Explainable Regime-Aware Investing

Battle Count: 80

Directly addresses daily cross-asset portfolio construction with regime-aware allocation, transaction cost management, and drawdown control. The framework is designed for operational deployment in quantitative trading with strict causality, adaptive complexity, and interpretable regime labels. The empirical results (Sharpe 2.18, max DD -5.43%, turnover 0.0079) are highly relevant for practitioners building systematic allocation strategies. The comparison with KNN baseline and passive benchmarks provides actionable implementation guidance.

Requires implementing: (1) Gaussian HMM estimation with EM algorithm on rolling windows, (2) predictive model-order selection with validation slices, (3) closed-form 2-Wasserstein distance computation between multivariate Gaussians, (4) template tracking with exponential smoothing, (5) Ledoit-Wolf covariance shrinkage, and (6) constrained mean-variance optimization with L1 transaction cost penalty. The mathematical components are well-defined but the engineering of rolling re-estimation, numerical stability of matrix square roots in Wasserstein distance, and optimization solver configuration add complexity. Pseudocode is provided but no reference implementation.

The paper provides detailed pseudocode (Algorithm 1 and 2), explicit mathematical formulations, and uses publicly available data from Yahoo Finance. However, no GitHub repository or code is provided. Hyperparameters (risk aversion gamma, transaction cost tau, max weight, smoothing rate eta, validation slice length, complexity penalty) are mentioned but exact values are not fully specified. The HMM estimation library and optimization solver are not explicitly named.