Predicting Style Returns with Time-Varying Sensitivities

This paper analyzes the predictability of different style portfolio returns (a.k.a. Fama-French factors) considering time-varying sensitivities of these returns to different macroeconomic variables and own momentums. Styles, as used in this paper, can be defined as groups of securities with a common characteristic, such as value (Graham and Dodd (1934)) and size (Banz (1979)), and have been popularized by Fama and French papers. This paper specifically looks at determinants of style investing, such as style momentum and predictor variables such as macroeconomic variables (e.g. yield spread, inflation, industrial production, etc.), and show how 'learning' about these variables affects the predictability of different style portfolio returns compared to models where there is no learning (e.g. linear models). A time-varying parameter model and a Kalman filter are used to take into account the effect of learning in this paper. At the end, it is found that returns on style portfolios such as value and size appear to be related, with time-varying sensitivities, to yield spread and other macroeconomic variables. This paper also finds that time-varying parameter models provide better in-sample and out-of-sample predictions then simple benchmark constant parameter models