Abstract
This paper estimates a term premium from an estimated DSGE model under adaptive learning (AL) with arbitrary beliefs. The estimated AL term premium shares important features with the corresponding measures obtained in the literature using no-arbitrage affine term structure models. Thus, the AL term premium is sizable, highly persistent, and countercyclical. The estimation results suggest that AL, by introducing much more economic uncertainty, presents an additional channel for explaining the bond premium. This feature can be viewed as an alternative to other features needed to generate a sizable term premium under rational expectations, such as a huge risk aversion coefficient. Shocks associated with wage markup, investment, and monetary policy explain roughly 75% of AL term premium fluctuations, but their relative contributions change quite dramatically with the forecast horizon used in the variance decomposition.