This paper compares the goodness of fit of a linear ar(1) model and a
Markov ar(1) alternative using data on the inflation rate during the H
ungarian hyperinflation (August 1921-March 1994) and stabilization (Ma
rch 1924-November 1926) periods. First, in-sample forecasting measures
(RMSE and MAPE) indicate that the Markov ar(1) uniformly dominates th
e linear ar(1) for horizons of one to ten periods. Second, bootstrappe
d samples yield a likelihood ratio test statistic which easily rejects
the linear ar(1) in favor of the Markov ar(1) alternative.