Inflation, Learning and Monetary Policy Regimes in the G-7 Economies

In this paper, the authors report estimates of two- and three-state Markov switching models applied to inflation, measured using consumer price indexes, in the G-7 countries. They report tests that show that two-state models are preferred to simple one-state representations of the data, and argue that three-state representations are more satisfactory than two-state representations for some countries. The preferred estimation results usually include a state that features a unit root in its dynamic structure, which concurs with results of direct tests for this property. However, the multistate representation of the data shows that for all G7 countries these quasi-unit-root properties arise primarily from a few brief episodes of history, concentrated in the 1970s and associated with the major oil-price shocks. For all countries there is evidence of progress towards establishing credibility of regimes with stable inflation, and in many countries there is evidence of progress in building credibility of regimes with low inflation. Credibility refers to the ex post probability assigned to the state by the Markov model, which has a large effect on how expectations of future inflation are formed. An interesting contrast arises from the results for the United States and Canada. Whereas in Canada the credibility of a regime with historically low inflation has risen sharply in the last few years, in the United States there has been convergence on a regime with a stable, but historically average, rate of inflation and not on the alternative low-inflation regime.