Examining the Forecasting Performance of a Modified Affine Model with Macroeconomic and Latent Factors

Abstract

Various studies model the dynamics of the yield curve assuming that some of the yields are measured without error but this methodology lacks economic interpretation. We overcome this problem by estimating a modified affine model with macroeconomic and latent factors which introduces measurement noise on both yields and macroeconomic determinants. Our results suggest that under the proposed model there is a significant reduction in the persistence of the latent factors and an increase in the effect of macroeconomic shocks to the entire yield curve. We provide a comparative analysis of these models, and we conduct out of sample comparative forecasts to investigate if our specification has a superior performance. We find important differences concerning the magnitude of the dynamics that move the yield curve. Our model provides better forecasts for the entire yield curve while it also beats random walk in many cases. This is an important finding since according to the relative literature it is very difficult for any affine model to outperform random walk.

Keywords: Affine models, Yield curve, Kalman filter, Out of sample forecasting

JEL Classification: C32, E43, E52