Abstract
This paper proposes and implements a parsimonious three-factor model of the term structure whose dynamics is driven uniquely by observable state variables. This approach allows comparing alternative views on the way state variables – macroeconomic variables, in particular – influence the yield curve dynamics, avoids curse of dimensionality problems, and provides more reliable inference by using both the cross-sectional and the time series dimension of the data. I simulate the small-sample properties of the procedure and conduct in- and out-of-sample studies using a comprehensive set of US data. I show that even a parsimonious model where the level, slope and curvature factors of the term structure are driven by, respectively, inflation, monetary policy and economic activity consistently outperforms the (latent-variable) benchmark model in an out-of-sample study.