Table 7.1 gives an indication of what the real-time and final output gap estimates look
like. It lists the average real-time and final output gaps over a calendar year. Consistent
with previous research, these figures indicate how unreliable real-time output gap
estimates can be. They often paint a different picture to the final estimate, and clearly
illustrate how one can misjudge the position of the business cycle in real-time. In general
the full sample estimates are more ‘long period’ volatile than are the real time estimates.
The latter tend to track actual output more closely, suggesting that perhaps too much
signal is read into the current data.
We can draw out further relevant differences between the real-time and final estimates
shown in Table 7.1. The German output gap estimates suggest that it was difficult to
estimate the degree of excess capacity in the period after German unification and that data
at the time may have been interpreted as suggesting trend growth had risen, when it
probably had not. The same could be said of the mini-boom around 2000, when
perceptions of the subsequent period of slow growth were absent.
The lack of cyclicality in the real time output gap estimator is also apparent in the
outturns for France and Italy, but in both cases the ‘short period’ volatility of the real time
estimator is apparent. The strength of the early 2000s boom in France is only really
apparent with hindsight, and the same is true in Italy, although subsequent below capacity
growth may be apparent there. The correlation of the real time and final estimates is
stronger in Italy than for the other countries.
The Spanish output gap, measured either in real time or over the full sample appears to be
much less ‘long term’ volatile than the other estimates presented here. This may reflect
too little quarterly noise in the data, or the generally smooth evolution of actual output.
Conversely the Netherlands shows long and large cycles which evolve in a smooth way
whichever technique is used, whilst Belgian data show much more ‘short period’
volatility in the estimates of the cycle.
The real-time unreliability of output gap (point) estimates is summarised more fully in
Table 7.2, both for full sample and real-time estimates. The table also examines the
filtered estimates or the quasi-final estimates in the parlance of Orphanides and van
Norden (2002). UC models use the data in two ways in the sense that first they estimate
the parameters of the model, denoted by the vector Θ, and secondly they use these
estimates to obtain the filtered and smoothed estimates of the output gap, namely the
quasi-final and final estimates of the output gap, respectively. The filtered and smoothed
estimates are the expected value of the output gap conditional on information available at
time t (t=1,2,...,T) and T, where T≥t, respectively. Let us denote the real-time estimates,
based on recursively updated estimates of the parameters of a given UC model Θ, by yt|t;
let yt|T denote the final estimates of the output gap at time t, that use full-sample
information T to estimate Θ, and then let yt∣t ( Θ T ) denote the filtered estimates of the
output gap at time t using these full-sample based parameter estimates.
Table 7.2 indicates that as the future becomes the present output gap estimates are
revised. Real-time point estimates of the output gap in the Euro area are unreliable, in the
sense that there is a large and important revision error. This is reflected by the correlation
coefficients against the final estimates being on average only 0.44. Reflecting the
importance of ex post information in re-defining the parameter values, the filtered
estimates are more reliable than the real-time estimates: correlation is higher averaging
0.66. Parameter uncertainty appears to be a dominant source of the unreliability of real-
189