Comparing forecasting ability of parametric and non-parametric methods: an application with Canadian monthly interest rates

Abstract

The primary objective of this article is to compare the forecasting ability of some recent parametric and non-parametric estimation methods by using monthly Canadian interest rate data between 1964:1–1999:1. The two-factor continuous time term structure model of Brennan and Schwartz was estimated where the first factor represents the short rate and the second factor the long rate using the continuous time estimation procedures developed by Bergstrom. The interest rates using the multivariate GARCH model developed by Engle and Kroner, and two non-parametric estimation methods namely, non-parametric kernel smoothing and the artificial neural networks was modelled. For the short-term rates, it has been found that, the Bergstrom's method and the artificial neural networks model have marginally better forecasting performance than that of the linear benchmark. For the long-term rates, none of the methods produced better forecasting precision than that of the benchmark.