Reconciling the Return Predictability Evidence under Structural Breaks

Abstract

This study shows that the poor out-of-sample performance of the realtime adjusted dividend-price ratio reported in Lettau and Nieuwerburgh (2008) is mainly a result of the gap period between the occurrence of a break and its detection, which implies that the poor out-of-sample performance of the adjusted dividend-price ratio is due to the requirement in Bai and Perron’s (1998) procedure that breaks must be away from the boundaries of the sample. A substantial improvement in the out-of-sample performance of the adjusted dividend-price ratio during the gap period is shown with the use of Andrews’s (2003) procedure in the real-time adjustment of the dividend-price ratio. The newly suggested procedure for the adjusted dividend-price ratio in this study has better out-ofsample performance than the simple sample mean, although it is not significant.