Loss Functions for Forecasting Treasury Yields

Abstract

Recent advances in the term structure literature focus on model specification and estimation. While forecasting is considered important, it is not explicitly taken into account at the estimation stage. We estimate affine term structure models by aligning in-sample and out-of-sample loss functions. Aligning loss functions provides substantial improvements in forecasting performance, especially for long forecast horizons. The resulting parameter estimates imply factors that differ from traditional term structure factors, especially for curvature. This suggests that the improvement in fit results from identification of the third factor, which captures information hidden to conventional loss functions, consistent with Cochrane and Piazzesi (2005).