The Economic Origin of Treasury Excess Returns: A Cycles and Trend Explanation

In this paper we try to understand the economic explanation of the difference in predictability afforded by the old and the new-generation return-predicting factors. To do so, first we show that the Cieslak-Povala (2010) approach can be expressed in terms of a conditional prediction of where the level and the slope of the yield curve should be, given long-term inflation. We then explore whether this interpretation is valid, or whether, as Cochrane (2015) argues, the Cieslak-Povala factor simply owes its effectiveness to its acting as a de-trender. We answer this question by decomposing excess returns into low- and high-frequency components; by showing that the old and new return-predicting factors capture very different periodicities of the return power spectrum; and by showing that a high speed of mean-reversion is required for the high-frequency part of the spectrum. We conclude that creating strongly mean-reverting cycles is key to predicting excess returns effectively, and explore to what extent the Cieslak-Povala approach may be 'special' in this respect. We give a financial interpretation to the low- and high-frequency sources of excess returns, and, based on the understanding this decomposition affords, we show how to build almost by inspection a whole class of extremely parsimonious, robust and financially-motivated return-predicting factors which forecast in- and out-of-sample returns as well or better than factors built using many more variables.