Sector Rotation through the Business Cycle: A Machine Learning Regime Approach

Abstract

Sector returns should theoretically differ during business cycle regimes. The notion of cyclical and defensive sectors is clearly established among practitioners and academics alike. On the other hand, the persistence, now- and forecastability of business cycles has been documented by a vast amount of literature. This study tests whether both strands can be merged to construct an investable sector rotation strategy based on the analysis of macroeconomic data. I find that both relationships hold: If one has forward looking information about GDP, outperformance from sector rotation is possible. Furthermore, one can nowcast the current position in the business cycle with some accuracy. While nowcasting accuracy is too small to translate into constant outperformance, the value of the examined methodology lies in the timely identification of major economic crises and provides economically superior performance by significantly reducing drawdowns during such.

Keywords: Nowcasting, Machine Learning, Sector Rotation, Random Forest, Vintage Data, Drawdowns

JEL Classification: G11, E23, E27