Estimating Affine Multifactor Term Structure Models Using Closed-Form Likelihood Expansions

Abstract

We develop and implement a technique for maximum likelihood estimation in closed-form of multivariate affine yield models of the term structure of interest rates. We derive closed-form approximations to the likelihood functions for all nine of the Dai and Singleton (2000) canonical affine models with one, two, or three underlying factors. Monte Carlo simulations reveal that this technique very accurately approximates true maximum likelihood, which is, in general, infeasible for affine models. We also apply the method to a dataset consisting of synthetic US Treasury strips, and find parameter estimates for nine different affine yield models, each using two different market price of risk specifications. One advantage of maximum likelihood estimation is the ability to compare non-nested models using likelihood ratio tests. We find, using these tests, that the choice of preferred canonical model can depend on the market price of riskspecification. Comparison to other approximation methods, Euler and QML, on both simulated and real data suggest that our approximation technique is much closer to true MLE than alternative methods.