Modelling the dynamic yield curve for the Canadian bond market

Abstract

The yield curve or the term structure of the interest has been one of the key leading macroeconomic indicators and forecasting the yield curve could provide vital information about future macroeconomic performance. Following the seminal works of Diebold & Li (2006a) and Diebold et. Al. (2006b)Many papers tried to evolve different techniques that can model the inner dynamics of the yield curve and forecast for future periods efficiently. This paper also followed the same path of Diebold & Li (2006) to model the inner dynamics of the Canadian zero-coupon yield curve and added some new structure following different stylized facts obtained from the factors. Primarily this paper estimated the model in a two-step way with the factor dynamics being VAR, VECM, EGARCH and DCC-EGARCH along with the one-step model using Kalman filter. A grid search is performed to calibrate the yield curve factors and using that all the models are compared using the root mean square forecast error. VECM and EGARCH turned out to be the best models revealing the different short term and long-term dynamics and the parsimonious nature of the model. Furthermore, a regime-switching model is also estimated to find the changing volatility structure of the yield curve as indicated in the previous models.