On the robustness of stylised business cycle facts for contemporary New Zealand

Abstract

ABSTRACTWe assess the robustness of stylised business cycle facts for contemporary New Zealand, traditionally computed from HP1600 trend-filtered data. The merits of these HP1600 estimates are considered, relative to those computed from two loess (local regression) trend filtering methods, one (loess11) chosen to exhibit greater fidelity and the other (loess47) to show more pronounced smoothness. The robustness of our key business cycle facts is further evaluated in terms of simple robust standard error estimates of measures of time-invariant volatility and correlation. Time-varying estimates of these quantities are also investigated. Statistically significant bivariate correlations are established for key real expenditure variables, labour market, fiscal and monetary policy, and some inflation variables, with almost all loess47 absolute magnitudes being somewhat greater than HP1600 magnitudes. Their pro- or counter-cyclicality and their lead/lag relationships are robust across HP1600 and loess47 trend filtering, though not for CPI and non-tradables price level variables.