Understanding the Predictability of Excess Returns

A seminal paper by Fama and Bliss (1987) showed that a simple regression model could explain a significant portion of 1-year ahead excess returns. Cochrane and Piazzesi (2005) showed that their regression model could explain a significantly larger por tion of excess returns than Fama and Bliss"model and that a single return-forecasting factor essentially encompassed the predictability of excess returns for all of the bonds considered. This paper makes several contributions to the literature. First, I show why excess return models based solely on bond prices are unlikely to provide information about the predictability of excess returns and, in so doing, show that neither FB"s model nor CP"s model provides information about the predictability of excess returns. Second, I show that the "predictive power" of FB"s model is due solely to the high correlation between excess returns and changes in bond prices, and that this correlation accounts for half of the "predictability" reported by CP. Third, I show that forecasting excess returns out of sample is identical to forecasting future bond prices. Consequently, the FB and CP models can be compared with any model that forecast future bond prices (or, equivalently, bond yields).