Multi-period Euler-equation learning and term structure

This paper builds on the Euler-equation approach to adaptive learning by introducing the term structure of interest rates into a medium-scale DSGE model, where bond yields are priced with separate Euler equations. Term structure information enables us to characterize agents’ forecasting models using only term spread information available at the time when expectations are formed. Our estimated DSGE model under adaptive learning substantially improves the model fit to the data, including both macroeconomic and yield curve data, compared to the rational expectations version. The out-of-sample forecasting performance also increases under our learning approach. Despite large modeling differences, the estimated non-standard term-risk premium from our adaptive learning model resembles those term premia estimated using no-arbitrage affine term structure models. The model fit gain obtained assuming adaptive learning with term structure information instead of rational expectations largely increases when disciplining model expectations with forecasts from the Survey of Professional Forecasters.