Forecasting the U.S. Treasury Yield Curve using Targeted Diffusion Indices

Abstract

We investigate possible empirical linkages between variation in the U.S. Treasury yield curve and several measures of economic and financial activity by the methodology targeted diffusion index forecasting. First, we model the entire yield curve with the Nelson-Siegel exponential components framework period-by-period, thereby distilling the yield curve into three, dynamic parameters. We show that these three parameters can be interpreted as yield curve factors corresponding to level, slope and curvature, and that their variation explain almost all yield curve variation. We then use targeted diffusion indices estimated from a set of 1196 different macroeconomic and financial variables to produce both in-sample and out-ofsample forecasts these three parameters, thus obtaining forecasts of the the entire yield curve. While we do find in-sample predictability of the Nelson-Siegel dynamic paramaters by the targeted diffusion indices, we do not find that they are able to produce better out-of-sample forecasts than the competitor models. Additionally, we find that the established Diebold-Li yield curve forecasting model, which has previously been found to produce superior forecasts, is outperformed by a simple random walk model. Our findings on a new, updated sample thus contradict earlier findings.