strategies.

High implementation complexity due to: (1) three independent AR(1)-MS-GARCH models with TVTP requiring Numba-JIT compiled Hamilton filter kernels; (2) 27-dimensional joint probability tensor construction via outer products; (3) 27-expert Mixture-of-Experts architecture with RidgeCV models and data-sparse fallback logic; (4) composite TVTP driver construction with multi-layer causal isolation; (5) rolling walk-forward analysis with warm-start optimization across 16 quarters; (6) Shannon entropy filtering; (7) penalized MLE with multiple penalty terms (stickiness, ordering, stationarity); (8) staggered parameter bounds and label-switching resolution. Requires expertise in HMMs, GARCH modeling, numerical optimization, and FX microstructure.

The paper provides detailed mathematical formulations, parameter specifications, transition matrices, and a comprehensive validation scorecard. However, no code repository is mentioned. The methodology relies on Numba-JIT compiled Hamilton filter kernels, L-BFGS-B optimization, and specific feature engineering (composite meta-signals, TVTP drivers). The walk-forward protocol is well-described with 16 quarters and rolling 5-year windows. Reproduction would require implementing the multi-scale architecture, TVTP specification, 27-expert MoE routing, and entropy filtering from scratch. Data source (EUR/USD 1-minute mid-price) is not explicitly identified as publicly available.