A spectral analysis of New Zealand output gaps using Fourier and wavelet techniques

This paper uses frequency domain techniques to illustrate the properties of various measures of New Zealand's output gap. Measures of the output gap are estimated using a number of different methods: a Structural VAR model, a multivariate unobserved components model, the Hodrick-Prescott filter, a multivariate time series filter, and a linear time trend filter. Spectral densities, calculated using the Fourier transform, highlight a number of important differences in the cyclical properties of the various output gap measures. However, the Fourier transform requires time series to be (weakly) stationary. This may be an unreasonable assumption for New Zealand data given our recent economic history. Accordingly, the paper also uses time-dependant spectra, calculated using wavelet analysis, to further illustrate the cyclical characteristics of the different techniques used to estimate the output gap.