RDP 2010-03: Modelling Inflation in Australia 6. Changes in the Inflation Process over Time

There is a large body of evidence documenting changes in the inflation process over time. For example, a number of papers document a flattening of the Phillips curve internationally (see Kuttner and Robinson, forthcoming, for a discussion), while both Heath et al (2004) and Dwyer and Leong (2001) document changes in the sensitivity of Australian inflation to movements in import prices since 1990. In light of this, we examine how the behaviour of our inflation models has evolved over the past three decades, starting in 1982 when disaggregated price data become available for the calculation of trimmed mean inflation.

Rolling Chow tests suggest that a change may have occurred in the sensitivity of inflation to its determinants around 1990, with little evidence of significant change since then. The maximum value of this statistic occurs in the March quarter of 1989 for both the standard Phillips curve and mark-up models, and in the June quarter of 1991 for the OLS NKPC, although it is only significant in the case of the standard Phillips curve model. Since 1990 – the starting period for the regressions presented earlier in the paper – the Chow test is clearly insignificant for all models.

Table 4 shows the model results when estimated over two non-overlapping sample periods, the first running from 1982 to 1992 (the higher-inflation period) and the second running from 1993 to 2009 (the inflation-targeting period).[22] Perhaps the clearest difference between the estimates in each sample is the decline in overall explanatory power, as indicated by the reduction in the adjusted R-squared from 0.83 to 0.45 for the standard Phillips curve model and from 0.86 to 0.33 for the mark-up model. However, the standard errors of the models have also declined over this time, by more than one-third. Most of the reduction in these two measures reflects the exclusion of the early 1990s disinflation period, as evidence by the still-high R-squared on the results shown earlier, which commence in 1990. Overall, these two results appear to be a reflection of the much smaller medium-term variation in inflation in the more recent period.

Table 4: Split-sample Inflation Model Estimates
Standard PC Mark-up model
1982:Q2–1992:Q4 1993:Q1–2009:Q4 1982:Q2–1992:Q4 1993:Q1–2009:Q4
Constant −0.038*** −0.003   −0.007 0.013***
Inflation expectations 0.683*** 0.247**   0.632*** 0.226
Output gap       0.089 0.135**
Unemployment rate 0.215* 0.136***      
Δ unemployment rate −0.003* −0.002      
Δ unit labour costs       0.103 0.220**
Δ import prices 0.237*** 0.087***   0.201*** 0.114***
Adjusted R2 0.831 0.450   0.865 0.328
Standard error 0.0024 0.0015   0.0022 0.0016
AIC −9.06 −10.11   −9.25 −9.89
Durbin-Watson statistic 1.98 1.84   2.33 1.69
Notes: ***, ** and * represent significance at the 1, 5 and 10 per cent levels respectively. Where multiple lags included, coefficients shown are sum of the lags. Coefficients on constant, output gap and level and change in unemployment rate are multiplied by 4 to represent annualised effects.

Looking at the coefficients, there are a number of notable changes between the higher-inflation and inflation-targeting samples. Most striking, the coefficient on inflation expectations declines considerably in the inflation-targeting period, from around 0.60 to 0.25, although it remains significant in the standard Phillips curve model and approaches significance in the mark-up model (with a p-value of 0.12). (As can be seen from a comparison of these results to those over our earlier results, around half of this difference reflects the early 1990s period.)

The other very stark change is the decline in the coefficients on import prices, which fall from over 0.2 in the high inflation era to 0.1 in the inflation-targeting era. This type of result is well documented in the literature. However, there is little agreement regarding why this decline in second-stage pass-through has occurred. The most prominent explanations are: changes in the composition of trade (Campa and Goldberg 2002); an increase in the prevalence of pricing-to-market at the retail level (Devereux and Engel 2001); a change in the nature of exchange rate shocks (Shambaugh 2008); and the introduction of inflation targeting, which might have encouraged firms to absorb such shocks in their margins (Taylor 2000; Gagnon and Ihrig 2004).

More broadly, there is also a decline in the sum of the coefficients on the nominal variables, with the vertical long-run Phillips curve not rejected in the first sample period. In addition, the coefficient on the unemployment rate provides further evidence for the well-documented flattening of the Phillips curve. This declines by half from the first to the second sample, with all of this decline occurring in the 1980s (as evidenced by our earlier results showing a similar coefficient to that in the inflation-targeting period).[23]


The split between these two samples is based on considerations of sample size rather than the Chow tests. [22]

This may partly reflect changes in the NAIRU, but we suspect that this is not a major effect. [23]