A differential evolution algorithm for yield curve estimation

Abstract

Modeling the term structure of government bond yields is of great interest to macroeconomists and financial market practitioners. It is crucial for bonds and derivatives pricing, risk management, and reveals market expectations, which is essential for monetary policy decisions. This paper suggests the use of a differential evolutionary algorithm to estimate yield curves for US Treasury bonds. It considers parsimonious modeling to avoid non-convergence and high instability of traditional optimization algorithms when estimating model parameters caused by the choice of their initial values during curve fitting. In this approach, the whole yield curve for different maturities is obtained by models parameters estimates. Computational experiments show that the differential evolutionary algorithm provides more accurate yield curves than the ones derived by nonlinear least squares and genetic algorithm approaches.