Measuring Core Inflation in Australia with Disaggregate Ensembles

We construct ensemble predictives for inflation in Australia based on the out-of-sample forecast performance of many component models, where each component model uses a particular disaggregate inflation series. Following Ravazzolo and Vahey (2009), the disaggregate ensemble can be interpreted as a forecast-based measure of core inflation. We demonstrate that the ensemble forecast densities for measured inflation using disaggregate information by city and by sector are well calibrated. The resulting forecast densities outperform considerably those from a benchmark autoregressive model. And the point forecasts are competitive. We show that the traditional weighted median and trimmed mean measures of core inflation sometimes differ substantially from the median of the forecast density.