Abstract

Estimating 19th century US federal bond yield curves involves challenges because few bonds were traded, bonds had peculiar features, government policies changed often, and there were wars. This paper compares statistical approaches for confronting these difficulties and shows that a dynamic Nelson-Siegel model with stochastic volatility and bond-specific pricing errors does a good job for historical US bond prices. This model is flexible enough to interpolate data across periods in a time-varying way without over-fitting. We exploit new computational techniques to deploy our model and estimate yield curves for US federal debt from 1790-1933.