Return Predictability: A Multi-resolution Analysis

摘要

We propose a general multi-resolution analysis approach to the return predictability problem. Starting with the finding that predictive power varies across timescales of a predicting variable, we identify those components in dividend price ratio that have significant predictive power and show if other components are removed, this 'filtered' dividend price ratio can produce remarkable evidence for its predictive power for stock returns over our sample period 1953-2000. Surprisingly, significant (in sample and out of sample) predictive power is also found over 1990-2000 subperiod, again, if one appropriately filters this predictor. Thus, our result suggests, contrary to existing evidence of no predictive power after 1990 from dividend price ratio, the opposite may be true. Another insightful finding of our empirical analysis is that at certain time-scale level the relation between dividend price ratio and stock return can be negative. This finding is interestingly related to Menzly et al. (2004) and assigns a new interpretation to their observation. Overall, we provide an alternative way to think about return predictability which could potentially allow us to exploit more important information from predicting variables. While the paper focuses on SP500 index, a full set of analysis is also applied to CRSP-EW data.