Estimates of the output gap in real time: how well have we been doing?

This paper addresses the real-time versus ex-post properties of the output gap as quantified by the Reserve Bank of New Zealand's multivariate (MV) filter, starting with the second quarter of 1997, when the current procedure was implemented. There are three sources of revisions of the output gap: revisions of real GDP data, the end point problem of symmetric filters and changes to the calibration of the MV filter. The performance of the output gap with respect to signalling inflationary pressure, as measured by future non- tradables inflation, has been reasonably good, both in real time and ex post. However, during the recorded history of the MV filter, the revisions to real-time output gap have been no smaller than had a standard Hodrick-Prescott (HP) filter been used. Moreover, the MV filter leads to permanently different levels of the output gap estimates if compared to a purely statistical trend. The MV filter is a hybrid construct. The empirical reference to indicators of inflationary pressure distances it from the original concept of the output gap, where a deviation of observed from potential output is taken as a cause of inflationary pressure. There is some indication that a major recalibration of the MV filter in 2002 helped to maintain the correlation with a target variable that it is supposed to "explain".