Market regime detection via realized covariances

Abstract

Identifying market regimes is crucial for asset pricing and portfolio management. Within efficient markets, the macroeconomic conditions drive the demand for risky assets. Consequently, the transitions between different regimes are reflected into covariance matrices, whose time-varying coefficients react to unexpected news. Accordingly, we identify market regimes by feeding latent information embedded in the covariances to various regime-switching models and an unsupervised learning methodology. The advantage over existing methods is that our approach considers all information in the covariances to detect regimes while allowing for smooth and abrupt regime changes. We display each model's ability to correctly detect regimes through a simulation study and by evaluating a regime-switching investment strategy. Our results point to hierarchical clustering as the best-performing model for labelling market regimes with both simulated and observed data. Furthermore, we find that regime-switching models based on an observable transition variable perform well during overall periods of stress.