Time-Series and Cross-Section Variation of Bond Risk Premia

A long-term bond that is sold before its maturity has an uncertain excess return over the certain return from a shorter-term risk-free bond maturing at the time of sale. There is strong evidence that this excess return is predictable and that the predicted excess return, or risk premium, is time varying. Cochrane and Piazzesi (2005) use five forward rates to predict excess returns on one to five-year maturity bonds. I extend and improve their approach in multiple aspects. Firstly, the data set for the US market is expanded in two dimensions: cross-section and time series. In the cross-section dimension, one- to twenty five-year maturity bonds are incorporated. In the time series dimension, the data is extended to December 2012. Including more recent data elucidates a structural break in the model in-line with the shifts in Federal Reserve strategies for monetary policy. Additionally, I apply a parsimonious variant of their model to nine global fixed income markets: Canada, the United Kingdom, Germany, Norway, Sweden, Switzerland, Japan, Australia, and New Zealand. Furthermore, I investigate the out-of-sample predictability of bond excess return in all ten markets, and I propose a superior specification to forecast risk premium for US market. My findings indicate that a variant of Cochrane and Piazzesi (2005) model has a reasonable in-sample and out-of-sample forecasting power for the entire cross-section of future bond returns