文摘
Scope and method of study. The study compares the parametric pricing approach to the modeling of the term structure of interest rates, with the non-parametric approach using artificial neural networks. The parametric pricing models include the Cox Ingersoll and Ross (1985a, b) model, Longstaff and Schwartz (1992) model, and restricted Heath Jarrow and Morton (1992) models (Ritchken and Sankarasubramanian (1995)). The network models include the multi layer perceptron and the radial basis function network. The study compares the different models on two grounds, first in terms of their ability to price zero coupon bonds and second in terms of their ability to hedge the underlying risk factors.;Findings and conclusions. The MLP network and the HJM-RS model are found to have significantly lesser pricing and hedging errors than the other models. The MLP and HJM-RS models capture different facets of the term structure such as steepness and curvature better than the principal components analysis model and the traditional duration measures. The principal components analysis models are more sensitive to shocks in the yield curve than the other models considered in the study.