摘要：A Bayesian inferential procedure for a gradual switching regression with an autocorrelated and heteroscedastic error term is derived and applied to estimate money demand functions in the U.S., Canada, and Japan using quarterly data. The empirical results show that the money demand functions in these countries switched regimes gradually. Unit root tests are applied to time series data on money and the results indicate that the tests are sensitive to regime changes as well as to the autoregressive error processes.
