Robust Rolling Regime Detection (R2-RD): A Data-Driven Perspective of Financial Markets

Abstract

The nonstationary and high-dimensional nature of financial markets poses significant challenges for navigation. Temporally stable regime classification offers a perspective to manage these challenges. We propose the Robust Rolling Regime Detection (R2-RD) framework that adaptively retrains with streaming data and employs temporal ensemble, label assignment, and threshold policies to address temporal instability resulting from nonstationarity, model mismatches, etc. Further, the R2-RD framework’s data-driven model selection procedure chooses the model that best describes the data from the wide variety of latent variable models. We demonstrate the application and ease of extensions of R2-RD via two different datasets: macroeconomic and futures markets. Numerical experiments also illustrate how different macroeconomic regimes separate the performance of mutual funds, allowing for regime-aware asset management. The findings make R2-RD an ideal support for data-driven decision models, and the implementations can be leveraged across investments, credit, risk, policies, etc.

Keywords: Regime Detection, Asset Management, Hidden Markov Model