The paper offers a new approach to analyze and estimate rational expectations models when changes in regime are present. In an uncertain environment the problem is then to find an internally consistent solution to the rational expectations hypothesis. The author’s solution consists in specifying an AR(m)-process with two modifications. First, the constant term around which the process is defined, say \(\mu (s_ t)\), and the standard deviation of its innovation, say \(\sigma (s_ t)\), are functions of the regime operative at date t. The discrete valued variable \(S_ t\) is modeled as the outcome of an unobserved discrete-valued two- state, first order Markov process. Given this framework, the author discusses how to draw optimal inference about the state of the system (assuming all parameters are known) and, given the state, how to obtain maximum likelihood parameter estimates.
The main part of the paper deals with the rational expectations hypothesis of the term structure of U.S. interest rates. The author compares traditional linear estimation procedures with hiw own approach and concludes that his approach outperforms the former ones and that linear constant-parameter models are inconsistent with the univariate time series properties of the interest rate series and with the observed bivariate relation between long and short rates under the expectation hypothesis.