Tuesday, May 6, 2014 | 2:00pm - 3:30pm | Room 335, HSBC Business School Building

Abstract

Since Diebold and Li (2006) showed the outstanding performance of a dynamic Nelson-Sieglel model (DNSM) in forecasting the yield curve, the DNSM has been widely used in many macro and finance area. Because of its parsimonious but flexible model specification the Bayesian model-averaging method based on the Bayes factor typically gives a weight of nearly one on the DNSM excluding a standard arbitrage-free affine term structure model (ATSM). Nevertheless, the ATSM has been also commonly used because it provides plenty of economically interpretable outcomes such as term premium and model-implied term structure of real interest rates. Meanwhile, the random-walk (RW) is often used as a benchmark in out-of-sample forecasting comparison. Despite the popularity of these three frameworks, none of them dominates the others across all maturities and forecast horizons. This fact indicates that those models are potentially misspecified. In this paper we investigate whether combining the possibly misspecified models in a linear form suggested by Geweke and Amisano (2011) and Waggoner and Zha (2012) help improve the predictive accuracy. For this we compare out-of sample prediction performance from the merged models with a constant model weight with those of the three individual prediction models and the merged models with a Markov-switching model weight for eight different maturities and forecast horizons of 1, 3, 6 and 12 months. We find that overall the constant mixture model is most supported. In particular, the constant mixture model consistently forecasts better than the individual prediction models across all maturities and forecast horizons.(JEL G12, C11, F37)