Regime-Based Factor Rotation Leveraging Sparse Jump Models

Abstract

This thesis explores integrating regime-switching signals into dynamic factor allocation. We apply a structured framework employing the Sparse Jump Model to identify regimes in financial markets and embed these insights into a Black–Litterman portfolio optimization. Our methodological contributions include the adoption of a Bayesian cross-validation framework for hyperparameter tuning, regime-based rebalancing dynamics, and novel allocation strategies. Through simulation studies tailored to reflect realistic market conditions, we conclude that Sparse Jump Models consistently outperform Hidden Markov and standard Jump Models. Empirically, we backtest this framework on U.S. equity-style factor ETFs from 2017 to 2025, where we find significant economic improvements in performance. The regime-aware strategies consistently deliver higher total returns, improved Sharpe ratios, and reduced drawdowns compared to benchmark strategies. However, the observed performance gains lack statistical significance. Therefore, we cautiously conclude that while regime signals demonstrate meaningful potential for enhancing factor allocations, practical implementation currently remains limited by structural constraints in the model’s design.