This article proposes a model and estimation method that facilitates the inclusion of a gen-eralized structure of observable, forward-looking factors into a dynamic term structure model. We illustrate our framework using multi-horizon survey forecasts of in ation, output growth and monetary policy where the dynamics of the forecasts are modeled jointly with the physical process driving their realizations. We extend this framework to allow for multiple-horizon forecasts to drive the short rate, yielding its novel interpretation as a forward-looking multiple-horizon monetary policy rule, which facilitates a decomposition of monetary policy and the yield curve into short and long horizon expectations. Our method eectively addresses the time-varying forecast horizon issue often found in survey forecasts by extracting xed-horizon forecasts that are consistent with their underlying P-dynamics. While short horizon forecasts can pin down the short rate, long horizon forecasts embed information that better describes longer maturity yields. In addition, although short horizon forecasts of real output growth are hidden in the cross section of yields, long horizon forecasts are strongly related to the yield curve's slope. We conclude by exploring the models' implications for bond risk premia.
(JEL G12, E37, E43, E44, E52, C13)

