Regime-Aware Factor Allocation with Optimal Feature Selection

This is the authors' abstract. We don't add key points for this paper.

In modern portfolio management, adapting to dynamic market conditions poses a significant challenge for investors seeking optimal risk-adjusted returns. Traditional static allocation strategies, rooted in modern portfolio theory and factor investing, often fail to capture the nuanced dynamics of changing regimes. This article presents a novel approach, regime-aware multifactor allocation with optimal feature selection. The goal is to optimize single-factor performance in response to changing regimes through the recently developed statistical jump model. The authors independently identify regimes over each of their factors by fitting a two-state jump model biannually and construct a multifactor investment portfolio. The authors' optimal feature selection methodology, whereby they remove the assumption of stationarity, allow for a temporally adjusting input feature set. The article extends prior work by showing (a) a regime-aware single-factor strategy outperforms a regime-agnostic single-factor strategy, (b) a regime-aware multifactor strategy outperforms a regime-agnostic multifactor strategy, and (c) an optimal feature selection drastically improves temporal regime identification and outperforms a fixed feature set. Through empirical out-of-sample analysis, the authors demonstrate the efficacy of the framework over six primary long-only equity factors. Their findings contribute to the growing body of research on regime-switching investment models, providing portfolio managers with a robust framework for navigating dynamic market conditions and enhancing portfolio performance.