The Ex Ante Predictive Accuracy of Alternative Models of the Term Structure of Interest Rates

This paper compares six term structure estimation methods empirically in terms of zero and forward rate curves as well as ex ante price and yield prediction accuracy. Specifically, we use daily government bond quotations to generate true out-of-sample prediction errors based on the model's ability to price bonds for one to five trading days ahead. Using several criteria, we find that the models' performance differs markedly between in- and out-of-sample predictions. Particularly, models with relatively smooth yield and forward rate curves do not perform well in sample but produce the best out-of-sample forecasts. Given that the estimation methods vary in computational complexity and parsimony, we examine the degree of loss in accuracy the modeler incurs by not using the best method. Of particular interest is the comparison between the more complex splining methods and the parsimonious Nelson-Siegel model estimated from Treasury STRIPS with only three parameters. Finally, we examine the models' performance in various maturity ranges which may be of interest to various types of investors.