A Unified HJM Approach to Non-Markov Gaussian Dynamic Term Structure Models: International Evidence

Abstract

Motivated by an extensive literature showing that government bond yields exhibit a strong non-Markov property, in the sense that moving averages of long-lagged yields significantly improve the predictability of excess bond returns. We then develop a systematic approach of constructing non-Markov Gaussian dynamic term structure models (GDTSMs) under the Heath-Jarrow-Morton (HJM) framework. Compared to the current literature, our approach is more flexible and parsimonious, enabling us to estimate an economically significant non-Markov effect that helps predict excess bond returns both in-sample and out-of-sample.