Nonparametric Smoothing of Yield Curves

This paper proposes a new nonparametric approach to the problem of inferring term structure estimates using coupon bond prices. The nonparametric estimator is defined on the basis of a penalized least squares criterion. The solution is a natural cubic spline, and the paper presents an iterative procedure for solving the non-linear first-order conditions. Besides smoothness, there are no a priori restrictions on the yield curve, and the position of the knots and the optimal smoothness can be determined from data. For these reasons the smoothing procedure is said to be completely data driven. The paper also demonstrates that smoothing a simple transformation of the yield curve greatly improves the stability of longer-term yield curve estimates.