RDP 2002-06: Output Gaps in Real Time: are they Reliable Enough to use for Monetary Policy? 1. Introduction

Successful macroeconomic management involves a process of continual reassessment of the state of the macroeconomy. Among many things that policy-makers would like to know about the current state of the economy is the extent to which the level of aggregate economic activity exceeds (or falls short of) the economy's productive capacity. This gap, between actual output and the economy's potential output, is the output gap.

For the output gap to be a concept of much value to policy-makers, however, it is not enough for output-gap estimates derived with the benefit of hindsight to provide useful information about the state of excess (or insufficient) demand in the economy in the past. It is instead important for estimates of the current output gap formed on the basis of current information – so-called real-time estimates – to provide a reasonable guide to the current state of excess demand in the economy. In this paper, we therefore ask the question: How well can we estimate the current output gap using currently available information?

The answer to this question is important because it determines, to a considerable extent, the appropriate strategy for the conduct of monetary policy. If, as a general rule, real-time estimates provide a bad guide to the ‘true’ current state of excess demand in the economy, then it makes no sense for policy-makers to place much weight on them in their monetary policy deliberations. In that case (and presuming that there are no alternative indicators that can give a reasonable guide to the current state of excess demand in the economy), policy-makers might be best advised to ignore these estimates and aim instead to stabilise the nominal growth rate of the economy, as has been argued for example by McCallum (1995, 2001).

However, if real-time output-gap estimates provide quite a good guide to the true current state of excess demand in the economy, then an alternative monetary-policy strategy seems superior. That alternative involves responding to the estimated output gap; ensuring that, other things given, monetary policy is expansionary when the estimated gap is negative and contractionary when it is positive. The Taylor rule is the most famous monetary-policy rule incorporating this logic, and many of the papers in the volume edited by Taylor (1999) suggest that monetary-policy rules of this type dominate most simple alternatives, in terms of their capacity to stabilise output and inflation in the economy. But these rules are clearly of use only if estimates of the output gap available in real time are of reasonable quality.

Estimating output gaps is not a straightforward exercise, either in real time or with the benefit of hindsight, simply because the level of potential output, on which they are based, is unobservable. Problems associated with estimating potential output arise from several sources, among which are: uncertainty about the true structure of the economy and hence about the relationship between potential output and observed economic data on actual output, inflation, etc; revisions to the data, particularly to actual output; and end-point problems common to most procedures used to estimate potential output.

In a companion paper, Stone and Wardrop (2002) provided an historical examination of the extent of real-time problems with the measurement of actual output. This examination suggested that the scale and persistence of actual output mismeasurement could make accurate estimation of the output gap difficult in real time, but did not specifically address the task of estimating potential output and, hence, the output gap.

In this paper we take up the challenge of obtaining real-time potential output estimates for 121 vintages of actual Australian GDP data, to assess explicitly the extent to which real-time problems hamper the use of output-gap estimates by policy-makers. The two novel features of this study are the method used to obtain real-time potential output estimates, and the fact that it is the first attempt to assess the scale of the real-time problem in estimating output gaps for Australian data. In this latter regard, it supplements earlier studies for the US undertaken by Orphanides and others, and for the UK by Nelson and Nikolov.

Australia experienced a significant productivity slowdown in the early 1970s, at a similar time to other industrial countries, and a significant productivity acceleration in the 1990s, which pre-dates the US acceleration. In light of the well-known difficulties associated with estimating output gaps in real time in the presence of changes in the trend rate of growth of potential output, these productivity developments make the Australian case of interest to researchers beyond the Antipodes.

The approach we adopt involves estimating a Phillips curve for each vintage of output data, and deriving a smooth path for potential output that generates a best fit for these Phillips curves. Variants of such a Phillips curve-based approach, but using the Kalman filter to derive results, have been recently investigated for US data by Orphanides and van Norden (2001), who report that such an approach does not reduce real-time problems, relative even to an unsophisticated method such as using an ordinary Hodrick-Prescott (H-P) filter.

There are, however, some important differences between our approach and that of Orphanides and van Norden. Orphanides and van Norden estimate Phillips curves which assume a simple relationship between inflation and the output gap. By contrast, we use specifications which allow richer dynamics and a role for influences on inflation other than the output gap. The Phillips curves we estimate include a role for the output gap and also (possible) roles for changes in the gap, estimated inflation expectations from the bond market, oil price inflation and import price inflation. We also allow for possible changes in the Phillips curve specification for different vintages of data as identifiable shocks, such as the oil shock of the early 1970s, hit the economy. In choosing our Phillips curve specifications, however, we are careful to rely solely on information available in real time. By using a richer specification for the Phillips curve, and one which can potentially change with changes in the data vintage, we hope to be able to improve the accuracy of real-time estimates of the output gap.

Overall, our examination of the Australian data, while confirming some of the broad conclusions of Orphanides and others about the difficulties of using output-gap estimates to guide monetary policy in real time, is more encouraging than these previous studies. Our results confirm that, although significant revisions to actual output estimates occur from time to time, these revisions are not the principal source of real-time problems in the estimation of the output gap. Rather, these problems arise primarily from the end-point problem of ‘not knowing the future’. In general, however, our results are quite promising, and suggest that useful information can be extracted from output-gap estimates in guiding judgements about policy in real time.