ŠT—v

This paper considers a vector error correction model where unknown autoregressive variables are in the cointegrating space. McCallum (1994, NBER Working Paper No.4938) proposed the time-varying risk premium that follows the AR(1) process. By applying a Bayesian Markov Chain Monte Carlo method, it is possible to consider McCallum's model within a cointegration framework. We propose a method to test the rank and to estimate the model including the unknown risk premium in the cointegrating space using the Kalman filter.