Implications of Return Predictability across Horizons for Asset Pricing Models

We use the evidence on predictability of returns at diff erent horizons to discriminate among competing asset pricing models. Speci cally, we employ predictors-based variance bounds, i.e. bounds on the variance of the Stochastic Discount Factors (SDFs) that price a given set of returns conditional on the information contained in a vector of return predictors. We show that return predictability delivers variance bounds that are much tighter than the classical, unconditional Hansen and Jagannathan (1991) bounds. We use the predictors-based bounds to discriminate among three leading classes of asset pricing models: rare disasters, long-run risks and external habit. We nd that the rare disasters model of Nakamura, Steinsson, Barro, and Ursua (2013) is the best performer since it satis es rather comfortably the predictors-based bounds at all horizons. As for long-run risks, while the classical version of Bansal and Yaron (2004) is the model most challenged by the introduction of conditioning information since it struggles to meet the bounds at all horizons, the more general version of Schorfheide, Song, and Yaron (2016), which accounts for multiple volatility components, satisfi es the 1- and 5-year bounds as long as the set of test assets includes only equities and T-Bills. The Campbell and Cochrane (1999) habit model lies somehow in the middle: it performs quite well at our longest 5-year horizon while it struggles at the 1-year horizon. Finally, when the set of test assets is augmented with Treasury Bonds, the only model that is able to satisfy the predictors-based bounds is the rare disasters model