RDP 9611: A Markov-Switching Model of Inflation in Australia 1. Introduction

Models of the inflation process typically specify inflation as a function of a wide set of macroeconomic and policy-related variables including wages, import prices, commodity prices and business cycle conditions, often involving complicated dynamic structures. These models can be highly successful in tracking actual inflation, given the behaviour of the explanatory variables. Recent work by de Brouwer and Ericsson (1995), for example, shows that an error-correction model including the variables listed above has good explanatory power for inflation in Australia since 1977. An issue not addressed by this kind of modelling, however, is that structural changes may have occured in the underlying processes generating inflation, with possible implications for inflation expectations.

To address these issues, this paper applies an alternative modelling approach based on some recent studies that use Markov-switching models to describe the inflation process.[1] The distinctive feature of this approach is the use of very simple equations for inflation, within a framework that allows for discrete ‘regime shifts’ – ie shifts among a set of alternative equations that can govern the inflation process at different points in time. Specifically, Markov-switching models allow for two (or more) processes to exist with a series of shifts between the states occurring in a probabilistic fashion, so that shifts occur endogenously rather than being imposed by the researcher. The modelling strategy thus imposes a simpler-than-conventional structure on the inflation process within any given regime, but gains power to fit the historical data by allowing regimes to change.

By removing many of the standard explanatory variables this approach clearly ignores information contained in more conventional models. Nonetheless, the approach may be relevant to understanding shifts in inflation expectations if it is true that members of the public also use simple forecasting rules to formulate their expectations, changing the rule when experience deviates significantly from an established pattern. The analysis allows a number of relevant issues to be addressed. These include the forms of simple rules that best fit the data in this kind of framework, the frequency of regime changes and the issue of what constitutes statistical evidence of a regime change from the point of view of an observer using simple forecasting rules.

The results in this paper suggest that the 1970s and 1980s can be characterised by a high-inflation process with relatively persistent deviations from the mean, although the process is ultimately mean reverting. In contrast, the 1960s and 1990s can be characterised as a process with a low mean and less persistent deviation from that mean. The data choose this model in preference to one where the high inflation 1970s and 1980s are characterised by a random walk (and hence do not revert to any particular long-run mean).

Section 2 introduces Markov-switching models and the particular model used in this paper is specified in Section 3. Empirical results are reported in Section 4 and Section 5 concludes.


See Hamilton (1989, 1990) and Ricketts and Rose (1995). [1]