We present a Markov-chain-Monte-Carlo technique for the exact estimation of multi-factor models of the term structure of interest rates. We apply this method to the Cox-Ingersoll-Ross-model which provides an interesting case because of the highly non-normal structure of the underlying state space model. Our technique is based on hybrid "Metropolis within Gibbs"-sampling and makes explicit use of the structure of the underlying economic model when constructing the proposal densities for the Metropolis-Hastings algorithm. In an empirical study using US interest rates we find that the difference between approximate quasi maximum likelihood estimation and exact estimation may be substantial.
Key words: Markov-chain-Monte-Carlo methods, term structure models
JEL classification: C15, E43