[Paper Review] Explainable Regime Aware Investing

Amine Boukardagha|arXiv (Cornell University)|Feb 21, 2026

Stock Market Forecasting Methods0 citations

TL;DR

The paper proposes a strictly causal Wasserstein Hidden Markov Model (Wasserstein HMM) to infer regime dynamics and embed them into a transaction-cost-aware mean–variance optimization, achieving higher risk-adjusted performance and lower drawdowns versus benchmarks.

ABSTRACT

We propose an explainable regime-aware portfolio construction framework based on a strictly causal Wasserstein Hidden Markov Model. The model combines rolling Gaussian HMM inference with predictive model-order selection and template-based identity tracking using the 2-Wasserstein distance between Gaussian components. This allows regime complexity to adapt dynamically while preserving stable economic interpretation. Regime probabilities are embedded into a transaction-cost-aware mean-variance optimization framework and evaluated on a diversified daily cross-asset universe. Relative to equal-weight and SPX buy-and-hold benchmarks, the Wasserstein HMM achieves materially higher risk-adjusted performance with Sharpe ratios of 2.18 versus 1.59 and 1.18 and substantially lower maximum drawdown of negative 5.43 percent versus negative 14.62 percent for SPX. During the early 2025 equity selloff labeled Liberation Day, the strategy dynamically reduced equity exposure and shifted toward defensive assets, mitigating peak-to-trough losses. Compared to a nonparametric KNN conditional-moment estimator using the same features and optimization layer, the parametric regime model produces materially lower turnover and smoother weight evolution. The results demonstrate that regime inference stability, particularly identity preservation and adaptive complexity control, is a first-order determinant of portfolio drawdown and implementation robustness in daily asset allocation.

Motivation & Objective

• Motivate regime-aware investing to address non-stationarity in daily portfolio allocation.
• Develop a strictly causal Wasserstein HMM with adaptive model-order selection for regime inference.
• Introduce Wasserstein template tracking to preserve regime identity over time.
• Integrate regime probabilities into a transaction-cost-aware mean–variance optimization framework.
• Evaluate performance against passive benchmarks and a non-parametric baseline to isolate the economic role of regime inference.

Proposed method

• Use strictly causal rolling Gaussian HMMs with expanding windows and one-step-ahead predictive log-likelihood for dynamic model-order selection.
• Map HMM components to persistent regime templates via 2-Wasserstein distance to preserve regime identity.
• Aggregate template-probabilities to form mixture moments (mean and covariance) for optimization.
• Solve a transaction-cost-aware mean–variance optimization with penalties on turnover and constraints on weights.
• Benchmark parametric Wasserstein HMM approach against a non-parametric KNN baseline using identical features and optimization layer.

Experimental results

Research questions

• RQ1 Does the parametric Wasserstein HMM provide more stable regime identification than non-parametric baselines in a daily rolling setting?
• RQ2 How does adaptive regime complexity control impact turnover and drawdowns in daily cross-asset allocation?
• RQ3 What is the economic impact of regime-conditioned allocations during market stress periods?
• RQ4 Do Wasserstein-based regime identities improve interpretability and stability of portfolio weights over time?

Key findings

• Parametric regime investing yields higher out-of-sample Sharpe ratio (2.18) than equal-weight (1.59) and SPX Buy & Hold (1.18).
• Maximum drawdown is substantially smaller for the parametric approach (-5.43%) versus SPX (-14.62%).
• During Liberation Day, the strategy reduced equity exposure and shifted toward defensive assets, mitigating losses.
• Compared with KNN, the Wasserstein HMM approach produces dramatically lower turnover and smoother weight evolution.
• Regime identity stability via Wasserstein templates is a first-order determinant of portfolio drawdown and implementation robustness.