RDP 2002-06: Output Gaps in Real Time: are they Reliable Enough to use for Monetary Policy? 5. Conclusions

We have examined the severity of the problems associated with estimating output gaps in real time. On the basis of results derived from 121 vintages of Australian GDP data, from 1971:Q4 to 2001:Q4, we have addressed the questions: How well can we estimate the output gap based solely on information available at the time? And how different are these real-time estimates from estimates generated with the benefit of hindsight?

Our broad conclusion is that quite good estimates of the output gap can be generated in real time. Over the past 28 years, the root-mean-square difference between real-time output-gap estimates derived using our preferred approach and final output gaps estimated with the benefit of hindsight using the latest available data, is less than 2 percentage points. Furthermore, the correlation between these two output-gap series over this time is about 0.8, and the average difference between them is virtually zero, indicating no apparent tendency for the real-time estimates to be biased.

There are three main features of our preferred approach to estimating output gaps in real time. Firstly, we allow for long-lived changes in the rate of growth of potential output. Secondly, we use a Phillips curve relationship to inform our estimates of potential output, rather than relying on a univariate approach based solely on the behaviour of actual output. And thirdly, we take considerable care in modelling the Phillips curve, allowing for a range of possible influences on inflation in addition to the output gap, and recognising the possibility that the inflation process (or our understanding of it) may change through time as identifiable shocks, of a type not previously seen, hit the economy.

The evidence we have presented suggests that all three of these features are useful in enhancing the prospects of generating reasonably accurate real-time output-gap estimates. The assumption of a constant rate of potential output growth may be an innocuous one over sufficiently short horizons, but can lead to serious error over longer periods. Similarly, estimating potential output using a univariate de-trending method, such as a H-P filter, may generate quite good output-gap estimates at times, but there should be no general presumption that it will do so in real time. Over the past 28 years, the correlation between real-time output-gap estimates derived using a H-P filter and our best final estimates of the output gap is virtually zero (in fact, it is slightly negative!).

Finally, if a Phillips curve approach is to be used to generate output-gap estimates in real time, it appears to be important that the Phillips curve be information-rich and as well specified as possible. Relying on a simple Phillips curve, which uses only limited information about the relationship between inflation and the output gap, can lead to poor real-time estimates of the output gap.

Notwithstanding our general optimism about the possibility of generating quite good output-gap estimates in real time, it is appropriate to end on a note of caution. Despite the apparent robustness of our approach to the range of changes in the Australian macroeconomic environment over the past 28 years, there remains an irreducible degree of uncertainty associated with output gaps generated in real time. The problem of ‘not knowing the future’ is still an important one and there will always be times when the best available estimates of the output gap made in real time will turn out, with the benefit of hindsight, to have been badly flawed.