Style Rotation Revisited

Abstract

Style rotation strategies have enjoyed growing interest in the academic and practitioner communities over the last decades. This study investigates the ability of innovative modeling approaches to effectively forecast equity style performance. Single–multifactor logit models and several machine learning techniques are employed to generate directional style spread forecasts. Their efficacy is assessed in both a statistical and economic evaluation context. The analysis reveals that certain univariate logit models and machine learning techniques, such as naïve Bayes, bagging, Bayes generalized linear models, discriminant analysis models, and k-nearest neighbor, enhance the accuracy of the generated forecasts and lead to profitable investment strategies.

TOPICS: Style investing, security analysis and valuation, big data/machine learning, performance measurement

Key Findings

• ▪ In this study, the authors employ logit and supervised machine learning models to generate directional forecasts of equity style performance, employing a plethora of well-known predictors.

• ▪ The generated forecasts are evaluated in both a statistical and economic evaluation setting.

• ▪ Apart from the widely known and tracked value and size spreads, the study also investigates the predictability of the betting-against-beta spread.

• ▪ The analysis reveals that certain univariate logit and machine learning techniques enhance the accuracy of the generated forecasts and lead to profitable investment strategies.