RDP 2005-05: Underlying Inflation: Concepts, Measurement and Performance 4. Conclusion

Several inferences can be drawn from the results of this paper. While statistical measures of underlying inflation may not have an agreed-upon theoretical foundation, it is clear that in practice they have an advantage over simpler exclusion-based measures. By removing elements from the distribution of price changes on a time-varying basis (rather than applying a fixed rule for all periods), trimmed means or weighted medians exploit the trade-off between efficiency and robustness better than some other core inflation measures. Moreover, they are relatively intuitive, making them easy to interpret in practice.

In some cases, these procedures can result in measures of underlying inflation that are biased with respect to CPI inflation. However, as this paper has shown, there is no evidence of bias for measures calculated using the distribution of annual (rather than the quarterly) price changes, or the distribution of seasonally adjusted quarterly price changes. Of these two techniques, the latter provides a more up-to-date reading of inflation, although it may be subject to revision as seasonal factors are updated over time.

Econometric testing suggests that a variety of measures provide a satisfactory gauge of the underlying trend in inflation. A number of measures assessed in this paper have the desirable property that CPI inflation tends to move towards underlying inflation over time, and thus have the potential to add value to the analysis of inflationary trends in Australia. Nonetheless, to ensure that such analysis is robust, it is desirable to consider a range of measures, and to interpret them in the light of other available information about broader economic developments.