RDP 2003-12: The Real-Time Forecasting Performance of Phillips Curves 6. Conclusions

Among the many techniques used to forecast inflation are Phillips curves based on estimates of the output gap. This paper suggests, however, that their real-time capacity to do so is limited, relative even to such simple alternative forecasting approaches as an AR(2) model or a random walk assumption.^[32] It appears that, while the Phillips curve relationship is useful in real time as a source of information upon which to condition estimates of the output gap, the lack of precision with which the relationship can be estimated in real time limits its usefulness as a means of forecasting inflation.^[33] This is so despite our having taken care to try to make our Phillips curve-based models as richly specified and realistic as possible.

Our Phillips curve-based forecasts may, however, perform a little better than AR model-based ones in at least predicting whether inflation will increase or decrease from its current level. Moreover, combining Phillips curve-based forecasts with those from our alternative, benchmark approaches, does seem to offer at least some scope for improving the real-time out-of-sample forecast accuracy of the latter.

Finally, an inflation-targeting central bank may, in any case, wish to react to anticipated spare capacity in the economy, beyond its expected implications for inflation.^[34] Whether the output gap can be estimated sufficiently accurately in real time for such a purpose remains open, but at least the findings of Gruen et al (2002) on that score were more promising than those of this paper.

These latter observations point to a possible ongoing role for output-gap-based Phillips curves, beyond their value as an ex post tool for understanding historical movements in inflation. They suggest that, in spite of their generally disappointing performance as a means of forecasting inflation in isolation, such Phillips curves may continue to be useful in real time – as a tool for conditioning gap estimates within a multivariate filtering framework, and as a possible complement to other, alternative inflation forecasting approaches.

Footnotes

This conclusion accords with the recent results of Orphanides and van Norden (2003) for the US, regarding the real-time forecasting power of Phillips curves, notwithstanding the different frameworks used to examine the issue in the two studies. [32]

By comparison, the direct impact of real-time output-gap mis-estimation, on the performance of Phillips curve-based inflation forecasts, appears to be of secondary importance. [33]

It is increasingly accepted that flexible inflation-targeting central banks, including the Reserve Bank of Australia, focus not only on deviations of forecast inflation from target, but also of forecast output from potential – see, for example, Bean (2003). [34]